CN113962460B - Urban fine granularity flow prediction method and system based on space-time comparison self-supervision - Google Patents
Urban fine granularity flow prediction method and system based on space-time comparison self-supervision Download PDFInfo
- Publication number
- CN113962460B CN113962460B CN202111233589.6A CN202111233589A CN113962460B CN 113962460 B CN113962460 B CN 113962460B CN 202111233589 A CN202111233589 A CN 202111233589A CN 113962460 B CN113962460 B CN 113962460B
- Authority
- CN
- China
- Prior art keywords
- flow distribution
- coarse
- granularity
- fine
- distribution map
- 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
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000010586 diagram Methods 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 30
- 230000006870 function Effects 0.000 claims description 12
- 230000002123 temporal effect Effects 0.000 claims description 9
- 238000005259 measurement Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 239000011362 coarse particle Substances 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 239000010419 fine particle Substances 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Educational Administration (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a city fine granularity flow prediction method and a system based on space-time comparison self-supervision, wherein the method comprises the following steps: obtaining a plurality of flow distribution diagrams of a city to be predicted for a period of time, and constructing a fine granularity flow distribution diagram and a coarse granularity flow distribution diagram; comparing and self-supervising learning is carried out on the coarse granularity flow distribution map according to the similarity between the areas, so that a space encoder is obtained; comparing and self-supervising learning is carried out on the coarse-granularity flow distribution map at a specific moment according to the similarity between the coarse-granularity flow distribution map at a specific moment and the flow distribution map at other moments to obtain a time encoder; and training a fine-grain flow prediction model for predicting the regional fine-grain flow based on the fine-grain flow distribution diagram and the coarse-grain flow distribution diagram. The method fully considers the similarity of regional flow in the space dimension and the similarity of global flow distribution in the time dimension, thereby improving the prediction accuracy.
Description
Technical Field
The invention belongs to the technical field of intelligent traffic, and particularly relates to a city fine granularity flow prediction method and system based on space-time comparison self-supervision.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
Smart cities are applications of artificial intelligence technology to traffic flow data. The system comprises the aspects of road flow monitoring, traffic signal control, traffic information acquisition and guidance and intelligent traffic. Wherein intelligent transportation is taken as an important component. The method needs fine-grained urban traffic monitoring on the local global, and is beneficial to planning urban road sections and reducing traffic jam risks.
At present, the technology for monitoring urban traffic is characterized in that a large number of sensor devices for monitoring traffic are deployed in the city, but the running and maintenance costs are high; on the other hand, the urban traffic data is structured by an image super-resolution method. The existing fine granularity flow prediction method has the problems of complex model, excessive parameter quantity, long training period and the like, and the problems of space-time characteristics of flow data and the like are not considered.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a city fine granularity flow prediction method and a city fine granularity flow prediction system based on space-time comparison self-supervision, a space encoder and a time encoder are constructed based on comparison self-supervision ideas, and similarity of regional level flow in a space dimension and similarity of global flow distribution in a time dimension are fully considered, so that prediction accuracy is improved.
To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:
a city fine granularity flow prediction method based on space-time comparison self-supervision comprises the following steps:
obtaining a plurality of flow distribution diagrams of a city to be predicted for a period of time, and constructing a fine granularity flow distribution diagram and a coarse granularity flow distribution diagram based on each flow distribution diagram;
performing comparison self-supervision learning on the coarse-granularity flow distribution map according to the similarity between areas in the map to obtain a space encoder;
Comparing and self-supervising learning is carried out on the coarse-granularity flow distribution map at a specific moment according to the similarity between the coarse-granularity flow distribution map at a specific moment and the flow distribution map at other moments to obtain a time encoder;
Training a fine-granularity flow prediction model for prediction of the regional fine-granularity flow based on the fine-granularity flow distribution map and the coarse-granularity flow distribution map; the fine-grain traffic prediction model includes a splice layer, a decoder, and an upsampling layer for combining a spatial encoder and a temporal encoder.
Further, the method for constructing the fine particle size flow distribution map and the coarse particle size flow distribution map comprises the following steps:
and obtaining a fine granularity flow distribution map according to the flow distribution data, and obtaining a coarse granularity flow distribution map according to a set coarse granularity scaling factor.
