CN110910659B - Traffic flow prediction method, device, equipment and storage medium - Google Patents
Traffic flow prediction method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN110910659B CN110910659B CN201911203515.0A CN201911203515A CN110910659B CN 110910659 B CN110910659 B CN 110910659B CN 201911203515 A CN201911203515 A CN 201911203515A CN 110910659 B CN110910659 B CN 110910659B
- Authority
- CN
- China
- Prior art keywords
- flow
- historical
- tensor
- traffic
- characteristic information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/065—Traffic control systems for road vehicles by counting the vehicles in a section of the road or in a parking area, i.e. comparing incoming count with outgoing count
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Traffic Control Systems (AREA)
Abstract
The embodiment of the application discloses a traffic flow prediction method, a traffic flow prediction device, traffic flow prediction equipment and a storage medium, and relates to an intelligent traffic management technology based on artificial intelligence, wherein the method comprises the following steps: acquiring a historical access flow tensor constructed based on the historical access flow of the target area and a historical OD flow tensor constructed based on the historical OD flow of the target area pair; extracting historical access flow characteristic information and historical OD flow characteristic information through a traffic flow prediction model according to the historical access flow tensor and the historical OD flow tensor; carrying out fusion processing on historical incoming and outgoing flow characteristic information and historical OD flow characteristic information through an indication matrix in a traffic flow prediction model, wherein the indication matrix is determined according to the incidence relation between a target area and a target area pair; and determining an inflow rate prediction result and/or an OD rate prediction result through the traffic flow prediction model based on the information obtained by the fusion processing. The method effectively improves the accuracy of traffic flow prediction.
Description
Technical Field
The present application relates to the field of artificial intelligence technologies, and in particular, to a method, an apparatus, a device, and a storage medium for traffic flow prediction.
Background
The prediction of the entrance and exit flow of each region of a city and the Origin and Destination (OD) flow between regions is a basic task of traffic planning management, is also an important target in the fields of smart cities, smart traffic and the like, and has an extremely important role in a plurality of fields of traffic management, city planning, city management and the like.
The prediction of the entrance/exit flow rate and the OD flow rate is substantially a space-time series prediction problem, that is, the traffic flow data at the present time is predicted based on the traffic flow data in the past period. Methods commonly used today to predict ingress and egress flow rates and OD flow rates include: the traditional time sequence analysis method utilizes a time sequence model obtained by deep learning algorithm training to predict and utilizes a space-time prediction model obtained by deep learning algorithm training to predict.
The inventor researches and discovers that the method generally takes the incoming and outgoing flow rate prediction and the OD flow rate prediction as two independent prediction tasks, ignores the correlation between the incoming and outgoing flow rate and the OD flow rate, and does not well utilize the correlation between the incoming and outgoing flow rate and the OD flow rate to assist in improving the accuracy of the incoming and outgoing flow rate and the OD flow rate prediction.
Disclosure of Invention
The embodiment of the application provides a traffic flow prediction method, a traffic flow prediction device, traffic flow prediction equipment and a storage medium based on artificial intelligence, and the accuracy of traffic flow prediction can be effectively improved.
In view of the above, a first aspect of the present application provides an artificial intelligence-based traffic flow prediction method, including:
acquiring a historical access flow tensor and a historical start and end point OD flow tensor of a target geographic range before a time to be predicted; the historical access flow tensor is obtained by construction according to historical access flow of a target area in the target geographic range, and the historical OD flow tensor is obtained by construction according to historical OD flow of a target area pair in the target geographic range;
extracting historical access flow characteristic information and historical OD flow characteristic information through a traffic flow prediction model according to the historical access flow tensor and the historical OD flow tensor;
converting the historical access flow characteristic information into to-be-fused historical access flow characteristic information based on an indication matrix in the traffic flow prediction model, and splicing the to-be-fused historical access flow characteristic information and the historical OD flow characteristic information to obtain first fusion information; and/or converting the historical OD flow characteristic information into historical OD flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical OD flow characteristic information to be fused and the historical access flow characteristic information to obtain second fusion information; the indication matrix is determined according to the incidence relation between the target area and the target area pair;
and determining at least one of an in-out flow prediction result and an OD flow prediction result of the target geographic range at the moment to be predicted based on the first fusion information and/or the second fusion information through the traffic flow prediction model.
A second aspect of the present application provides an artificial intelligence-based traffic flow prediction apparatus, the apparatus including:
the acquisition module is used for acquiring a historical access flow tensor and a historical start and end point OD flow tensor of the target geographic range before the time to be predicted; the historical access flow tensor is obtained by construction according to historical access flow of a target area in the target geographic range, and the historical OD flow tensor is obtained by construction according to historical OD flow of a target area pair in the target geographic range;
the characteristic extraction module is used for extracting historical access flow characteristic information and historical OD flow characteristic information through a traffic flow prediction model according to the historical access flow tensor and the historical OD flow tensor;
the information fusion module is used for converting the historical access flow characteristic information into historical access flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical access flow characteristic information to be fused and the historical OD flow characteristic information to obtain first fusion information; and/or converting the historical OD flow characteristic information into historical OD flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical OD flow characteristic information to be fused and the historical access flow characteristic information to obtain second fusion information; the indication matrix is determined according to the incidence relation between the target area and the target area pair;
and the prediction result determining module is used for determining at least one of an access flow prediction result and an OD flow prediction result of the target geographic range at the moment to be predicted based on the first fusion information and/or the second fusion information through the traffic flow prediction model.
A third aspect of the application provides an apparatus comprising a processor and a memory:
the memory is used for storing a computer program;
the processor is configured to execute the traffic flow prediction method according to the first aspect.
A fourth aspect of the present application provides a computer-readable storage medium for storing a computer program for executing the traffic-flow prediction method according to the first aspect.
A fifth aspect of the present application provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the steps of the traffic flow prediction method of the first aspect described above.
According to the technical scheme, the embodiment of the application has the following advantages:
the embodiment of the application provides a traffic flow prediction method, which integrates the correlation between the input flow and the output flow and the OD flow into the traffic flow prediction process, and assists to improve the accuracy of traffic flow prediction by using the correlation between the input flow and the output flow and the OD flow. Specifically, in the traffic flow prediction method provided in the embodiment of the present application, a historical access flow tensor and a historical OD flow tensor of a target geographic range before a time to be predicted are obtained, where the historical access flow tensor is obtained by constructing a historical access flow of a target area in the target geographic range, and the historical OD flow tensor is obtained by constructing a historical OD flow of a target area pair in the target geographic range; and then, processing the historical entrance and exit flow tensor and the historical OD flow tensor by using a traffic flow prediction model to obtain an entrance and exit flow prediction result and/or an OD flow prediction result of the target geographic range at the moment to be predicted. In order to integrate the correlation between the incoming and outgoing flow and the OD flow in the traffic flow prediction process, the traffic flow prediction model performs fusion processing on historical incoming and outgoing flow characteristic information extracted based on a historical incoming and outgoing flow tensor and historical OD flow characteristic information extracted based on a historical OD flow tensor by using an indication matrix constructed based on the geographic position incidence relation between the target area and the target area pair; therefore, the correlation between the inlet flow and the outlet flow and the OD flow is effectively utilized, and the accuracy of the traffic flow prediction result can be improved to a certain extent.
Drawings
Fig. 1 is a schematic view of an application scenario of a traffic flow prediction method according to an embodiment of the present application;
fig. 2 is a schematic flow chart of a traffic flow prediction method according to an embodiment of the present application;
fig. 3 is a schematic structural diagram of a feature extraction layer in a traffic flow prediction model according to an embodiment of the present application;
FIG. 4 is a schematic diagram of a traffic flow prediction model according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a first traffic flow prediction device according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of a second traffic flow prediction device according to an embodiment of the present application;
fig. 7 is a schematic structural diagram of a third traffic flow prediction apparatus according to an embodiment of the present application;
fig. 8 is a schematic structural diagram of a fourth traffic flow prediction device according to an embodiment of the present application;
fig. 9 is a schematic structural diagram of a fifth traffic flow prediction apparatus according to an embodiment of the present application;
fig. 10 is a schematic structural diagram of a sixth traffic flow prediction apparatus according to an embodiment of the present application;
fig. 11 is a schematic structural diagram of a server according to an embodiment of the present application;
fig. 12 is a schematic structural diagram of a terminal device according to an embodiment of the present application.
Detailed Description
In order to make the technical solutions of the present application better understood, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The terms "first," "second," "third," "fourth," and the like in the description and in the claims of the present application and in the drawings described above, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Artificial Intelligence (AI) is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.
The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.
The technical scheme provided by the embodiment of the application relates to a traffic flow prediction technology in the field of artificial intelligence.
In order to facilitate understanding of the traffic flow prediction method provided in the embodiments of the present application, the following first explains related terms referred to in the present application.
The access flow includes a traffic flow flowing out of a certain area and a traffic flow flowing into the area within a certain time period, where the traffic flow may specifically include a human flow, a vehicle flow, a mobile device flow, and the like, and the application does not make any limitation on an object to which the access flow specifically corresponds.
The OD flow refers to a flow of traffic between the starting and ending regions, that is, a flow of traffic flowing from the starting region to the ending region within a certain period of time, where the flow of traffic may specifically include a human flow, a vehicle flow, a mobile device flow, and the like, and the application also does not make any limitation on an object to which the OD flow specifically corresponds.
A block refers to an area enclosed by roads of a certain grade, and may be a traffic cell in the field of traffic planning and management, and is often used as a basic space unit for research and analysis in the fields of traffic management, city planning, city management, and the like.
A Recurrent Neural Network (RNN) is a type of recurrent neural Network in which sequence data is input, recursion is performed in the evolution direction of the sequence, and all nodes are connected in a chain.
Graph Convolutional neural Network (GCN), a deep learning Network, is applicable to processing objects in non-euclidean spaces.
The core technical idea of the traffic flow prediction method provided by the embodiment of the present application is introduced below.
The inventor of the present application has found that there is a specific correlation between the flow rate of the incoming and outgoing flow and the flow rate of the OD, specifically, the sum of the flow rates of the OD flowing from a certain starting region to each ending region should be equal to the flow rate of the outgoing flow from the starting region, and the sum of the flow rates of the OD flowing from each starting region to a certain ending region should be equal to the flow rate of the incoming flow to the ending region; the prediction of the incoming and outgoing flow rates and the prediction of the OD flow rates should be essentially organically integrated. However, in the related art, only the flow rate of the incoming and outgoing traffic is generally and individually predicted, and the correlation between the flow rate of the incoming and outgoing traffic and the flow rate of the OD traffic is not well utilized in the process of predicting the traffic flow, so that the accuracy of traffic flow prediction still needs to be improved.
In view of the problems in the prior art, embodiments of the present application provide a traffic flow prediction method, which integrates the correlation between the incoming and outgoing flow and the OD flow into the traffic flow prediction process, and organically combines the incoming and outgoing flow prediction task and the OD flow prediction task, thereby improving the accuracy of traffic flow prediction.
Specifically, in the technical scheme provided by the embodiment of the application, a historical access flow tensor which is constructed based on historical access flow of a target area in a target geographic range and a historical OD flow tensor which is constructed based on historical OD flow of a target area pair in the target geographic range are obtained; then, the traffic flow prediction model obtained through pre-training realizes the prediction of the entrance flow and the exit flow and/or the prediction of the OD flow in the target geographic range according to the historical entrance flow tensor and the historical OD flow tensor. The traffic flow prediction model is provided with an indication matrix constructed according to the geographical position incidence relation between the target area and the target area pair, the indication matrix is used for carrying out fusion processing on historical input and output flow characteristic information extracted from a historical input and output flow tensor and historical OD flow characteristic information extracted from a historical OD flow tensor so as to realize organic fusion between the input and output flow characteristic information and the OD flow characteristic information, then an input flow prediction result and/or an OD flow prediction result are determined based on the information obtained after the fusion processing, and the correlation between the input and output flow and the OD flow is fused into the traffic flow prediction process, so that the accuracy of traffic flow prediction is improved.