Further, performing contrast self-supervision learning on the coarse-granularity flow distribution map according to the similarity between the areas in the map, and obtaining the spatial encoder includes:
Extracting high-level semantic features from the coarse-grain flow distribution map; and randomly selecting an anchor point region, carrying out similarity measurement on the region and other regions in the same graph based on high-level semantic features, acquiring a positive sample set and a negative sample set, and carrying out comparison self-supervision learning.
Further, deriving the temporal encoder by contrast self-supervised learning includes:
Extracting high-level semantic features from the coarse-grain flow distribution map; taking the coarse granularity flow distribution map at a specific moment as an anchor point; and carrying out similarity measurement on the moment and other moments based on high-level semantic features, obtaining a positive sample set and a negative sample set, and carrying out comparison self-supervision learning.
Further, a multi-layer perceptron is adopted to extract high-level semantic features.
Further, the decoder is a convolutional layer with ReLu activation functions.
Further, in the training process of the fine granularity flow prediction model, structural constraint conditions are further set, and the sum of the regional flows in the downsampled coarse granularity flow distribution map is required to be equal to the corresponding regional flow in the coarse granularity flow distribution map.
One or more embodiments provide a system for urban fine-grained flow prediction based on spatiotemporal contrast self-supervision, comprising:
The training data acquisition module is used for acquiring a plurality of flow distribution diagrams of a city to be predicted for a period of time and constructing a fine-granularity flow distribution diagram and a coarse-granularity flow distribution diagram based on each flow distribution diagram;
the spatial comparison self-supervision learning module is used for comparing and self-supervision learning the coarse granularity flow distribution map according to the similarity between areas in the map to obtain a spatial encoder;
The time comparison self-supervision learning module is used for comparing and self-supervision learning on the coarse granularity flow distribution map at a specific moment according to the similarity between the coarse granularity flow distribution map and the flow distribution map at other moments to obtain a time encoder;
The flow prediction model training module is used for training a fine-grain flow prediction model based on the fine-grain flow distribution diagram and the coarse-grain flow distribution diagram and predicting the fine-grain flow of the region; the fine-grain traffic prediction model includes a splice layer, a decoder, and an upsampling layer for combining a spatial encoder and a temporal encoder.
One or more embodiments provide an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, which when executed implements the spatiotemporal contrast self-supervised based urban fine granularity traffic prediction method.
One or more embodiments provide a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the spatio-temporal contrast self-supervised based urban fine granularity traffic prediction method.
The one or more of the above technical solutions have the following beneficial effects:
the method for urban fine-granularity flow prediction based on space-time comparison self-supervision is provided, a space encoder and a time encoder are constructed based on comparison self-supervision thought aiming at flow data characteristics, similarity of regional level flow in space dimension and similarity of global flow distribution in time dimension are fully considered, and an urban fine-granularity flow prediction model has rich space-time data characteristics, so that prediction accuracy is improved.
In the process of carrying out final fine granularity flow prediction model training by combining a space encoder and a time encoder, by setting structural constraint, the sum of flows in fine granularity areas is ensured to be strictly equal to the corresponding coarse granularity flow, and the prediction effect is further improved. Compared with the traditional urban flow prediction model, the method has the characteristics of simple network, less parameter quantity, short training period, good prediction effect and the like.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
FIG. 1 is a flow chart of a method for urban fine granularity flow prediction based on space-time contrast self-supervision in an embodiment of the invention;
FIG. 2 is a schematic diagram of coarse and fine particle size flow distribution and structural constraints in an embodiment of the present invention;
FIG. 3 is a flow chart of spatial encoder training in accordance with an embodiment of the present invention;
FIG. 4 is a graph showing the trend of total flow in urban areas over time according to an embodiment of the present invention;
FIG. 5 is a timing encoder training flowchart in accordance with an embodiment of the present invention;
fig. 6 is a schematic diagram of an overall architecture of a model for urban fine-grained traffic prediction based on space-time contrast self-supervision in an embodiment of the invention.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Embodiments of the invention and features of the embodiments may be combined with each other without conflict.
Introduction of contrast self-supervision technology: basic methods in machine learning are supervised learning, semi-supervised learning, and unsupervised learning. The biggest difference is whether the model needs manually marked label information or not when training, and the supervision and learning utilize a large amount of label data to train the model, so that the model finally learns the correlation between the input label and the output label; semi-supervised learning trains a network with a small amount of tagged data and a large amount of untagged data; and the unsupervised learning does not depend on any label value, and the relation among samples is found through mining the internal characteristics of the data. The contrast self-supervision belongs to one of the non-supervision learning, and the characteristic relation between positive samples is enhanced by establishing positive and negative sample pairs, so that the distance between the positive samples and the negative samples is prolonged.
Example 1
City fine-grained traffic prediction aims at deducing the fine-grained traffic in the city from the observed coarse-grained traffic (traffic, bicycle traffic, car traffic). The embodiment discloses a city fine-granularity flow prediction method based on space-time comparison self-supervision, which utilizes the self-supervision method to predict city fine-granularity flow, designs two auxiliary task networks according to the space-time characteristics of flow data, enables a simple network structure to obtain strong representation capability from the data, and then combines two pre-trained encoder fine-tuning networks to predict city fine-granularity flow. In particular, the flow data has spatiotemporal characteristics. Spatially, there are many urban areas (e.g., office, entertainment, residential areas, etc.) that have the same function in one city, which are geographically diverse but have similar traffic distribution conditions. The regional comparison self-supervision is designed aiming at the space angle, and the regional comparison self-supervision aims to enable the urban regional features with the same function to be similar to each other and the regional features with different functions to be far away. Traffic flow data is periodic and contiguous from a temporal perspective. In major urban thoroughfares, traffic is similar during commuting to and from work. The time comparison self-supervision is designed for the time angle, and the purpose of the time comparison self-supervision is to enable the characteristics of similar flow moments to be similar to each other and enable the characteristics of dissimilar flow moments to be far away. Through training of two self-supervision tasks, two encoders respectively learn the space-time characteristics of data, and finally fine-tune two encoding networks to conduct urban fine-grained flow prediction.
As shown in fig. 1 and 6, the method specifically includes the following steps:
Step 1: a plurality of flow distribution maps of a city to be predicted for a period of time are acquired, and a fine-granularity flow distribution map and a coarse-granularity flow distribution map are constructed based on each flow distribution map.
Wherein the flow profile is obtained from a public data website, including pedestrian data, bicycle data, and motor vehicle data, and is pre-processed.
According to different longitude and latitude dividing scales, we can obtain a coarse-granularity flow distribution map and a fine-granularity flow distribution map, and determine a scaling factor N e Z + between the coarse-granularity flow distribution maps, for example, scaling factor n=2 in fig. 2. Finally obtaining a coarse grain flow distribution diagramAnd fine grain flow distribution map/>. Constructing a city fine granularity flow prediction task:
In this embodiment, a fine-granularity flow distribution map is first obtained, and then the fine-granularity flow distribution map is processed according to a set scaling factor to obtain a coarse-granularity flow distribution map.
Step 2: and comparing and self-supervising learning is carried out on the coarse granularity flow distribution diagram according to the similarity between areas in the diagram, so as to obtain the space encoder.
In the existing city fine-granularity traffic prediction, only traffic conditions of adjacent areas are considered, and areas with similar semantic information and dissimilar distances are ignored. The embodiment designs a space contrast self-supervision network (Reg), which is called a space encoder, and comprises a high-level semantic feature extraction module and a contrast self-supervision learning module. The high-level semantic feature extraction module adopts a multi-layer perceptron to extract high-level semantic features of the coarse-granularity flow distribution map; and the comparison self-supervision learning module randomly selects a region as an anchor point, performs similarity measurement on the region and other regions in the same graph based on high-level semantic features, acquires a positive sample set and a negative sample set, performs comparison self-supervision learning, and evaluates the space encoder by adopting a comparison loss function. If the value of the contrast loss function does not meet the set condition, selecting other areas as anchor points, continuing learning, and if the value of the contrast loss function meets the set condition, stopping iteration.
Specifically, for coarse particle size flow distributionIt is first changed to a high-level semantic representation Z reg∈RHW×C by a multi-level perceptron. Selecting an anchor point x q from the high-level semantic representation, and obtaining a positive sample/>, by calculating Euclidean distance between the anchor point region and other regionsAnd negative sample/>. Calculating a contrast loss function:
where sim (u, v) is a similarity function (e.g., inner product or cosine similarity) between the computed features. In this way, features between positive samples are pulled closer to each other, and away from features of negative samples.
As shown in fig. 3, there are many similar flow regions in the current coarse-grain flow profile during the spatially-contrasted self-supervised pre-training process. Randomly selecting an area to be marked as an anchor point x q, and naming an area with similar flow and an area with dissimilar flow as positive samplesAnd negative sample/>. Because positive and negative samples exhibit different distribution conditions in traffic size, this property should be the same at their high level semantic information representation.