It should be understood that the traffic flow prediction method provided by the embodiment of the present application may be specifically applied to various devices with data processing capability, such as terminal devices, servers, and the like. The terminal device may be a computer, a Personal Digital Assistant (PDA), a tablet computer, a smart phone, or the like; the server may specifically be an application server or a Web server, and in actual deployment, the server may be an independent server or a cluster server.
In order to facilitate understanding of the technical solutions provided in the embodiments of the present application, taking an example that the traffic flow prediction method provided in the embodiments of the present application is applied to a server, an application scenario in which the traffic flow prediction method provided in the embodiments of the present application is applied is described below.
Referring to fig. 1, fig. 1 is a schematic view of an application scenario of a traffic flow prediction method according to an embodiment of the present application. As shown in fig. 1, the application scenario includes a server 110, where the server 110 is configured to execute the traffic flow prediction method provided in the embodiment of the present application, and a traffic flow prediction model 111 trained in advance is run in the server 110, and the server 110 may determine an access flow prediction result and/or an OD flow prediction result of a target geographic range at a time to be predicted based on a historical access flow tensor and a historical OD flow tensor of the target geographic range before the time to be predicted.
Specifically, when the ingress and egress traffic and/or the OD traffic in the target geographic range needs to be predicted, the historical ingress and egress traffic of the target area in the target geographic range and the historical OD traffic of the target area pair in the target geographic range before the time to be predicted may be input to the server 110; the server 110 constructs a historical ingress and egress traffic tensor from the received historical ingress and egress traffic and constructs a historical OD traffic tensor from the received historical OD traffic.
The server 110 inputs the constructed historical input/output flow rate tensor and the historical OD flow rate tensor into the traffic flow prediction model 111 trained in advance, extracts historical input/output flow rate feature information from the historical input/output flow rate tensor through the first feature extraction layer 1111 in the traffic flow prediction model 111, wherein the historical input/output flow rate feature information may be a tensor capable of reflecting historical input/output flow rate space-time features, and the historical OD flow rate feature information may be a tensor capable of reflecting the historical OD flow rate space-time features through the second feature extraction layer 1112 in the traffic flow prediction model 111. Then, the historical incoming and outgoing flow characteristic information and the historical OD flow characteristic information are fused through an indication matrix 1113 in the traffic flow prediction model 111; specifically, historical access flow characteristic information can be converted into to-be-fused historical access flow characteristic information based on the indication matrix 1113, and the to-be-fused historical access flow characteristic information and the historical OD flow characteristic information are spliced to obtain first fusion information; and/or, historical OD traffic characteristic information may be converted into to-be-fused historical OD traffic characteristic information based on the indication matrix 1113, and the to-be-fused historical OD traffic characteristic information and the historical ingress and egress traffic characteristic information are spliced to obtain second fusion information, so that the ingress and egress traffic prediction task and the OD traffic prediction task are organically fused. Finally, the output layer 1114 of the traffic flow prediction model 111 determines the input and output flow prediction result and/or the OD flow prediction result of the target geographic range based on the first fusion information and/or the second fusion information obtained after the fusion processing.
Because the indication matrix 1113 in the traffic flow prediction model 111 is constructed based on the geographical position incidence relation between the target area of the input flow to be predicted and the target area pair of the OD flow to be predicted, and the geographical position incidence relation between the target area and the target area pair can reflect the relation between the input flow and the output flow of the target area and the OD flow of the target area pair to a certain extent, the historical input flow characteristic information and the historical OD flow characteristic information are fused by using the indication matrix 1113, and the input flow prediction result and/or the OD flow prediction result are determined based on the fused information, which is substantially equivalent to the fusion of the correlation between the input flow and the output flow into the prediction process of the input flow and/or the output flow and the OD flow, so as to improve the accuracy of the input flow and/or the output flow prediction.
It should be understood that the application scenario shown in fig. 1 is only an example, and in practical applications, the traffic flow prediction method provided in the embodiment of the present application may also be applied to other scenarios, and no limitation is made to the application scenario to which the traffic flow prediction method provided in the embodiment of the present application is applied.
The following describes in detail the traffic flow prediction method provided by the present application by way of example.
Referring to fig. 2, fig. 2 is a schematic flow chart of a traffic flow prediction method according to an embodiment of the present application. For convenience of description, the following embodiments are described taking a server as an execution subject. As shown in fig. 2, the method comprises the steps of:
step 201: and acquiring a historical ingress and egress flow tensor and a historical origin and destination OD flow tensor of the target geographic range before the time to be predicted.
When the traffic of a target area in the target geographic range at the time to be predicted needs to be predicted, and/or the OD traffic of a target area pair in the target geographic range at the time to be predicted needs to be predicted, the server needs to obtain a historical traffic tensor constructed based on the historical traffic of the target area before the time to be predicted, and a historical traffic tensor constructed based on the historical OD traffic of the target area before the time to be predicted.
It should be understood that in practical applications, the server may construct the historical ingress and egress traffic tensor using the historical ingress and egress traffic of the target area within the target geographic range, and construct the historical OD traffic tensor using the historical OD traffic of the target area pair within the target geographic range. The server can also obtain the constructed historical ingress and egress traffic tensor and the historical OD traffic tensor from other devices. The method for acquiring the historical ingress and egress traffic tensor and the historical OD traffic tensor by the server is not limited in any way.
Optionally, in order to ensure that the prediction result obtained by the technical solution provided in the embodiment of the present application is more appropriate to the actual prediction requirement in the related art, the area within the target geographic range in the present application may be divided according to the street blocks, and accordingly, the target area mentioned in the present application may be substantially understood as the target street block, and the target area pair mentioned in the present application may be substantially understood as the target street block pair. The block is a basic unit in an actual traffic management task, so that the block is used as a basic research unit, a historical access flow tensor is constructed based on the access flow of the target block, and a historical OD flow tensor is constructed based on the OD flow of the target block pair, so that the access flow prediction result and/or the OD flow prediction result predicted based on the access flow tensor and the OD flow prediction result can be guaranteed to be more practical, and the actual requirements of the related fields are better met.
The historical access flow of the target area is the access flow in a plurality of preset time periods before the time to be predicted, for example, the historical access flow of the target area may be the access flow in two hours before the time to be predicted, the access flow in a time corresponding to the time to be measured in the previous day or the previous days, and the like, and in general, the historical access flows of the target areas for constructing the historical access flow tensor belong to the same time period. Similarly, the historical OD traffic of the target area pair is also the OD traffic in a plurality of preset time periods before the time to be predicted, for example, the historical OD traffic may be the OD traffic of the target area pair in two hours before the time to be predicted, or the OD traffic of the target area pair in the previous day or multiple days corresponding to the time to be predicted, and in general, the historical OD traffic of each target area pair for constructing the historical OD traffic tensor belongs to the same time period.
The time period to which the history ingress/egress traffic volume belongs for constructing the history ingress/egress traffic volume tensor may be the same as or different from the time period to which the history OD traffic volume belongs for constructing the history OD traffic volume tensor. In order to ensure that the correlation between the ingress and egress traffic and the OD traffic can be better utilized, it is generally necessary to ensure that the period of the historical ingress and egress traffic used to construct the historical ingress and egress traffic tensor is the same as the period of the historical OD traffic used to construct the historical OD traffic tensor.
It should be noted that the historical ingress and egress flow tensor in the single preset time period may be specifically a historical ingress and egress flow matrix with a size of N × 2, where N represents the number of target areas in the target geographic range where ingress and egress flows are to be predicted, and N is an integer greater than or equal to 1. An element with an index of [ i,0] in the historical incoming and outgoing flow matrix can represent the historical incoming flow of the ith target area, and an element with an index of [ i,1] can represent the historical outgoing flow of the ith target area; alternatively, the element with index [ i,0] in the historical ingress and egress traffic matrix may represent the historical egress traffic of the ith target area, and the element with index [ i,1] may represent the historical ingress traffic of the ith target area. The historical OD flow tensor in the single preset time period may be specifically a historical OD flow matrix of M × 1, where a jth element in the historical OD flow matrix represents the historical OD flow of a jth target region. Optionally, in order to ensure that more dimensions of information can be referred to in the subsequent prediction of the ingress and egress traffic and/or the OD traffic, the historical ingress and egress traffic tensor acquired by the server may specifically include historical ingress and egress traffic tensors respectively corresponding to multiple different sampling periods, and the historical OD traffic tensor acquired may specifically include historical OD traffic tensors respectively corresponding to multiple different sampling periods.
Assuming that a server needs to obtain historical ingress and egress traffic tensors corresponding to P (P is an integer greater than 1) different sampling periods, the method can be implemented as follows: and acquiring the access flow of each target area in the target geographic range in P sampling periods, and constructing and obtaining a historical access flow tensor corresponding to the sampling period according to the access flow of each target area in the target geographic range in the sampling period aiming at each sampling period in the P sampling periods.
For example, it is assumed that the server needs to obtain the historical ingress and egress traffic tensors corresponding to three sampling periods, which are an hour sampling period, a day sampling period, and a week sampling period. At this time, the server may obtain the access flow rate of each target area within the target geographic range within several hours, and further construct a historical access flow rate tensor of the target geographic range with hours as a sampling period based on the access flow rate of each target area within several hours. The server can obtain the access flow of each target area in the target geographic range within a plurality of days, and then based on the access flow of each target area within a plurality of days, a historical access flow tensor of the target geographic range with the days as a sampling period is constructed. The server can obtain the access flow of each target area in the target geographic range within a plurality of weeks, and then based on the access flow of each target area within a plurality of weeks, a historical access flow tensor of the target geographic range with the weeks as sampling periods is constructed.
In order to ensure that the obtained historical access flow tensor has more reference value for the prediction of the access flow and/or the OD flow at the time to be predicted, a first target time period in each sampling period may be determined according to the time to be predicted, then the access flow of each target area in the first target time period in the sampling period is obtained for each sampling period, and the historical access flow tensor corresponding to each sampling period is constructed based on the access flow tensor. That is, the historical access flow tensor corresponding to each sampling period in the P sampling periods is constructed according to the access flow of each target area in the first target time period under the sampling period.
Still take the example that the server needs to obtain the historical ingress and egress traffic tensors corresponding to three sampling periods, where the three sampling periods are respectively the hour sampling period, the day sampling period, and the week sampling period. Assuming that the time to be predicted is 8 a.m. on wednesday, when the server acquires the historical access flow tensor taking the hour as the sampling period, it may determine 6 a.m. and 7 a.m. on wednesday as the first target time period, and then construct and obtain the historical access flow tensor taking the hour as the sampling period in the target geographic range based on the access flow of each target area at 6 a.m. and 7 a.m. on wednesday. When the server acquires the historical access flow tensor taking the day as the sampling period, the historical access flow tensor taking the day as the sampling period can be determined to serve as a first target time interval at 8 am on Monday and 8 am on Tuesday, and then the historical access flow tensor taking the day as the sampling period in the target geographic range is constructed and obtained based on the access flow of each target area at 8 am on Monday and 8 am on Tuesday. When the server acquires the historical access flow tensor taking the week as the sampling period, the server may determine that the access flow of each target area at the time of 8 am on wednesday of the previous week and at the time of 8 am on wednesday of the previous two weeks is used as a first target time period, and then construct and obtain the historical access flow tensor taking the week as the sampling period in the target geographic range based on the access flow of each target area at the time of 8 am on wednesday of the previous week and at the time of 8 am on wednesday of the previous two weeks.
It should be understood that, in practical applications, the period length and the P value of the sampling period may be set according to actual requirements, and the period length and the P value of the sampling period are not specifically limited herein. In addition, the selection manner of the first target time period in each sampling period may also be set according to actual requirements, and the application does not limit the selection manner of the first target time period.
Assuming that a server needs to obtain historical OD traffic tensors corresponding to Q (Q is an integer greater than 1) different sampling periods, the method can be implemented as follows: and obtaining the OD flow of each target area pair in the target geographic range in Q sampling periods, and constructing and obtaining a historical OD flow tensor corresponding to the sampling period according to the OD flow of each target area pair in the target geographic range in the sampling period aiming at each sampling period in the Q sampling periods.