Step 3: and performing contrast self-supervision learning on the coarse-granularity flow distribution map at the specific moment according to the similarity between the coarse-granularity flow distribution map at the specific moment and the flow distribution map at other moments to obtain the time encoder, wherein the time encoder comprises a high-level semantic feature extraction module and a contrast self-supervision learning module.
Existing urban fine-grained flow predictions infer fine-grained flow profiles from coarse-grained flow profiles only, and neglect the impact of different moments on flow inference. The embodiment designs a time comparison self-monitoring network (TCN), which is called a time encoder, and comprises a high-level semantic feature extraction module and a comparison self-monitoring learning module. The high-level semantic feature extraction module adopts a multi-layer perceptron to extract high-level semantic features of the coarse-granularity flow distribution map; the comparison self-supervision learning module performs high-level semantic feature extraction on the coarse-granularity flow distribution map; taking the coarse granularity flow distribution map at a specific moment as an anchor point; and carrying out similarity measurement on the moment and other moments based on high-level semantic features, obtaining a positive sample set and a negative sample set, and carrying out comparison self-supervision learning. In addition, in the supervision process, a self-supervision auxiliary task is also executed, and is used for reducing the coding characteristic difference between the anchor point and the positive sample and keeping the negative sample characteristic away from the anchor point characteristic.
As shown in fig. 4 and 5, a curve is drawn according to the total flow of the flow chart at each moment to find that the flow data has periodicity. In the time comparison self-supervision process, at any one time stamp, we can obtain an anchor point A, and calculate the Euclidean distance between the anchor point and the flow distribution map on other time stamps to obtain a positive sample set (such as the moment of the point B in the figure)And negative sample set (time points of C and D in the figure)/>. And selecting a flow distribution map which is most similar to the anchor point and a flow distribution map which is least similar to the anchor point according to the positive and negative sample sets.
The self-supervision auxiliary task adopts a Triplet Loss, so that the high-level semantic representation of the anchor point and the positive sample is more similar, and the representation between the anchor point and the negative sample is far away. Given a set of coarse-grained viewsThe Triplet Loss is calculated as follows:
Where f (·) is a learnable nonlinear transformation and α is a positive boundary parameter.
Step 4: and training a fine-grain flow prediction model for predicting the regional fine-grain flow based on the fine-grain flow distribution diagram and the coarse-grain flow distribution diagram.
By pre-training step 2 and step 3 we get spatial encoder Enc reg and temporal encoder Enc tcn, respectively, which contain rich spatio-temporal data features. And then training a fine granularity flow prediction model by combining the space encoder and the time encoder which are obtained through pre-training.
The fine-grained traffic prediction model includes a splice layer, a decoder, and an upsampling layer, connected in sequence, for combining a spatial encoder and a temporal encoder.
The splicing layer is used for performing tensor splicing operation on the two encoder characteristics:
Hreg＝Encreg(Xc) (5)
Htcn＝Enctcn(Xc) (6)
Ha＝Concat(Hreg,Htcn) (7)
wherein Concat is the tensor stitching operation resulting in H a∈RH×W×2C.
The decoder is a convolution layer with ReLu activation functions and is used for carrying out nonlinear transformation on the convolution layer to obtain H D∈RH×W×C:
HD＝ReLu(Conv(Ha)) (8)
The upsampling layer is configured to, for H D obtained by the decoder, perform size expansion and feature dimension reduction on the H D by using an upsampling operation:
Uf＝Upsampling(HD) (9)
Wherein the U f∈RNH×NW×C is used for controlling the temperature of the liquid,
Structural constraints are also introduced during the fine-grained flow prediction phase in order to further improve the accuracy of flow prediction. As shown in fig. 2, we divide the flow rate of a certain area into (a) a coarse-grain flow rate distribution chart and (b) a fine-grain flow rate distribution chart according to the division scale. Urban fine-grained flow prediction is based on (a) coarse-grained flow profiles to infer (b) fine-grained flow profiles, but urban fine-grained flow prediction has strict structural constraints requiring that the sum of flows of fine-grained regions be exactly equal to the corresponding coarse-grained flow size. A normalization constraint is thus designed:
Wherein the method comprises the steps of Is/>The region of the ith row and the jth column, and/>Representing the probability value. Final predicted fine particle flow distribution map/>Wherein +.. Calculation is performed using the mean square error as a loss function:
Where F represents the fine-tuning model and θ represents a parameter that can be learned in the model.
And training the fine granularity flow distribution map and the coarse granularity flow distribution map as the output and input of a model respectively to obtain a fine granularity flow prediction model.
Example two
It is an object of this embodiment to provide a small sample based urban fine grained flow prediction system, the system comprising:
The training data acquisition module is used for acquiring a plurality of flow distribution diagrams of a city to be predicted for a period of time and constructing a fine-granularity flow distribution diagram and a coarse-granularity flow distribution diagram based on each flow distribution diagram;
the spatial comparison self-supervision learning module is used for comparing and self-supervision learning the coarse granularity flow distribution map according to the similarity between areas in the map to obtain a spatial encoder;
The time comparison self-supervision learning module is used for comparing and self-supervision learning on the coarse granularity flow distribution map at a specific moment according to the similarity between the coarse granularity flow distribution map and the flow distribution map at other moments to obtain a time encoder;
The flow prediction model training module is used for training a fine-grain flow prediction model based on the fine-grain flow distribution diagram and the coarse-grain flow distribution diagram and predicting the fine-grain flow of the region; the fine-grain traffic prediction model includes a splice layer, a decoder, and an upsampling layer for combining a spatial encoder and a temporal encoder.
Example III
An object of the present embodiment is to provide an electronic apparatus.
An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method as in embodiment one when executing the program.
Example IV
An object of the present embodiment is to provide a computer-readable storage medium.
A computer readable storage medium having stored thereon a computer program which when executed by a processor implements the method as described in embodiment one.
The steps involved in the second to fourth embodiments correspond to the first embodiment of the method, and the detailed description of the second embodiment refers to the relevant description of the first embodiment. The term "computer-readable storage medium" should be taken to include a single medium or multiple media including one or more sets of instructions; it should also be understood to include any medium capable of storing, encoding or carrying a set of instructions for execution by a processor and that cause the processor to perform any one of the methods of the present invention.
The existing urban fine granularity flow prediction method has the problems of complex model, excessive parameter quantity, long training period, no consideration of flow data time-space characteristics and the like. Aiming at the flow data characteristics, the invention designs two self-supervision auxiliary tasks by adopting a comparison self-supervision idea to help learn the flow data characteristics, and provides a city fine granularity flow prediction method based on space-time comparison self-supervision. The spatial contrast emphasis of the method is based on the similarity of the flow rate of the internal space exploration regional level; time contrast focuses on exploring global traffic distribution similarity. By establishing positive and negative sample pairs, features between the anchor point and the positive sample are similar, so that the negative sample features are far away. Through training two simple network structures, the urban fine granularity flow prediction model has rich space-time data characteristics, and the prediction effect is greatly improved. Compared with the traditional urban flow prediction model, the method has the characteristics of simple network, less parameter quantity, short training period, good prediction effect and the like.
It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented by general-purpose computer means, alternatively they may be implemented by program code executable by computing means, whereby they may be stored in storage means for execution by computing means, or they may be made into individual integrated circuit modules separately, or a plurality of modules or steps in them may be made into a single integrated circuit module. The present invention is not limited to any specific combination of hardware and software.
While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.
Claims (8)
1. A city fine granularity flow prediction method based on space-time comparison self-supervision is characterized by comprising the following steps:
obtaining a plurality of flow distribution diagrams of a city to be predicted for a period of time, and constructing a fine granularity flow distribution diagram and a coarse granularity flow distribution diagram based on each flow distribution diagram;
performing comparison self-supervision learning on the coarse-granularity flow distribution map according to the similarity between areas in the map to obtain a space encoder;
Comparing and self-supervising learning is carried out on the coarse-granularity flow distribution map at a specific moment according to the similarity between the coarse-granularity flow distribution map at a specific moment and the flow distribution map at other moments to obtain a time encoder;
training a fine-granularity flow prediction model for prediction of the regional fine-granularity flow based on the fine-granularity flow distribution map and the coarse-granularity flow distribution map; the fine-grained traffic prediction model includes a splice layer, a decoder, and an upsampling layer for combining a spatial encoder and a temporal encoder;
performing contrast self-supervision learning on the coarse-grain flow distribution map according to the similarity between areas in the map, and obtaining a space encoder comprises:
Extracting high-level semantic features from the coarse-grain flow distribution map; randomly selecting an anchor point region, carrying out similarity measurement on the region and other regions in the same graph based on high-level semantic features, acquiring a positive sample set and a negative sample set, and carrying out comparison self-supervision learning;
comparing and self-supervising learning is carried out on the coarse-granularity flow distribution map at the specific moment according to the similarity between the coarse-granularity flow distribution map at the specific moment and the flow distribution map at other moments, and the time encoder comprises:
Extracting high-level semantic features from the coarse-grain flow distribution map; taking the coarse granularity flow distribution map at a specific moment as an anchor point; and carrying out similarity measurement on the moment and other moments based on high-level semantic features, obtaining a positive sample set and a negative sample set, and carrying out comparison self-supervision learning.