For example, it is assumed that the server needs to obtain the historical OD traffic tensors corresponding to three sampling periods, which are an hour sampling period, a day sampling period, and a week sampling period. At this time, the server may obtain the OD traffic of each target region pair within the target geographic range within several hours, and then construct a historical OD traffic tensor of the target geographic range with hours as a sampling period based on the OD traffic of each target region pair within several hours. The server can obtain the OD flow of each target area pair in the target geographic range within a plurality of days, and then based on the OD flow of each target area pair in the plurality of days, a historical OD flow tensor of the target geographic range with the days as a sampling period is constructed. The server can obtain the OD flow of each target region pair in a target geographic range in a plurality of weeks, and then based on the OD flow of each target region pair in the plurality of weeks, a historical OD flow tensor of the target geographic range with the weeks as sampling periods is constructed.
In order to ensure that the obtained historical OD traffic tensor has a reference value for the prediction of the incoming and outgoing traffic and/or the OD traffic at the time to be predicted, a second target time period in each sampling period may be determined according to the time to be predicted, then for each sampling period, the OD traffic of each target region in the second target time period in the sampling period may be obtained, and the historical OD traffic tensor corresponding to each sampling period may be constructed based on the OD traffic tensor. That is to say, the historical OD traffic tensor corresponding to each sampling period in the Q sampling periods is constructed according to the OD traffic of each target area pair in the second target time period in the sampling period.
Still take the example that the server needs to obtain the historical OD traffic tensors corresponding to three sampling periods, where the three sampling periods are respectively the hour sampling period, the day sampling period, and the week sampling period. Assuming that the time to be predicted is 8 a.m. on wednesday, when the server acquires the historical OD traffic tensor taking hours as the sampling period, it may determine 6 a.m. and 7 a.m. on wednesday as the second target time period, and then construct and obtain the historical OD traffic tensor taking hours as the sampling period in the target geographic range based on the OD traffic of each target region pair at 6 a.m. and 7 a.m. on wednesday. When the server acquires the historical OD traffic tensor taking the day as the sampling period, it may determine that 8 am on monday and 8 am on tuesday as the second target time period, and then construct and obtain the historical OD traffic tensor taking the day as the sampling period in the target geographic range based on the OD traffic of each target region pair at 8 am on monday and 8 am on tuesday. When the server acquires the historical OD traffic tensor taking the week as the sampling period, it may determine that the last three morning 8 hours and the last two weeks three morning 8 hours are used as the second target time period, and then construct and obtain the historical OD traffic tensor taking the week as the sampling period in the target geographic range based on the OD traffic of each target region for the last three morning 8 hours and the last two weeks three morning 8 hours.
It should be understood that, in practical applications, the period length and the Q value of the sampling period may be set according to actual requirements, and the period length and the Q value of the sampling period are not limited in any way herein. In addition, the selection manner of the second target time period in each sampling period may also be set according to actual requirements, and the application does not limit the selection manner of the second target time period.
When the traffic flow prediction task focuses on fully predicting the traffic flow in the target geographic area, if the historical OD flows of all the target region pairs are directly processed based on the target geographic area, the required calculation amount is huge, for example, if the target geographic area includes N target regions, and the target geographic area correspondingly includes N × N target region pairs, if the historical OD flows of the N × N target region pairs are directly processed by related calculation, the required calculation amount is N2(ii) of rank; in addition, when blocks are used as the basic research units, the number of blocks included in the target geographic scope is typically on the order of ten thousand, and correspondingly the number of block pairs included in the target geographic scope will be on the order of billions based on billions of streetsThe area pair traffic flow is subjected to relevant calculation processing, and the calculation amount required is more difficult to bear. Therefore, the traversing processing mode is difficult to be applied to the traffic flow prediction task of a large-scale traffic unit.
In order to reduce the computational resources consumed by processing a large number of historical OD flows while meeting the need for a comprehensive prediction of traffic flow within a target geographic area, the present application may determine the above-described target zone pairs by:
obtaining OD flow of each region pair in a target geographic range in a historical training period; and further, selecting the area pair with the OD flow meeting the preset condition in the historical training period from the area pairs in the target geographic range as the target area pair. The regional pairs in the target geographic range are screened in advance, the number of screened target regional pairs is usually far smaller than the number of street pairs actually included in the target geographic range, so that the scheme provided by the application can be better applied to traffic flow prediction tasks of large-scale traffic units, and the OD flow of the target regional pairs screened in this way is usually more valuable for comprehensively predicting the access flow and/or the OD flow in the target geographic range.
In specific implementation, the server may obtain an average OD traffic of each region pair in the target geographic range in the historical training period. Then, M area pairs with an average OD flow greater than a preset threshold may be selected from the area pairs within the target geographical range as target area pairs, for example, an area pair with an average hourly pedestrian flow greater than 1 is selected as a target area pair; or, the respective corresponding average OD flows of each region pair in the target geographic range may be sorted in a descending order, and the top M region pairs corresponding to the average OD flows are selected as the target region pairs.
It should be understood that, in practical applications, the server may select the target region with reference to other data related to the OD traffic in the historical training period, for example, the median OD traffic, the minimum OD traffic, the maximum OD traffic, and the like of each region pair in the historical training period, in addition to the average OD traffic of each region pair in the historical training period, and the reference data of the target region pair is not limited in this application. In addition, both the historical training time period and the preset threshold value can be set according to actual requirements, and the historical training time period and the preset threshold value are not specifically limited in the application.
Step 202: and extracting historical access flow characteristic information and historical OD flow characteristic information through a traffic flow prediction model according to the historical access flow tensor and the historical OD flow tensor.
After acquiring a historical access flow tensor and a historical OD flow tensor of a target geographic range before a time to be predicted, a server inputs the acquired historical access flow tensor and historical OD flow tensor into a traffic flow prediction model which is trained in advance; extracting historical access flow characteristic information from a historical access flow tensor through a structure for extracting the access flow characteristic in the traffic flow prediction model, wherein the historical access flow characteristic information can be information capable of reflecting time sequence characteristics or space-time characteristics of historical access flow; through a structure for extracting OD flow characteristic information in the traffic flow prediction model, historical OD flow characteristic information is extracted from the historical OD flow tensor, and the historical OD flow characteristic information can be specifically information capable of reflecting time sequence characteristics or space-time characteristics of the historical OD flow.
In a possible implementation manner, when a block is used as a basic research unit in the technical solution provided in the embodiment of the present application, that is, when a historical access flow tensor acquired by a server is constructed according to historical access flows of target blocks in a target geographic range, and a historical OD flow tensor is constructed according to historical OD flows of target block pairs in the target geographic range, considering that the block usually belongs to an irregular shape, a GCN may be set in a traffic flow prediction model, and historical access flow characteristic information and historical OD flow characteristic information are extracted by using the GCN.
Specifically, the traffic flow prediction model may include a first GCN layer and a second GCN layer; when historical access flow characteristic information and historical OD flow characteristic information are extracted through a traffic flow prediction model, a historical access flow space-time tensor can be obtained by processing a historical access flow tensor through a first GCN layer and used as historical access flow characteristic information, and a historical OD flow space-time tensor can be obtained by processing a historical OD flow tensor through a second GCN layer and used as historical OD flow characteristic information.
The number of the first GCN layers and the number of the second GCN layers are hyper-parameters of the traffic flow prediction model, and may be set according to actual needs. The processing formula of the GCN layer is shown as formula (1):
H(l)=σ(H(l-1)W(l)0+AH(l-1)W(l)1) (1)
wherein H(l)Denotes the output of the l-th layer in the GCN layer, H(l-1)The output of the l-1 layer in the GCN layer is represented, and the output of the l-1 layer in the GCN layer is the input of the l layer in the GCN layer; w(l)0And W(l)1Is the model parameter tensor in the GCN layer, which is constantly updated with the training of the model; a is an adjacency matrix used for representing the relation between the regions after normalization processing, the size of A in the first GCN layer is N x N, and the size of A in the second GCN layer is M x M; σ is the activation function.
When constructing the adjacency matrix in the first GCN layer, the OD traffic in the historical training period may be constructed from N × N region pairs obtained by combining N target blocks two by two. Specifically, the neighbor matrix with size N × N may be initialized, and then the average OD traffic of N × N region pairs in the historical training period is obtained and is used as an element in the initialized neighbor matrix, for example, assuming that the average OD traffic from the target block 16 to the target block 12 in the historical training period is 5, E [16] [12] in the initialized neighbor matrix E in the first GCN layer is 5. And after determining each element value in the adjacency matrix, carrying out normalization processing on the adjacency matrix to obtain the adjacency matrix suitable for the first GCN layer.
When constructing the adjacency matrix in the second GCN layer, the elements in the adjacency matrix may be determined according to M × M region pair combinations obtained by combining M target block pairs two by two. Specifically, the M target block pairs may be combined two by two to obtain M × M target block pair combinations, and then each target block pair combination in the M × M target block pair combinations is processed as follows: and judging whether the end point region contained in the previous target block pair in the target block pair combination is the same as the start point region contained in the next target block pair, if so, determining the element corresponding to the target block pair combination in the adjacent matrix according to the OD flow of the previous target block pair in the target block pair combination in the historical training period, and if not, directly determining the element corresponding to the target block pair combination in the adjacent matrix to be 0. And after determining each element in the adjacency matrix, carrying out normalization processing on the adjacency matrix to obtain the adjacency matrix suitable for the second GCN layer.
It should be noted that, each element in the M × M size adjacency matrix corresponds to one target region pair combination, and if the end region of a previous target region pair in one target region pair combination is the same as the start region of a next target region pair, it represents that the OD traffic of the previous target region pair in the target region pair combination will affect the OD traffic of the next target region pair. For example, assuming that the target zone pair E [38] [18] 'is 5 (indicating an average OD flow rate from the zone 38 to the zone 18 within the historical training period is 5) and the target zone pair E [6] [18 ]' is 12 (indicating an average OD flow rate from the zone 6 to the zone 18 within the historical training period is 12), S [ (38,18), (18,42) ] -5, S [ (6,18), (18,42) ] -12 in the adjacency matrix S. On the contrary, if the end region of the previous target region pair in the combination of the target region pairs is different from the start region of the next target region pair, it represents that the OD traffic of the previous target region pair in the combination of the target region pairs does not affect the OD traffic of the next target region pair, and at this time, it may be directly determined that the element corresponding to the combination of the target street pair in the adjacency matrix is 0.
In order to ensure that the traffic flow prediction model can better extract the time series characteristic information of the incoming and outgoing flow from the historical incoming and outgoing flow tensor and better extract the time series characteristic information of the OD flow from the historical OD flow tensor, in the traffic flow prediction model provided in the embodiment of the present application, a first RNN layer may be further disposed before the first GCN layer, and a second RNN layer may be further disposed before the second GCN layer, so that the time series characteristic of the incoming flow and the time series characteristic of the OD flow are better extracted by using the first RNN layer and the second RNN layer, respectively.
Specifically, when the historical ingress and egress flow tensor is processed, the historical ingress and egress flow tensor can be processed through the first RNN layer to obtain the historical ingress and egress flow time sequence tensor. In order to avoid losing original information, the historical access flow rate time sequence tensor is spliced with the input historical access flow rate tensor, for example, assuming that the size of the historical ingress and egress flow tensor is H × N × 2 (where H denotes the number of preset periods related to the historical ingress and egress flow tensor), processing the historical ingress and egress flow tensor by the first RNN layer will obtain a historical ingress and egress flow timing tensor of size N × T1 (where T1 is the output dimension of the first RNN layer), moving the first dimension of the historical ingress and egress flow tensor to the third dimension, reshaping (rehape) to obtain a tensor of size N × 2 × H, and Reshape to obtain a two-dimensional tensor of size N × 2H, and concatenating the historical ingress and egress flow timing tensor and the tensor in the last dimension to obtain a tensor of size N × 2H + T1. And then the spliced tensor is processed through the first GCN layer to obtain a historical access flow space-time tensor which is used as historical access flow characteristic information.
Similarly, when the historical OD traffic tensor is specifically processed, the input historical OD traffic tensor may be processed by the second RNN layer to obtain the historical OD traffic time sequence tensor. In order to avoid losing original information, the historical OD flow time sequence tensor is spliced with the input historical OD flow tensor, for example, assuming that the size of the historical OD flow tensor is H × M × 1 (where H denotes the number of preset periods related to the historical OD flow tensor), the historical OD flow tensor is processed by the second RNN layer to obtain the historical OD flow tensor with the size of M × T2 (where T2 is the output dimension of the second RNN layer), the first dimension of the historical OD flow tensor is moved to the third dimension, the reshaping (Reshape) is obtained to obtain the tensor with the size of M × 1 × H, then the Reshape is further obtained to obtain the two-dimensional tensor with the size of M × H, and the historical OD flow time sequence and the two-dimensional tensor are spliced in the last dimension to obtain the tensor with the size of M × M (H + T2). And processing the spliced tensor through the second GCN layer to obtain a historical OD flow tensor as historical OD flow characteristic information.
It should be noted that the first RNN layer and the second RNN layer may specifically have a standard RNN network structure, or may also have structures such as a Long Short-Term Memory network (LSTM), a Gated Round Unit (GRU), and the like.
In practical applications, the structure for extracting the inflow rate feature information in the traffic flow prediction model may include only the first RNN layer, and the structure for extracting the OD flow feature information may include only the second RNN layer, and in this case, the historical inflow/outflow rate feature information extracted by the first RNN layer is substantially information that can represent an inflow/outflow rate time series feature, and the historical OD flow feature information extracted by the second RNN layer is substantially information that can represent an OD flow rate time series feature.
In addition, when the technical solution provided in the embodiment of the present application uses other regular shapes (such as artificially divided grids) as basic research units, the traffic flow prediction model may extract historical ingress and egress flow characteristic information by using a first RNN layer and a first Convolutional Neural Network (CNN) layer, and may extract historical OD flow characteristic information by using the first RNN layer and a second CNN layer.
Optionally, when the historical access flow tensor input into the traffic flow prediction model includes P historical access flow tensors corresponding to P sampling periods respectively, and the historical OD flow tensors include Q historical OD flow tensors corresponding to Q sampling periods respectively, the traffic flow prediction model may correspondingly include P first feature extraction layers and Q second feature extraction layers, where the P first feature extraction layers correspond to the P sampling periods one to one, and the Q second feature extraction layers correspond to the Q sampling periods one to one.
Specifically, when historical access flow characteristic information is extracted through a traffic flow prediction model, for each sampling period in P sampling periods, a historical access flow tensor corresponding to the sampling period is processed through a first characteristic extraction layer corresponding to the sampling period, so that sub-historical access flow characteristic information corresponding to the sampling period is obtained; and further, fusing the sub-historical access flow characteristic information corresponding to the P sampling periods to obtain historical access flow characteristic information.
Similarly, specifically, when the historical OD flow characteristic information is extracted through the traffic flow prediction model, for each sampling period of the Q sampling periods, the historical OD flow tensor corresponding to the sampling period is processed through the second characteristic extraction layer corresponding to the sampling period, so as to obtain the sub-historical OD access flow characteristic information corresponding to the sampling period; and then, fusing the sub-historical OD flow characteristic information corresponding to the Q sampling periods respectively to obtain the historical OD flow characteristic information.
The first feature extraction layer may specifically include at least one structure of a first RNN layer and a first GCN layer, and the second feature extraction layer may specifically include at least one structure of a second RNN layer and a second GCN layer. That is, when the input historical access flow tensor includes a plurality of input historical access flow tensors, for each historical access flow tensor, corresponding historical access flow characteristic information can be extracted by using the first RNN layer and/or the first GCN layer; when the input historical OD traffic tensor includes a plurality of quantities, for each of the historical OD traffic tensors, the corresponding historical OD traffic characteristic information may be extracted by using the second RNN layer and/or the second GCN layer.
For convenience of understanding, the following description will be given of an implementation process of extracting the historical ingress and egress flow characteristic information of the traffic flow prediction model and an implementation process of extracting the historical OD flow characteristic information, with reference to fig. 3, taking a sampling period corresponding to the historical ingress and egress flow tensor and a sampling period corresponding to the historical OD flow tensor as examples, where each of the sampling periods includes an hour sampling period, a day sampling period and a week sampling period.
As shown in fig. 3, the input layer 310 of the traffic flow prediction model is configured to receive the input historical ingress and egress flow tensor and the historical OD flow tensor, and the historical ingress and egress flow tensor received by the input layer 310 includes a historical ingress and egress flow tensor taking hours as a sampling period, a historical ingress and egress flow tensor taking days as a sampling period, and a historical ingress and egress flow tensor taking weeks as a sampling period. For the historical access flow tensor taking hours as a sampling period, the traffic flow prediction model may perform space-time feature extraction processing on the historical access flow tensor by using the first feature extraction layer 320, specifically, the historical access flow tensor taking hours as the sampling period may be processed by using the first RNN layer 321 in the first feature extraction layer 320 to obtain a historical access flow time sequence tensor, and then the historical access flow time sequence tensor and the historical access flow tensor taking hours as the sampling period are spliced together, and the spliced tensor is processed by using the first GCN layer 322 in the first feature extraction layer 320 to obtain historical access flow feature information taking hours as the sampling period. Similarly, the historical access flow tensor taking the day as the sampling period and the historical access flow tensor taking the week as the sampling period may be processed by the first feature extraction layer 330 and the first feature extraction layer 340, respectively, to obtain historical access flow feature information taking the day as the sampling period and historical access flow feature information taking the week as the sampling period.
The historical OD traffic tensors received by the input layer 310 include historical OD traffic tensors with hour as a sampling period, historical OD traffic tensors with day as a sampling period, and historical OD traffic tensors with week as a sampling period. For the historical OD flow tensors with the hour as the sampling period, the traffic flow prediction model may perform space-time feature extraction processing on the historical OD flow tensors with the hour as the sampling period by using the second feature extraction layer 350, specifically, the historical OD flow tensors with the hour as the sampling period may be processed by using the second RNN layer 351 in the second feature extraction layer 350 to obtain the historical OD flow time sequence tensors, and then the historical OD flow time sequence tensors and the historical OD flow tensors with the hour as the sampling period are spliced together, and then the spliced tensors are processed by using the second GCN layer 352 in the second feature extraction layer 350 to obtain the historical OD flow feature information with the hour as the sampling period. Similarly, the second feature extraction layer 360 and the second feature extraction layer 370 may be respectively used to process the historical OD traffic tensor with the sampling period of days and the historical OD traffic tensor with the sampling period of weeks, so as to obtain the historical OD traffic feature information with the sampling period of days and the historical OD traffic feature information with the sampling period of weeks.
Optionally, in order to better fuse the sub-history access flow characteristic information corresponding to the P sampling periods and the sub-history OD flow characteristic information corresponding to the Q sampling periods, that is, to ensure that the history access flow characteristic information and the history OD flow characteristic information obtained after the fusion process have higher reference values, an external characteristic layer is further provided in the traffic flow prediction model provided in the embodiment of the present application. The external feature layer is used for acquiring external feature information of a moment to be predicted, determining weights corresponding to P sampling periods according to the external feature information, and determining weights corresponding to Q sampling periods according to the external feature information.
Specifically, when the sub-history access flow characteristic information corresponding to each of the P sampling periods is subjected to fusion processing, the sub-history access flow characteristic information corresponding to each of the P sampling periods may be subjected to weighting processing by using the weights corresponding to each of the P sampling periods determined by the external feature layer, so as to obtain the history access flow characteristic information. Similarly, when the sub-history OD traffic characteristic information corresponding to each of the Q sampling periods is fused, weights corresponding to each of the Q sampling periods may be determined by using the external feature layer, and the sub-history OD traffic characteristic information corresponding to each of the Q sampling periods is weighted to obtain the history OD traffic characteristic information.
In specific implementation, the external characteristic information of the time to be predicted may be obtained first, for example, the time characteristic information of the time to be predicted may be obtained, such as the time to be predicted is the hour of the day, the date to which the time to be predicted belongs is the day of the week, whether the time to be predicted is a holiday, and the like; in addition, information such as temperature, wind speed, weather state and the like at the moment to be predicted can be acquired as the external characteristic information, and the acquired external characteristic information is not limited in any way. Then, the acquired external feature information can be processed through One-Hot (One-Hot) coding, data obtained through the One-Hot coding processing are processed through a full connection layer in an external feature layer to obtain a tensor with the implicit external features, and then weights corresponding to the P sampling periods and weights corresponding to the Q sampling periods are determined through a classification (Softmax) layer in the external feature layer.
It should be understood that, in practical applications, if the sampling periods corresponding to the historical ingress and egress flow tensors and the historical OD flow tensors are the same, only one Softmax layer may be provided; if the sampling periods of the historical access flow tensors and the historical OD flow tensors are different, two Softmax layers need to be arranged, one Softmax layer is used for determining the weights corresponding to the P sampling periods corresponding to the historical access flow rates, and the other Softmax layer is used for determining the weights corresponding to the Q sampling periods corresponding to the historical OD flow tensors.
The external features mainly considered by the external feature layer in the embodiment of the present application are usually time features of the time to be predicted, and the basic idea is that, considering that different times to be predicted should have different weights in different sampling periods, for example, three types of historical access flow tensors, which respectively take an hour as a sampling period, a day as a sampling period, and a week as a sampling period, should have different reference weights, in one scene, the relevance of each type of traffic flow index at monday morning 10 to the traffic flow index at 10 am before one day (i.e., sunday) and at two days before (i.e., saturday) morning 10 may be low, and the relevance of each type of traffic flow index at 10 am before one week (i.e., monday) may be high; in another scenario, a traffic flow indicator at a certain time of day of a holiday (e.g., national festival holiday) may be less correlated with a traffic flow indicator at the same time one day or one week before, and more correlated with a traffic flow indicator one hour or two hours before the day. Therefore, the traffic flow prediction model in the embodiment of the application captures the time characteristic information of the time to be predicted by means of the external characteristic layer so as to determine reasonable information weight for different sampling periods.
Step 203: converting the historical access flow characteristic information into to-be-fused historical access flow characteristic information based on an indication matrix in the traffic flow prediction model, and splicing the to-be-fused historical access flow characteristic information and the historical OD flow characteristic information to obtain first fusion information; and/or converting the historical OD flow characteristic information into historical OD flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical OD flow characteristic information to be fused and the historical access flow characteristic information to obtain second fusion information; the indication matrix is determined according to the incidence relation between the target area and the target area pair.
After the server extracts the historical access flow characteristic information and the historical OD flow characteristic information through the traffic flow prediction model, the historical access flow characteristic information and the historical OD flow characteristic information can be fused by using an indication matrix in the traffic flow prediction model.
The indication matrix I is a predefined matrix with size N × M, which is constructed based on the geographical position correlation between the target region and the target region pair, and may specifically represent the relationship between the input historical ingress and egress flow tensor and the historical OD flow tensor. For example, assuming that the 51 st element in the input historical OD traffic matrix corresponds to the target zone pair (18,32) that is the traffic flowing from zone 18 to zone 32, then the initialized indication matrix I may be set to 1 for element I [18] [51] and element I [32] [51], the element I [18] [51] representing the 18 th row in the historical OD traffic tensor is associated with the 51 st element in the historical OD traffic tensor, and the element I [32] [51] representing the 32 nd row in the historical OD traffic tensor is also associated with the 51 th element in the historical OD traffic tensor. And by analogy, determining each element in the initialized indication matrix according to the incidence relation between the target block and the target block pair.
Specifically, when the historical ingress and egress traffic characteristic information and the historical OD traffic characteristic information are fused by the indication matrix, the fusion processing may be implemented by at least one of the following manners:
in the first mode, historical access flow characteristic information to be fused is obtained through calculation based on the indication matrix and the historical access flow characteristic information, and the historical access flow characteristic information to be fused and the historical OD flow characteristic information are spliced to obtain first fusion information.
The method comprises the steps of performing feature extraction processing on a historical access flow tensor to obtain historical access flow feature information, specifically, obtaining historical access flow feature tensor LA with the size of N × HA, wherein HA is the output size of the last layer in a first feature extraction layer, multiplying the historical access flow feature tensor LA by a transposed matrix of an indication matrix to obtain tensor LA 'corresponding to the historical access flow feature information to be fused, and obtaining tensor LA' with the size of M × HA, wherein the specific processing process is as shown in formula (2):
LA’=I.T*LA (2)
where I denotes the indicator matrix, I.T denotes the transposed matrix of the indicator matrix, and LA' is M × HA in size.