2. The urban fine-granularity flow prediction method based on space-time comparison self-supervision as claimed in claim 1, wherein the method for constructing the fine-granularity flow distribution map and the coarse-granularity flow distribution map is as follows:
and obtaining a fine granularity flow distribution map according to the flow distribution data, and obtaining a coarse granularity flow distribution map according to a set coarse granularity scaling factor.
3. The urban fine-granularity traffic prediction method based on space-time contrast self-supervision as recited in claim 1, wherein a multi-layer perceptron is adopted for high-level semantic feature extraction.
4. The method of urban fine-grained traffic prediction based on spatio-temporal contrast self-supervision according to claim 1, wherein the decoder is a convolutional layer with ReLu activation functions.
5. The urban fine-granularity flow prediction method based on space-time comparison self-supervision according to claim 1, wherein structural constraint conditions are further provided in the training process of the fine-granularity flow prediction model, and the sum of regional flows in the downsampled coarse-granularity flow distribution map is required to be equal to the corresponding regional flow in the coarse-granularity flow distribution map.
6. A space-time contrast self-supervision based urban fine-grained flow prediction system, comprising:
The training data acquisition module is used for acquiring a plurality of flow distribution diagrams of a city to be predicted for a period of time and constructing a fine-granularity flow distribution diagram and a coarse-granularity flow distribution diagram based on each flow distribution diagram;
the spatial comparison self-supervision learning module is used for comparing and self-supervision learning the coarse granularity flow distribution map according to the similarity between areas in the map to obtain a spatial encoder;
The time comparison self-supervision learning module is used for comparing and self-supervision learning on the coarse granularity flow distribution map at a specific moment according to the similarity between the coarse granularity flow distribution map and the flow distribution map at other moments to obtain a time encoder;
the flow prediction model training module is used for training a fine-grain flow prediction model based on the fine-grain flow distribution diagram and the coarse-grain flow distribution diagram and predicting the fine-grain flow of the region; the fine-grained traffic prediction model includes a splice layer, a decoder, and an upsampling layer for combining a spatial encoder and a temporal encoder;
performing contrast self-supervision learning on the coarse-grain flow distribution map according to the similarity between areas in the map, and obtaining a space encoder comprises:
Extracting high-level semantic features from the coarse-grain flow distribution map; randomly selecting an anchor point region, carrying out similarity measurement on the region and other regions in the same graph based on high-level semantic features, acquiring a positive sample set and a negative sample set, and carrying out comparison self-supervision learning;
comparing and self-supervising learning is carried out on the coarse-granularity flow distribution map at the specific moment according to the similarity between the coarse-granularity flow distribution map at the specific moment and the flow distribution map at other moments, and the time encoder comprises:
Extracting high-level semantic features from the coarse-grain flow distribution map; taking the coarse granularity flow distribution map at a specific moment as an anchor point; and carrying out similarity measurement on the moment and other moments based on high-level semantic features, obtaining a positive sample set and a negative sample set, and carrying out comparison self-supervision learning.
7. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method for urban fine-grained traffic prediction based on spatiotemporal contrast self-supervision as defined in any one of claims 1-5 when executing the program.
8. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements a method for urban fine-grained traffic prediction based on spatio-temporal contrast self-supervision according to any of the claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111233589.6A CN113962460B (en) | 2021-10-22 | 2021-10-22 | Urban fine granularity flow prediction method and system based on space-time comparison self-supervision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111233589.6A CN113962460B (en) | 2021-10-22 | 2021-10-22 | Urban fine granularity flow prediction method and system based on space-time comparison self-supervision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113962460A CN113962460A (en) | 2022-01-21 |
| CN113962460B true CN113962460B (en) | 2024-05-28 |
Family
ID=79466297
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111233589.6A Active CN113962460B (en) | 2021-10-22 | 2021-10-22 | Urban fine granularity flow prediction method and system based on space-time comparison self-supervision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113962460B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114662805B (en) * | 2022-05-26 | 2022-09-13 | 山东融瓴科技集团有限公司 | Traffic flow prediction method based on similar time sequence comparison |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110288121A (en) * | 2019-05-16 | 2019-09-27 | 北京交通大学 | Flight based on multiple time granularity attention mechanism visits rate prediction technique |
| CN110414747A (en) * | 2019-08-08 | 2019-11-05 | 东北大学秦皇岛分校 | A kind of space-time shot and long term urban human method for predicting based on deep learning |
| CN112767682A (en) * | 2020-12-18 | 2021-05-07 | 南京航空航天大学 | Multi-scale traffic flow prediction method based on graph convolution neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11842271B2 (en) * | 2019-08-29 | 2023-12-12 | Nec Corporation | Multi-scale multi-granularity spatial-temporal traffic volume prediction |
-
2021
- 2021-10-22 CN CN202111233589.6A patent/CN113962460B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110288121A (en) * | 2019-05-16 | 2019-09-27 | 北京交通大学 | Flight based on multiple time granularity attention mechanism visits rate prediction technique |
| CN110414747A (en) * | 2019-08-08 | 2019-11-05 | 东北大学秦皇岛分校 | A kind of space-time shot and long term urban human method for predicting based on deep learning |
| CN112767682A (en) * | 2020-12-18 | 2021-05-07 | 南京航空航天大学 | Multi-scale traffic flow prediction method based on graph convolution neural network |
Non-Patent Citations (2)
| Title |
|---|
| 王梦园.基于深度学习的精细化出行需求预测.全文. * |
| 黄玲玲.基于手机信令数据的城市网格化人流移动预测分析.全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113962460A (en) | 2022-01-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Du et al. | Deep irregular convolutional residual LSTM for urban traffic passenger flows prediction | |
| CN109658695B (en) | Multi-factor short-term traffic flow prediction method | |
| Guo et al. | GPS-based citywide traffic congestion forecasting using CNN-RNN and C3D hybrid model | |
| CN110543543A (en) | user movement behavior prediction method and device based on multi-granularity neural network | |
| CN113468978B (en) | Fine granularity car body color classification method, device and equipment based on deep learning | |
| CN110570035B (en) | People flow prediction system for simultaneously modeling space-time dependency and daily flow dependency | |
| CN113159403B (en) | Intersection pedestrian track prediction method and device | |
| CN113762338B (en) | Traffic flow prediction method, equipment and medium based on multiple graph attention mechanism | |
| CN113962460B (en) | Urban fine granularity flow prediction method and system based on space-time comparison self-supervision | |
| Rahman et al. | Attention based deep hybrid networks for traffic flow prediction using google maps data | |
| CN112559585A (en) | Traffic space-time sequence single-step prediction method, system and storage medium | |
| Chen et al. | Short‐Term Traffic Flow Prediction with Recurrent Mixture Density Network | |
| Liu et al. | Dynamic traffic flow prediction based on long-short term memory framework with feature organization | |
| CN115762147B (en) | Traffic flow prediction method based on self-adaptive graph meaning neural network | |
| CN116682261A (en) | Holiday road flow data prediction analysis method and analysis device | |
| CN115331460B (en) | Large-scale traffic signal control method and device based on deep reinforcement learning | |
| CN113947250A (en) | Urban fine-grained flow prediction method and system based on limited data resources | |
| CN113094422B (en) | Urban road traffic flow map generation method, system and equipment | |
| CN114023074B (en) | Traffic jam prediction method, device and medium based on multiple signal sources | |
| CN111210513A (en) | Information processing method, device, equipment and storage medium | |
| Jiang et al. | A deep learning-assisted mathematical model for decongestion time prediction at railroad grade crossings | |
| CN112633579A (en) | Domain-confrontation-based traffic flow migration prediction method | |
| Van Toan et al. | Multi-scale synergy approach for real-time semantic segmentation | |
| Yao et al. | TL-Detector: Lightweight Based Real-Time Traffic Light Detection Model for Intelligent Vehicles | |
| Shi et al. | Attention-YOLOX: Improvement in On-Road Object Detection by Introducing Attention Mechanisms to YOLOX |
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 | ||
| GR01 | Patent grant |