And splicing the historical OD flow characteristic tensor LB and a tensor LA' corresponding to the historical access flow characteristic information to be fused in the last dimension to obtain a tensor OB with the size of M (HA + HB), namely the first fusion information.
And in the second mode, historical OD flow characteristic information to be fused is obtained through calculation based on the indication matrix and the historical OD flow characteristic information, and the historical OD flow characteristic information to be fused and the historical access flow characteristic information are spliced to obtain second fusion information.
Obtaining historical OD flow characteristic information after characteristic extraction processing is carried out on the historical OD flow tensor, specifically, obtaining a historical OD flow characteristic tensor LB with the size of M HB, wherein HB is the output size of the last layer in the second characteristic extraction layer, multiplying the historical OD flow characteristic tensor LB by using an indication matrix to obtain a tensor LB 'corresponding to the historical OD flow characteristic information to be fused, wherein the size of the tensor LB' is N HB, and the specific processing process is shown as a formula (3):
LB’=I*LB (3)
and splicing the historical access flow characteristic tensor LA and the tensor LB' corresponding to the historical OD flow characteristic information to be fused in the last dimension to obtain a tensor OA with the size of N (HA + HB), namely the second fused information.
It should be noted that, in practical applications, the traffic flow prediction model may implement only the above-mentioned first mode to realize fusion of the historical ingress and egress flow characteristic information and the historical OD flow characteristic information, so as to obtain first fusion information; the traffic flow prediction model can also realize the fusion of the historical access flow characteristic information and the historical OD flow characteristic information only by executing the second mode to obtain second fusion information; the traffic flow prediction model can also execute the first mode and the second mode simultaneously to realize the fusion of the historical access flow characteristic information and the historical OD flow characteristic information to obtain first fusion information and second fusion information.
Step 204: and determining at least one of an in-out flow prediction result and an OD flow prediction result of the target geographic range at the moment to be predicted based on the first fusion information and/or the second fusion information through the traffic flow prediction model.
The server completes fusion processing on historical access flow characteristic information and historical OD flow characteristic information through an indication matrix in a traffic flow prediction model to obtain first fusion information and/or second fusion information, and then determines access flow prediction results and/or OD flow prediction results of a target geographic range at a to-be-predicted time through the traffic flow prediction model based on the first fusion information and/or the second fusion information obtained through fusion processing, namely determines access flow of each target area in the target geographic range at the to-be-predicted time and/or determines OD flow of each target area in the target geographic range at the to-be-predicted time.
It should be understood that the final obtained ingress and egress traffic prediction result should be an ingress and egress traffic prediction result tensor of size N x 2, wherein the elements are in one-to-one correspondence with the elements in the historical in-out flow tensor of a single preset time period input into the traffic flow prediction model, for example, it is assumed that an element with an index [ i,0] in the historical access flow tensor of the single preset period represents the historical access flow of the ith target area, an element with an index [ i,1] in the historical access flow tensor of the single preset period represents the historical egress flow of the ith target area, accordingly, an element with an index [ i,0] in the access flow prediction result tensor represents the access flow of the ith target area at the time to be predicted, and an element with an index [ i,1] in the access flow prediction result tensor represents the egress flow of the ith target area at the time to be predicted.
Similarly, the finally obtained OD traffic prediction result should be an OD traffic prediction result tensor of M × 1 size, where elements of the OD traffic prediction result tensor correspond to elements of the historical OD traffic tensor of the input traffic flow prediction model in a single preset time period in a one-to-one manner, for example, assuming that the jth element in the historical OD traffic tensor of the single preset time period represents the historical OD traffic of the jth target area pair, and accordingly, the jth element in the OD traffic prediction result tensor represents the OD traffic of the jth target area pair at the time to be predicted.
During specific implementation, on the basis of obtaining first fusion information and/or second fusion information through fusion processing, the first fusion information or the second fusion information can be processed through a first full-link layer in a traffic flow prediction model to obtain an in-out flow prediction result; the first fusion information or the second fusion information can be processed through a second full-link layer in the traffic flow prediction model to obtain an OD flow prediction result.
Preferably, when the first fusion information and the second fusion information are obtained simultaneously through the fusion processing and the ingress flow prediction result and the OD flow prediction result need to be determined, the first full-link layer may be used to process the first fusion information to obtain the ingress and egress flow prediction result, and the second full-link layer may be used to process the second fusion information to obtain the OD flow prediction result.
It should be noted that in the training process of the traffic flow prediction model in the embodiment of the present application, L2 Loss may be sampled as Loss tax, and the traffic flow prediction model may be trained by a gradient learning algorithm or other optimization algorithms, for example, Adam algorithm may be used as the learning optimization algorithm of the traffic flow prediction model.
In the traffic flow prediction method provided in the embodiment of the present application, a historical access flow tensor which is constructed based on historical access flow of a target area in a target geographic range and a historical OD flow tensor which is constructed based on historical OD flow of a target area pair in the target geographic range are obtained first; then, the traffic flow prediction model obtained through pre-training realizes the prediction of the entrance flow and the exit flow and/or the prediction of the OD flow in the target geographic range according to the historical entrance flow tensor and the historical OD flow tensor. The traffic flow prediction model is provided with an indication matrix constructed according to the geographical position incidence relation between the target area and the target area pair, the indication matrix is used for carrying out fusion processing on historical input and output flow characteristic information extracted from a historical input and output flow tensor and historical OD flow characteristic information extracted from a historical OD flow tensor so as to realize organic fusion between the input and output flow characteristic information and the OD flow characteristic information, then an input flow prediction result and/or an OD flow prediction result are determined based on the information obtained after the fusion processing, and the correlation between the input and output flow and the OD flow is fused into the traffic flow prediction process, so that the accuracy of traffic flow prediction is improved.
In order to further understand the technical solution provided by the embodiment of the present application, the following takes a neighborhood as a basic research unit, and combines the structure of the traffic flow prediction model shown in fig. 4 to integrally and exemplarily introduce the traffic flow prediction method provided by the embodiment of the present application.
Assuming that a target geographic range includes N target blocks and M target block pairs, currently, the flow rates of the N target blocks and the OD flow rates of the M target block pairs within the target geographic range at a time t to be predicted need to be predicted, and for a single historical preset time period, a historical access flow tensor with a size of N × 2 and a historical OD flow tensor with a size of M × 1 need to be constructed as inputs of a traffic flow prediction model. For the historical ingress and egress flow tensor of N x 2, the element with the index [ i,0] represents the historical ingress flow of the ith target block, and the element with the index [ i,1] represents the historical egress flow of the ith target block. For the historical OD traffic tensor of M x 1, the jth element represents the historical OD traffic for the jth target block pair.
It should be noted that, in this embodiment, based on the concept that the neighborhood pairs within the target geographic range have sparseness (that is, no traffic connection occurs between a large number of area pairs within the target geographic range), it is determined that the prediction target of the OD traffic should focus on the neighborhood pairs whose OD traffic mean values are greater than the preset threshold value within the historical training period, that is, M neighborhood pairs whose OD traffic mean values are greater than the preset threshold value within the target geographic range are selected as the target neighborhood pairs, and in general, M is much smaller than the number N of the neighborhood pairs actually included within the target geographic range, so that the OD traffic prediction based on large-scale data can be avoided, and the cost of computing resources is reduced.
In the technical scheme provided by the embodiment, the selection of the historical incoming and outgoing flow and the historical OD flow mainly comprises the following three ways: 1) selecting the flow of the target block in and out within one hour and two hours before the moment to be predicted by taking the hour as an acquisition period, and selecting the OD flow of the target block in one hour and two hours before the moment to be predicted; 2) selecting the flow of the target block in and out at the same time as the time to be predicted before one day and the same time as the time to be predicted before two days by taking the day as an acquisition period, and selecting the OD flow of the target block pair at the same time as the time to be predicted before one day and the same time as the time to be predicted before two days; 3) and taking a week as an acquisition cycle, selecting the incoming and outgoing flow of the target block in the time before the week and the same as the time to be predicted, and selecting the OD flow of the target block pair in the time before the week and the same as the time to be predicted, and in the time before the two weeks and the same as the time to be predicted.
Based on the historical access flow of the target block under different sampling periods, a historical access flow tensor 411 with an hour sampling period, a historical access flow tensor 412 with a day sampling period and a historical access flow tensor 413 with a week sampling period are respectively constructed and obtained as the input of the input layer 410 in the traffic flow prediction model. Based on the historical OD flows of the target block pairs in different sampling periods, a historical OD flow tensor 414 taking hours as a sampling period, a historical OD flow tensor 415 taking days as a sampling period, and a historical OD flow tensor 416 taking weeks as a sampling period are respectively constructed and obtained as inputs of the input layer 410 in the traffic flow prediction model.
Expanding historical access flow tensors corresponding to three different sampling periods in the input layer 410 according to time dimensions, and inputting the expanded historical access flow tensors into a first RNN layer 421, a first RNN layer 422 and a first RNN layer 423 in the RNN layer 420; the historical OD flow tensors corresponding to three different sampling periods in the input layer 410 are expanded according to the time dimension, and input into the second RNN layer 424, the second RNN layer 425, and the second RNN layer 426 in the RNN layer 420. To avoid loss of original information, the output of the last state of each first RNN layer in the RNN layer 420 will be spliced together with the historical incoming and outgoing traffic tensor of the corresponding input, and the output of the last state of each second RNN layer in the RNN layer 420 will be spliced together with the historical OD traffic tensor of the corresponding input.
For the historical access flow tensors corresponding to different sampling periods obtained through the splicing processing, the historical access flow tensors are respectively input into a first GCN layer 431, a first GCN layer 432 and a first GCN layer 433 in the GCN layer 430 by a traffic flow prediction model; for the historical OD flow tensors corresponding to different sampling periods obtained through the splicing processing, the traffic flow prediction model inputs the historical OD flow tensors into a second GCN layer 434, a second GCN layer 435 and a second GCN layer 436 of the GCN layer 430 respectively.
The number of layers of each first GCN layer and the number of layers of each second GCN layer in the GCN layers 430 are both super parameters of a traffic flow prediction model, and a processing formula of each GCN layer is shown in formula (1):
H(l)=σ(H(l-1)W(l)0+AH(l-1)W(l)1) (1)
wherein H(l)Denotes the output of the l-th layer in the GCN layer, H(l-1)The output of the l-1 layer in the GCN layer is represented, and the output of the l-1 layer in the GCN layer is the input of the l layer in the GCN layer; w(l)0And W(l)1Is the model parameter tensor in the GCN layer, which is constantly updated with the training of the model; a is an adjacency matrix used for representing the relation between the regions after normalization processing, the size of A in the first GCN layer is N x N, and the size of A in the second GCN layer is M x M; σ is the activation function.
When constructing the neighborhood adjacency matrix in the first GCN layer, the historical average OD traffic of N × N neighborhood pairs obtained by combining N target neighborhoods two by two may be used as an element in the adjacency matrix. For example, the initialized neighborhood matrix is a two-dimensional matrix E with size N × N, and if the historical average OD traffic from neighborhood 16 to neighborhood 12 is 5, E [16] [12] is 5. On the basis, the neighborhood adjacent matrix E is normalized, and the neighborhood adjacent matrix in the first GCN layer can be obtained.
When the OD adjacency matrix in the second GCN layer is constructed, elements in the OD adjacency matrix may be determined according to the combination of M × M block pairs obtained by combining M target block pairs two by two. For example, when the initialized OD adjacency matrix is a two-dimensional matrix S having a size of M × M and a target block pair (18,42) flowing from block 18 to block 42 is considered, if it is known that E [38] [18] is 5 and E [6] [18] is 12 in the history average OD traffic matrix E, S [ (38,18), (18,42) ] -5, S [ (6,18), (18,42) ] -12, and so on can be defined. The core idea is that traffic flowing from the starting point a to the end point B has a causal effect on traffic flowing from the starting point B to the end point C, and therefore the relative magnitude of this effect is determined based on the historical average OD traffic matrix E. On the basis, the OD adjacent matrix S is normalized, and the OD adjacent matrix in the second GCN layer can be obtained.
The sub-history OD flow space-time tensors corresponding to 3 different sampling periods are obtained after the sub-history OD flow space-time tensors are processed by the first GCN layer 431, the first GCN layer 432 and the first GCN layer 433, and the sub-history OD flow space-time tensors corresponding to 3 different sampling periods are obtained after the sub-history OD flow space-time tensors are processed by the second GCN layer 434, the second GCN layer 435 and the second GCN layer 436. The 3 weighted values generated by the external eigen layer 440 are used to perform weighted fusion processing on the 3 sub-historical access flow space-time tensors corresponding to different sampling periods respectively to obtain a historical access flow space-time tensor 451, and the 3 sub-historical OD flow space-time tensors corresponding to different sampling periods are performed weighted fusion processing respectively to obtain a historical OD flow space-time tensor 452.
In this embodiment, the extrinsic feature layer 440 mainly selects the temporal feature information 441 as the extrinsic feature information, and the selected temporal feature information 441 may include: the time to be predicted is the hour of the day, the day to which the time to be predicted belongs is the day of the week, whether the day to which the time to be predicted belongs is a holiday, and the like. The time characteristic information is processed through One-Hot coding and then used as input of an external characteristic layer 440, firstly, a tensor implying external characteristics is obtained through a full connection layer 442, and then, 3 weight values are obtained through a Softmax layer 443 and respectively represent the weight of historical traffic flow information with an hour sampling period, a day sampling period and a week sampling period.
In order to merge the historical access flow space-time tensor 451 and the historical OD flow space-time tensor 452, the information merging and output layer 450 in the traffic flow prediction model is provided with an information merging module 453 that performs information merging processing based on an indication matrix having a size of N × M, which means a relationship between the historical access flow matrix and the historical OD flow matrix. For example, in the historical OD traffic matrix, the 51 st element corresponds to the target block pair (18,32), meaning the traffic flowing from block 18 to block 32, then for the initialization indicator matrix I, one can define where I [18] [51] ═ 1 and I [32] [51] ═ 1, and so on, to determine the individual elements in the indicator matrix.
When the information fusion module 453 specifically performs information fusion, the historical ingress and egress flow space-time tensor 451 (a tensor having a size of N × HA, and HA is an output size of the last layer of the historical ingress and egress flow tensor on the GCN layer) may be multiplied by the transposed matrix I.T of the indication matrix I to obtain a historical ingress and egress flow tensor to be fused, which HAs a size of M × HA. The historical ingress and egress flow tensor to be fused and the historical OD flow space-time tensor 452 (the matrix with the size of M × HB, HB is the output size of the historical OD flow tensor at the last layer of the GCN layer) are spliced in the last dimension to obtain the first fusion tensor 454 with the size of M (HA + HB).
Conversely, for the historical OD traffic space-time tensor 452, the indication matrix I is multiplied by it to obtain the historical OD traffic tensor to be fused, whose size is N × HB. And splicing the historical OD flow tensor to be fused and the historical access flow space-time tensor 451 in the last dimension to obtain a second fusion tensor 455 with the size of N (HA + HB).
The first fusion tensor 454 and the second fusion tensor 455 both represent tensors in which historical access flow spatiotemporal feature information and historical OD flow spatiotemporal feature information are fused, and further, an access flow prediction result of a target geographical range at a time to be predicted is output through the first fully-connected layer 456 based on the first fusion tensor 454, and an OD flow prediction result of the target geographical range at the time to be predicted is output through the second fully-connected layer 457 based on the second fusion tensor 455.
The inventor conducts experiments on the traffic flow prediction model provided by the embodiment of the application based on the traffic flow data in Beijing, compares the traditional time series analysis model with some representative deep learning models applicable to the irregular block traffic flow prediction, and the results are shown in Table 1:
TABLE 1
Model (model) | RMSE |
HA | 41.46 |
ARIMA | 39.70 |
STGCN | 31.01 |
DCRNN | 27.85 |
MVGCN | 27.12 |
Model in the present application | 24.47 |
The ha (history average) is a model for predicting the traffic flow based on the historical average, the arima (automated Integrated Moving average) is a differential Moving regression model, the stgcn (spatial Temporal Graph conditional Network) is a space-time convolution Neural Network model, the dcrnn (diffusion conditional recursive Neural Network) is a diffusion convolution Neural Network model, and the MVGCN (Multi-View Graph conditional Network) is a Multi-Graph convolution Neural Network model.
Through comparison, the traffic flow prediction model provided by the application is obviously superior to other models in Root Mean Square Error (RMSE) index.
In terms of the effect of the prediction experiment of the OD flow in the neighborhood, a conventional time series analysis model (here, a Seasonal automated guided Moving Average (SARIMA) with a better effect is selected as the Baseline) is compared with the traffic flow prediction model provided by the present application, and the results are shown in table 2:
TABLE 2
Model (model) | RMSE |
Average OD>=2,SARIMA | 25.03 |
Average OD>2, model in the present application | 4.91 |
Average OD>=50,SARIMA | 31.31 |
Average OD>Model in this application 50 | 20.30 |
Through comparison, the prediction effect of the traffic flow prediction model provided by the embodiment of the application is ahead of SARIMA, and the advantage of the leading is relatively more obvious when the average OD flow of the predicted region pair is smaller, so that the traffic flow prediction model provided by the embodiment of the application has obvious advantages in the aspects of capturing space-time correlation, periodic correlation and the like.
Aiming at the traffic flow prediction method described above, the application also provides a corresponding traffic flow prediction device, so that the traffic flow prediction method can be applied and realized in practice.
Referring to fig. 5, fig. 5 is a schematic view showing a traffic flow prediction apparatus 500 corresponding to the traffic flow prediction method shown in fig. 2, the apparatus including:
an obtaining module 501, configured to obtain a historical ingress and egress traffic tensor and a historical origin and destination OD traffic tensor of a target geographic range before a time to be predicted; the historical access flow tensor is obtained by construction according to historical access flow of a target area in the target geographic range, and the historical OD flow tensor is obtained by construction according to historical OD flow of a target area pair in the target geographic range;
a feature extraction module 502, configured to extract, according to the historical ingress and egress flow tensor and the historical OD flow tensor, historical ingress and egress flow feature information and historical OD flow feature information through a traffic flow prediction model;
an information fusion module 503, configured to convert the historical access flow characteristic information into historical access flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splice the historical access flow characteristic information to be fused and the historical OD flow characteristic information to obtain first fusion information; and/or converting the historical OD flow characteristic information into historical OD flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical OD flow characteristic information to be fused and the historical access flow characteristic information to obtain second fusion information; the indication matrix is determined according to the incidence relation between the target area and the target area pair;
a prediction result determining module 504, configured to determine, by the traffic flow prediction model, at least one of an input/output flow prediction result and an OD flow prediction result of the target geographic range at the time to be predicted based on the first fusion information and/or the second fusion information.
Optionally, on the basis of the traffic flow prediction apparatus shown in fig. 5, the regions within the target geographic range are divided according to blocks, the target regions are target blocks, and the target region pairs are target block pairs; the traffic flow prediction model includes: a first GCN layer and a second GCN layer. Referring to fig. 6, fig. 6 is a schematic structural diagram of another traffic flow prediction apparatus 600 according to an embodiment of the present application, in which the feature extraction module 502 includes:
a first feature extraction submodule 601, configured to process the historical access flow tensor through the first GCN layer to obtain a historical access flow time-space tensor as the historical access flow feature information;
a second feature extraction submodule 602, configured to process the historical OD traffic tensor through the second GCN layer to obtain a historical OD traffic space-time tensor, which is used as the historical OD traffic feature information.
Optionally, on the basis of the traffic flow prediction apparatus shown in fig. 6, the traffic flow prediction model further includes: a first Recurrent Neural Network (RNN) layer and a second RNN layer; the access flow characteristic extraction sub-module 601 is specifically configured to: processing the historical access flow tensor through the first RNN layer to obtain a historical access flow time sequence tensor; splicing the historical access flow time sequence tensor and the historical access flow tensor; processing the spliced tensor through the first GCN layer to obtain the historical access flow space-time tensor which is used as the historical access flow characteristic information;
the OD traffic feature extraction sub-module 602 is specifically configured to: processing the historical OD flow tensor through the second RNN layer to obtain a historical OD flow time sequence tensor; splicing the historical OD traffic time sequence tensor and the historical OD traffic tensor; and processing the spliced tensor through the second GCN layer to obtain a historical OD flow space-time tensor which is used as the historical OD flow characteristic information.
Optionally, in addition to the traffic flow prediction device shown in fig. 5, the historical ingress and egress flow of the target area includes: the flow rate of the inlet and the outlet in P sampling periods is greater than 1; the historical OD traffic for the target zone pair includes: and the OD flow under Q sampling periods is an integer greater than 1. Referring to fig. 7, fig. 7 is a schematic structural diagram of another traffic flow prediction apparatus 700 according to an embodiment of the present application, where the obtaining module 501 includes:
a first obtaining submodule 701, configured to construct, for each sampling period of the P sampling periods, a historical access flow tensor corresponding to the sampling period according to the access flow in the sampling period;
the second obtaining sub-module 702 is configured to, for each sampling period of the Q sampling periods, construct a historical OD traffic tensor corresponding to the sampling period according to the OD traffic in the sampling period.
Optionally, on the basis of the traffic flow prediction apparatus shown in fig. 7, the historical access flow tensor corresponding to each sampling period in the P sampling periods is constructed according to the access flow in the first target time period in the sampling period; the first target time interval is determined according to the time to be predicted; the historical OD flow tensor corresponding to each sampling period in the Q sampling periods is constructed according to the OD flow in the second target time period under the sampling period; the second target time period is determined according to the time to be predicted.
Optionally, on the basis of the traffic flow prediction apparatus shown in fig. 7, the traffic flow prediction model includes: p first feature extraction layers and Q second feature extraction layers, P first feature extraction layers with P kind of sampling period one-to-one, Q second feature extraction layers with Q kind of sampling period one-to-one. Referring to fig. 8, fig. 8 is a schematic structural diagram of another traffic flow prediction apparatus 800 according to an embodiment of the present application, in which the feature extraction module 502 includes:
a third feature extraction submodule 801, configured to, for each sampling period in the P sampling periods, process the historical access flow tensor corresponding to the sampling period through the first feature extraction layer corresponding to the sampling period, and obtain sub-historical access flow feature information corresponding to the sampling period;
a first fusion submodule 802, configured to perform fusion processing on sub-history access flow characteristic information corresponding to each of the P sampling periods to obtain the history access flow characteristic information;
a fourth feature extraction submodule 803, configured to, for each sampling period in the Q sampling periods, process the historical OD traffic tensor corresponding to the sampling period through the second feature extraction layer corresponding to the sampling period, and obtain sub-historical OD traffic feature information corresponding to the sampling period;
and a second fusion submodule 804, configured to perform fusion processing on the sub-historical OD traffic characteristic information corresponding to each of the Q sampling periods, so as to obtain the historical OD traffic characteristic information.
Optionally, on the basis of the traffic flow prediction apparatus shown in fig. 8, the first feature extraction layer includes: at least one of a first RNN layer and a first GCN layer; the second feature extraction layer includes: at least one of a second RNN layer and a second GCN layer.
Optionally, on the basis of the traffic flow prediction apparatus shown in fig. 8, the traffic flow prediction model further includes: and the external characteristic layer is used for acquiring external characteristic information of the moment to be predicted, determining weights corresponding to the P types of sampling periods according to the external characteristic information, and determining weights corresponding to the Q types of sampling periods according to the external characteristic information.
The first fusion submodule 802 is specifically configured to: weighting the sub-history access flow characteristic information corresponding to the P sampling periods by using the weights corresponding to the P sampling periods to obtain the history access flow characteristic information;
the second fusion submodule 804 is specifically configured to: and weighting the sub-history OD flow characteristic information corresponding to the Q sampling periods by using the weights corresponding to the Q sampling periods to obtain the history OD flow characteristic information.
Optionally, on the basis of the traffic flow prediction apparatus shown in fig. 6, referring to fig. 9, fig. 9 is a schematic structural diagram of another traffic flow prediction apparatus 900 provided in an embodiment of the present application, where the feature extraction module 502 further includes:
a combining submodule 901, configured to combine pairs of target area pairs within the target geographic range to obtain a target area pair combination;
an adjacency matrix construction sub-module 902, configured to determine, according to the OD traffic of the previous region pair in the target region pair combination within the historical training period, an element of an adjacency matrix in the second GCN layer if an end region included in the previous target region pair in the target region pair combination is the same as a start region included in the next target region pair.
Optionally, on the basis of the traffic flow predicting apparatus shown in fig. 5, referring to fig. 10, fig. 10 is a schematic structural diagram of another traffic flow predicting apparatus 1000 provided in the embodiment of the present application, and the apparatus further includes:
a historical data obtaining module 1001, configured to obtain an OD traffic of each region pair in the target geographic range in a historical training period;
a target area pair screening module 1002, configured to select, from the area pairs in the target geographic range, an area pair whose OD traffic in the historical training period meets a preset condition as the target area pair.
Optionally, on the basis of the traffic flow prediction apparatus shown in fig. 5, the prediction result determining module 504 is specifically configured to:
processing the first fusion information or the second fusion information through a first full-link layer in the traffic flow prediction model to obtain the input and output flow prediction result;
and/or processing the first fusion information or the second fusion information through a second full-link layer in the traffic flow prediction model to obtain the OD flow prediction result.
In the traffic flow prediction apparatus provided in the embodiment of the present application, a historical access flow tensor which is constructed based on the historical access flow of the target area in the target geographic range and a historical OD flow tensor which is constructed based on the historical OD flow of the target area pair in the target geographic range are obtained first; then, the traffic flow prediction model obtained through pre-training realizes the prediction of the entrance flow and the exit flow and/or the prediction of the OD flow in the target geographic range according to the historical entrance flow tensor and the historical OD flow tensor. The traffic flow prediction model is provided with an indication matrix constructed according to the geographical position incidence relation between the target area and the target area pair, the indication matrix is used for carrying out fusion processing on historical input and output flow characteristic information extracted from a historical input and output flow tensor and historical OD flow characteristic information extracted from a historical OD flow tensor so as to realize organic fusion between the input and output flow characteristic information and the OD flow characteristic information, then an input flow prediction result and/or an OD flow prediction result are determined based on the information obtained after the fusion processing, and the correlation between the input and output flow and the OD flow is fused into the traffic flow prediction process, so that the accuracy of traffic flow prediction is improved.
The embodiment of the present application further provides a device for predicting traffic flow, which may specifically be a server and a terminal device, and the server and the terminal device provided in the embodiment of the present application will be described below from the perspective of hardware materialization.
Referring to fig. 11, fig. 11 is a schematic structural diagram of a server 1100 according to an embodiment of the present disclosure. The server 1100 may vary widely in configuration or performance and may include one or more Central Processing Units (CPUs) 1122 (e.g., one or more processors) and memory 1132, one or more storage media 1130 (e.g., one or more mass storage devices) storing applications 1142 or data 1144. Memory 1132 and storage media 1130 may be, among other things, transient storage or persistent storage. The program stored on the storage medium 1130 may include one or more modules (not shown), each of which may include a series of instruction operations for the server. Still further, the central processor 1122 may be provided in communication with the storage medium 1130 to execute a series of instruction operations in the storage medium 1130 on the server 1100.
The server 1100 may also include one or more power supplies 1126, one or more wired or wireless network interfaces 1150, one or more input-output interfaces 1158, and/or one or more operating systems 1141, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, and so forth.
The steps performed by the server in the above embodiment may be based on the server structure shown in fig. 11.
The CPU 1122 is configured to execute the following steps:
acquiring a historical access flow tensor and a historical start and end point OD flow tensor of a target geographic range before a time to be predicted; the historical access flow tensor is obtained by construction according to historical access flow of a target area in the target geographic range, and the historical OD flow tensor is obtained by construction according to historical OD flow of a target area pair in the target geographic range;
extracting historical access flow characteristic information and historical OD flow characteristic information through a traffic flow prediction model according to the historical access flow tensor and the historical OD flow tensor;
converting the historical access flow characteristic information into to-be-fused historical access flow characteristic information based on an indication matrix in the traffic flow prediction model, and splicing the to-be-fused historical access flow characteristic information and the historical OD flow characteristic information to obtain first fusion information; and/or converting the historical OD flow characteristic information into historical OD flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical OD flow characteristic information to be fused and the historical access flow characteristic information to obtain second fusion information; the indication matrix is determined according to the incidence relation between the target area and the target area pair;
and determining at least one of an in-out flow prediction result and an OD flow prediction result of the target geographic range at the moment to be predicted based on the first fusion information and/or the second fusion information through the traffic flow prediction model.
Optionally, the CPU 1122 may also be configured to execute the steps of any implementation manner of the traffic flow prediction method provided in the embodiment of the present application.
Referring to fig. 12, fig. 12 is a schematic structural diagram of a terminal device according to an embodiment of the present application. For convenience of explanation, only the parts related to the embodiments of the present application are shown, and details of the specific technology are not disclosed. The terminal may be any terminal device including a computer, a tablet computer, a Personal Digital Assistant (PDA), and the like, taking the terminal as the computer as an example:
fig. 12 is a block diagram showing a partial structure of a computer related to a terminal provided in an embodiment of the present application. Referring to fig. 12, the computer includes: radio Frequency (RF) circuit 1210, memory 1220, input unit 1230, display unit 1240, sensor 1250, audio circuit 1260, wireless fidelity (WiFi) module 1270, processor 1280, and power supply 1290. Those skilled in the art will appreciate that the computer architecture shown in FIG. 12 is not intended to be limiting of computers, and may include more or fewer components than those shown, or some components in combination, or a different arrangement of components.
The memory 1220 may be used to store software programs and modules, and the processor 1280 performs various functional applications of the computer and data processing by operating the software programs and modules stored in the memory 1220. The memory 1220 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, a phonebook, etc.) created according to the use of the computer, etc. Further, the memory 1220 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.
The processor 1280 is a control center of the computer, connects various parts of the entire computer using various interfaces and lines, performs various functions of the computer and processes data by running or executing software programs and/or modules stored in the memory 1220 and calling data stored in the memory 1220, thereby monitoring the entire computer. Optionally, processor 1280 may include one or more processing units; preferably, the processor 1280 may integrate an application processor, which mainly handles operating systems, user interfaces, application programs, etc., and a modem processor, which mainly handles wireless communications. It is to be appreciated that the modem processor described above may not be integrated into the processor 1280.
In this embodiment, the processor 1280 included in the terminal further has the following functions:
acquiring a historical access flow tensor and a historical start and end point OD flow tensor of a target geographic range before a time to be predicted; the historical access flow tensor is obtained by construction according to historical access flow of a target area in the target geographic range, and the historical OD flow tensor is obtained by construction according to historical OD flow of a target area pair in the target geographic range;
extracting historical access flow characteristic information and historical OD flow characteristic information through a traffic flow prediction model according to the historical access flow tensor and the historical OD flow tensor;
converting the historical access flow characteristic information into to-be-fused historical access flow characteristic information based on an indication matrix in the traffic flow prediction model, and splicing the to-be-fused historical access flow characteristic information and the historical OD flow characteristic information to obtain first fusion information; and/or converting the historical OD flow characteristic information into historical OD flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical OD flow characteristic information to be fused and the historical access flow characteristic information to obtain second fusion information; the indication matrix is determined according to the incidence relation between the target area and the target area pair;
and determining at least one of an in-out flow prediction result and an OD flow prediction result of the target geographic range at the moment to be predicted based on the first fusion information and/or the second fusion information through the traffic flow prediction model.
Optionally, the processor 1280 is further configured to execute the steps of any implementation manner of the traffic flow prediction method provided in the embodiment of the present application.
The embodiment of the present application further provides a computer-readable storage medium for storing a computer program, where the computer program is used to execute any one implementation of the traffic flow prediction method described in the foregoing embodiments.
The present application also provides a computer program product including instructions, which when run on a computer, causes the computer to execute any one of the embodiments of a traffic flow prediction method described in the foregoing embodiments.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (15)
1. An artificial intelligence based traffic flow prediction method, characterized in that the method comprises:
acquiring a historical access flow tensor and a historical OD flow tensor of a target geographic range before a time to be predicted; the historical access flow tensor is obtained by construction according to historical access flow of a target area in the target geographic range, the historical OD flow tensor is obtained by construction according to historical OD flow of a target area pair in the target geographic range, and the target area pair comprises two different target areas;
extracting historical access flow characteristic information and historical OD flow characteristic information through a traffic flow prediction model according to the historical access flow tensor and the historical OD flow tensor;
converting the historical access flow characteristic information into to-be-fused historical access flow characteristic information based on an indication matrix in the traffic flow prediction model, and splicing the to-be-fused historical access flow characteristic information and the historical OD flow characteristic information to obtain first fusion information; and/or converting the historical OD flow characteristic information into historical OD flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical OD flow characteristic information to be fused and the historical access flow characteristic information to obtain second fusion information; the indication matrix is determined according to the incidence relation between the target area and the target area pair, wherein the incidence relation is a geographical position incidence relation;
and determining at least one of an in-out flow prediction result and an OD flow prediction result of the target geographic range at the moment to be predicted based on the first fusion information and/or the second fusion information through the traffic flow prediction model.
2. The method of claim 1, wherein the regions within the target geographic area are divided according to neighborhoods, the target regions are target neighborhoods, and the target region pairs are target neighborhood pairs; the traffic flow prediction model includes: a first GCN layer and a second GCN layer; then, the extracting historical ingress and egress flow characteristic information and historical OD flow characteristic information by the traffic flow prediction model includes:
processing the historical access flow tensor through the first GCN layer to obtain a historical access flow space-time tensor which is used as the historical access flow characteristic information; and processing the historical OD flow tensor through the second GCN layer to obtain a historical OD flow space-time tensor which is used as the historical OD flow characteristic information.
3. The method according to claim 2, wherein the traffic flow prediction model further comprises: a first RNN layer and a second RNN layer; then, the extracting historical ingress and egress flow characteristic information and historical OD flow characteristic information by the traffic flow prediction model includes:
processing the historical access flow tensor through the first RNN layer to obtain a historical access flow time sequence tensor; splicing the historical access flow time sequence tensor and the historical access flow tensor; processing the spliced tensor through the first GCN layer to obtain the historical access flow space-time tensor which is used as the historical access flow characteristic information;
processing the historical OD flow tensor through the second RNN layer to obtain a historical OD flow time sequence tensor; splicing the historical OD traffic time sequence tensor and the historical OD traffic tensor; and processing the spliced tensor through the second GCN layer to obtain a historical OD flow space-time tensor which is used as the historical OD flow characteristic information.
4. The method of claim 1, wherein the historical ingress and egress traffic of the target area comprises: the flow rate of the inlet and the outlet in P sampling periods is greater than 1; the historical OD traffic for the target zone pair includes: OD flow under Q sampling periods, wherein Q is an integer greater than 1; then, the obtaining of the historical ingress and egress traffic tensor and the historical origin and destination OD traffic tensor of the target geographic range before the time to be predicted includes:
for each sampling period in the P types of sampling periods, constructing a historical access flow tensor corresponding to the sampling period according to the access flow in the sampling period;
and aiming at each sampling period in the Q types of sampling periods, constructing a historical OD flow tensor corresponding to the sampling period according to the OD flow under the sampling period.
5. The method according to claim 4, wherein the historical ingress and egress traffic tensor corresponding to each of the P sampling periods is constructed according to the ingress and egress traffic in the first target period under the sampling period; the first target time interval is determined according to the time to be predicted;
the historical OD flow tensor corresponding to each sampling period in the Q sampling periods is constructed according to the OD flow in the second target time period under the sampling period; the second target time period is determined according to the time to be predicted.
6. The method according to claim 4, wherein the traffic flow prediction model comprises: the P first feature extraction layers correspond to the P sampling periods one by one, and the Q second feature extraction layers correspond to the Q sampling periods one by one; then, the extracting historical ingress and egress flow characteristic information and historical OD flow characteristic information by the traffic flow prediction model includes:
for each sampling period in the P types of sampling periods, processing the historical access flow tensor corresponding to the sampling period through a first feature extraction layer corresponding to the sampling period to obtain sub-historical access flow feature information corresponding to the sampling period;
fusing the sub-historical access flow characteristic information corresponding to the P sampling periods respectively to obtain the historical access flow characteristic information;
for each sampling period in the Q sampling periods, processing the historical OD flow tensor corresponding to the sampling period through a second feature extraction layer corresponding to the sampling period to obtain sub-historical OD flow feature information corresponding to the sampling period;
and performing fusion processing on the sub-historical OD flow characteristic information corresponding to the Q sampling periods respectively to obtain the historical OD flow characteristic information.
7. The method of claim 6, wherein the first feature extraction layer comprises: at least one of a first RNN layer and a first GCN layer;
the second feature extraction layer includes: at least one of a second RNN layer and a second GCN layer.
8. The method according to claim 6, wherein the traffic flow prediction model further comprises: the external feature layer is used for acquiring external feature information of the moment to be predicted, determining weights corresponding to the P sampling periods according to the external feature information, and determining weights corresponding to the Q sampling periods according to the external feature information;
then, the fusing the sub-historical access flow characteristic information corresponding to each of the P sampling periods to obtain the historical access flow characteristic information includes:
weighting the sub-history access flow characteristic information corresponding to the P sampling periods by using the weights corresponding to the P sampling periods to obtain the history access flow characteristic information;
performing fusion processing on the sub-historical OD traffic characteristic information corresponding to each of the Q sampling periods to obtain the historical OD traffic characteristic information, including:
and weighting the sub-history OD flow characteristic information corresponding to the Q sampling periods by using the weights corresponding to the Q sampling periods to obtain the history OD flow characteristic information.
9. The method according to claim 2 or 7, wherein the adjacency matrix in the second GCN layer is determined by:
combining the target area pairs in the target geographic range pairwise to obtain a target area pair combination;
and if the destination region contained in the previous destination region pair in the destination region pair combination is the same as the starting region contained in the next destination region pair, determining the elements of the adjacent matrix in the second GCN layer according to the OD traffic of the previous region pair in the destination region pair combination in the historical training period.
10. The method of claim 1, wherein the target area pair is determined by:
obtaining OD flow of each region pair in the target geographic range in a historical training period;
and selecting the area pair with the OD flow meeting the preset condition in the historical training period from the area pairs in the target geographic range as the target area pair.
11. The method of claim 1, wherein said determining an incoming and outgoing traffic prediction for said target geographic area at said time to be predicted comprises:
processing the first fusion information or the second fusion information through a first full-link layer in the traffic flow prediction model to obtain the input and output flow prediction result;
the determining of the OD traffic prediction result of the target geographical range at the time to be predicted includes:
and processing the first fusion information or the second fusion information through a second full-link layer in the traffic flow prediction model to obtain the OD flow prediction result.
12. An artificial intelligence-based traffic flow prediction apparatus, characterized in that the apparatus comprises:
the acquisition module is used for acquiring a historical access flow tensor and a historical OD flow tensor of the target geographic range before the time to be predicted; the historical access flow tensor is obtained by construction according to historical access flow of a target area in the target geographic range, the historical OD flow tensor is obtained by construction according to historical OD flow of a target area pair in the target geographic range, and the target area pair comprises two different target areas;
the characteristic extraction module is used for extracting historical access flow characteristic information and historical OD flow characteristic information through a traffic flow prediction model according to the historical access flow tensor and the historical OD flow tensor;
the information fusion module is used for converting the historical access flow characteristic information into historical access flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical access flow characteristic information to be fused and the historical OD flow characteristic information to obtain first fusion information; and/or converting the historical OD flow characteristic information into historical OD flow characteristic information to be fused based on an indication matrix in the traffic flow prediction model, and splicing the historical OD flow characteristic information to be fused and the historical access flow characteristic information to obtain second fusion information; the indication matrix is determined according to the incidence relation between the target area and the target area pair, wherein the incidence relation is a geographical position incidence relation;
and the prediction result determining module is used for determining at least one of an access flow prediction result and an OD flow prediction result of the target geographic range at the moment to be predicted based on the first fusion information and/or the second fusion information through the traffic flow prediction model.
13. The apparatus of claim 12, wherein the regions within the target geographic area are divided according to neighborhoods, the target regions are target neighborhoods, and the target region pairs are target neighborhood pairs; the traffic flow prediction model includes: a first GCN layer and a second GCN layer; the feature extraction module comprises:
the first feature extraction submodule is used for processing the historical access flow tensor through the first GCN layer to obtain a historical access flow space-time tensor which is used as the historical access flow feature information;
and the second feature extraction submodule is used for processing the historical OD flow tensor through the second GCN layer to obtain a historical OD flow space-time tensor which is used as the historical OD flow feature information.
14. An apparatus for predicting traffic flow, the apparatus comprising a processor and a memory;
the memory is used for storing a computer program;
the processor is configured to execute the traffic flow prediction method according to any one of claims 1 to 11 in accordance with the computer program.
15. A computer-readable storage medium for storing a computer program for executing the traffic-flow prediction method according to any one of claims 1 to 11.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911203515.0A CN110910659B (en) | 2019-11-29 | 2019-11-29 | Traffic flow prediction method, device, equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911203515.0A CN110910659B (en) | 2019-11-29 | 2019-11-29 | Traffic flow prediction method, device, equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110910659A CN110910659A (en) | 2020-03-24 |
CN110910659B true CN110910659B (en) | 2021-08-17 |
Family
ID=69821009
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911203515.0A Active CN110910659B (en) | 2019-11-29 | 2019-11-29 | Traffic flow prediction method, device, equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110910659B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113496305A (en) * | 2020-04-03 | 2021-10-12 | 浙江大学 | Data processing method and device |
CN111612670B (en) * | 2020-04-27 | 2024-03-15 | 清华大学 | Method, device and computer equipment for constructing emission list of motor vehicle |
CN112488419B (en) * | 2020-12-15 | 2023-02-10 | 平安国际智慧城市科技股份有限公司 | Passenger flow distribution prediction method, device, equipment and storage medium based on OD analysis |
CN112712695B (en) * | 2020-12-30 | 2021-11-26 | 桂林电子科技大学 | Traffic flow prediction method, device and storage medium |
CN112767692A (en) * | 2020-12-30 | 2021-05-07 | 兰州理工大学 | Short-term traffic flow prediction system based on SARIMA-GA-Elman combined model |
CN112633608B (en) * | 2021-01-06 | 2022-04-12 | 南方科技大学 | People flow transfer prediction method, device, equipment and storage medium |
CN113011501B (en) * | 2021-03-22 | 2022-05-24 | 广东海启星海洋科技有限公司 | Method and device for predicting typhoon water level based on graph convolution neural network |
CN113326449B (en) * | 2021-05-27 | 2023-07-25 | 北京百度网讯科技有限公司 | Method, device, electronic equipment and medium for predicting traffic flow |
CN113590962B (en) * | 2021-08-03 | 2023-07-25 | 合肥泰瑞数创科技有限公司 | Flow data prediction system, method, computer equipment and medium |
CN113744525A (en) * | 2021-08-17 | 2021-12-03 | 东南大学 | Traffic distribution prediction method based on feature extraction and deep learning |
CN114327840B (en) * | 2022-03-11 | 2022-06-10 | 北京骑胜科技有限公司 | Data processing method and data processing device |
CN114579643A (en) * | 2022-05-05 | 2022-06-03 | 国家邮政局邮政业安全中心 | Express delivery traffic prediction method and device and electronic equipment |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO1999026210A1 (en) * | 1997-11-18 | 1999-05-27 | DDG GESELLSCHAFT FüR VERKEHRSDATEN MBH | Method for predicting a parameter representing the state of a system, especially a traffic parameter representing the state of a traffic network, and a device for carrying out said method |
JP2004355537A (en) * | 2003-05-30 | 2004-12-16 | Matsushita Electric Ind Co Ltd | Traffic flow data prediction device and method |
CN108230698A (en) * | 2016-12-21 | 2018-06-29 | 杭州海康威视数字技术股份有限公司 | traffic flow forecasting method and device |
CN109559512A (en) * | 2018-12-05 | 2019-04-02 | 北京掌行通信息技术有限公司 | A kind of regional traffic flow prediction technique and device |
CN109598381A (en) * | 2018-12-05 | 2019-04-09 | 武汉理工大学 | A kind of Short-time Traffic Flow Forecasting Methods based on state frequency Memory Neural Networks |
-
2019
- 2019-11-29 CN CN201911203515.0A patent/CN110910659B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO1999026210A1 (en) * | 1997-11-18 | 1999-05-27 | DDG GESELLSCHAFT FüR VERKEHRSDATEN MBH | Method for predicting a parameter representing the state of a system, especially a traffic parameter representing the state of a traffic network, and a device for carrying out said method |
JP2004355537A (en) * | 2003-05-30 | 2004-12-16 | Matsushita Electric Ind Co Ltd | Traffic flow data prediction device and method |
CN108230698A (en) * | 2016-12-21 | 2018-06-29 | 杭州海康威视数字技术股份有限公司 | traffic flow forecasting method and device |
CN109559512A (en) * | 2018-12-05 | 2019-04-02 | 北京掌行通信息技术有限公司 | A kind of regional traffic flow prediction technique and device |
CN109598381A (en) * | 2018-12-05 | 2019-04-09 | 武汉理工大学 | A kind of Short-time Traffic Flow Forecasting Methods based on state frequency Memory Neural Networks |
Also Published As
Publication number | Publication date |
---|---|
CN110910659A (en) | 2020-03-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110910659B (en) | Traffic flow prediction method, device, equipment and storage medium | |
Tekouabou et al. | Improving parking availability prediction in smart cities with IoT and ensemble-based model | |
Li et al. | Autost: Efficient neural architecture search for spatio-temporal prediction | |
Wang et al. | Deep learning for real time crime forecasting | |
CN111860951A (en) | Rail transit passenger flow prediction method based on dynamic hypergraph convolutional network | |
CN110570035B (en) | People flow prediction system for simultaneously modeling space-time dependency and daily flow dependency | |
CN113257016B (en) | Traffic signal control method and device and readable storage medium | |
Zhang et al. | Multistep speed prediction on traffic networks: A graph convolutional sequence-to-sequence learning approach with attention mechanism | |
CN115204478A (en) | Public traffic flow prediction method combining urban interest points and space-time causal relationship | |
CN111242292A (en) | OD data prediction method and system based on deep space-time network | |
CN112911626A (en) | Wireless network flow prediction method based on multi-graph convolution | |
JP2021531523A (en) | Image processing method, candidate evaluation method and related equipment | |
CN111242395A (en) | Method and device for constructing prediction model for OD (origin-destination) data | |
CN113887704A (en) | Traffic information prediction method, device, equipment and storage medium | |
CN115018039A (en) | Neural network distillation method, target detection method and device | |
CN111444302A (en) | Mobility prediction method, system and device based on user classification | |
CN116486624A (en) | Traffic flow prediction method and system based on space-time diagram convolutional neural network | |
Bansal et al. | Htfm: Hybrid traffic-flow forecasting model for intelligent vehicular ad hoc networks | |
CN115423162A (en) | Traffic flow prediction method and device, electronic equipment and storage medium | |
Wang et al. | A novel time efficient machine learning-based traffic flow prediction method for large scale road network | |
He et al. | Forecasting nationwide passenger flows at city-level via a spatiotemporal deep learning approach | |
CN117195976A (en) | Traffic flow prediction method and system based on layered attention | |
CN117037483A (en) | Traffic flow prediction method based on multi-head attention mechanism | |
Ma et al. | Cellular traffic prediction via deep state space models with attention mechanism | |
Liu et al. | Acfm: A dynamic spatial-temporal network for traffic prediction |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40022966 Country of ref document: HK |
|
GR01 | Patent grant | ||
GR01 | Patent grant |