CN116524734A - Traffic flow prediction device constructed based on dynamic space-time staggered graph - Google Patents

Traffic flow prediction device constructed based on dynamic space-time staggered graph Download PDF

Info

Publication number
CN116524734A
CN116524734A CN202310412465.7A CN202310412465A CN116524734A CN 116524734 A CN116524734 A CN 116524734A CN 202310412465 A CN202310412465 A CN 202310412465A CN 116524734 A CN116524734 A CN 116524734A
Authority
CN
China
Prior art keywords
time
space
traffic flow
time step
module
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.)
Pending
Application number
CN202310412465.7A
Other languages
Chinese (zh)
Inventor
陈岭
吴彬青
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202310412465.7A priority Critical patent/CN116524734A/en
Publication of CN116524734A publication Critical patent/CN116524734A/en
Pending legal-status Critical Current

Links

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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Analytical Chemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a traffic flow prediction device constructed based on a dynamic space-time staggered graph, which belongs to the field of intelligent traffic systems, and comprises the steps of constructing space and time dependent operation through attention screening operation on the basis of acquiring training samples, standardized samples and combined samples, obtaining graph rolling characteristics of a time step t through a space graph rolling module, a space-time staggered graph rolling module, a characteristic fusion module and a time sequence characteristic extraction module, and obtaining the graph rolling characteristics of the time step t according to time characteristics H t The traffic flow prediction data of H time steps in the future are predicted, the prediction efficiency and the prediction accuracy are improved, the traffic jam is relieved, and reliable traffic planning suggestions are provided for daily commute.

Description

Traffic flow prediction device constructed based on dynamic space-time staggered graph
Technical Field
The invention relates to the field of intelligent traffic systems, in particular to a traffic flow prediction device constructed based on dynamic space-time interleaving diagrams.
Background
Traffic flow prediction is a key technology in intelligent traffic systems and is also an integral part of the development of smart cities. Accurate traffic flow prediction can help to efficiently schedule traffic resources, alleviate traffic jams, provide public safety early warning, and provide reliable advice for daily commute of citizens. Therefore, traffic flow prediction has become a research hotspot in academia and industry, and has wide practical application.
Traffic flow predictions face complex spatial and temporal dependencies, and thus their accuracy presents a great challenge. The spatial dependence is represented by the fact that traffic flow data acquired by traffic flow sensors can be influenced by nearby traffic conditions; the time dependence is manifested in that current traffic flow data is affected by historical traffic flow data. Furthermore, the spatial and temporal dependencies in real-world intelligent transportation systems are often coupled to each other and change over time.
Over the past decades, researchers have proposed many traffic flow prediction methods, among which are shallow machine learning and convolutional neural network-based methods. Over the past few decades, researchers have proposed a number of traffic flow prediction methods. These methods include methods based on shallow machine learning and methods based on Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs). While these approaches make it possible to build time-dependent and grid-based spatial dependencies, they fail to capture the non-european spatial dependencies of real-world irregular traffic networks. To solve this problem, existing works introduce Graph Neural Networks (GNNs), representing traffic flow sensors on the traffic road network as nodes, and representing spatial dependencies between the traffic flow sensors as edges. Recently, researchers have integrated GNNs with RNNs, CNNs, and attents to capture spatial and temporal dependencies. Such neural networks are known as space-time-graph neural networks (Spatial Temporal GraphNeural, STGNNs), and exhibit the most advanced performance in terms of traffic prediction.
The existing STGNNs often use distances between traffic sensors, similarities in distribution of points of interest (Point ofInteresting, POIs) near the traffic sensors, similarities in static learnable embedded representations of the traffic sensors, etc. to construct a static space map when patterning, ignoring the fact that spatial dependence in traffic networks can change over time. In addition, some models attempt to model dynamic spatial dependencies, such as using dynamic covariates to adjust static graphs or building dynamic graph structures directly with attention mechanisms, tend to focus on modeling of spatial dependencies only, ignoring dependencies across spatial and temporal dimensions.
Recent correlation studies have proposed a method of modeling time-space interleaved dependencies by representing the dependencies across spatial and temporal dimensions with a combined graph. And constructing a space-time synchronization diagram through the distance diagram and the time connection diagram, and constructing a space-time combination diagram through the distance diagram, the time similarity diagram and the time connection diagram. However, these methods rely on static distance maps, time-connected maps, time-similarity maps, and cannot model dynamic space-time interleaving dependencies.
Disclosure of Invention
In view of the above, the present invention aims to provide a traffic flow prediction device constructed based on a dynamic space-time staggered graph, which establishes spatial dependence and time dependence between sensors and the sensors themselves by introducing the dynamic space-time staggered graph, thereby predicting accuracy.
To achieve the above object, the present invention provides a traffic flow prediction device constructed based on a dynamic space-time interleaving map, comprising a memory for storing a computer executable program for performing traffic flow prediction constructed based on a dynamic space-time interleaving map, and a processor communicatively connected to the memory and configured to execute the computer executable program stored in the memory, wherein the processor implements the following steps when executing the computer executable program:
dividing traffic flow data into training sets, randomly taking a batch of traffic flow data from the training sets as training samples to perform time sequence standardization operation, and combining the standardized samples with the training samples to obtain combined samples;
the traffic flow prediction model is constructed by a attention screening module, a dynamic space diagram construction module, a dynamic time connection diagram construction module, a dynamic space-time staggered diagram construction module, a diagram convolution module, a time sequence feature extraction module and a traffic flow prediction module, wherein the attention screening module is used for screening the combined samples to obtain related time step weights W sel Corresponding related traffic flow data X sel The method comprises the steps of carrying out a first treatment on the surface of the For dynamic space diagram construction modules Embedding the node into the representation E N And time step tFusion is carried out to obtain E t Will E t Space diagram for constructing time step t +.>The dynamic time connection diagram construction module is used for constructing a space diagram-based +.>The resulting diagonal matrix and the associated time step weight of time step t +.>Multiplication to construct a plurality of time-connected graphs of time step t +.>The dynamic space-time interleaving diagram construction module is used for constructing a space diagram +.>With multiple time connectivity graphsCombining to obtain a spatio-temporal interleaving pattern of time step t>The graph convolution module comprises a space graph convolution module, a space-time staggered graph convolution module and a feature fusion module, wherein the space graph convolution module is used for carrying out time step t on traffic flow data X t And a space diagram of time step t->Performing convolution operation to obtain spatial feature +.>The space-time interleaving chart convolution module is used for relating the time step t to the traffic flow data +.>And time step t +.>Performing convolution operation to obtain the space-time interleaving feature +.>The feature fusion module is used for interleaving the time step t space-time characteristics +.>And spatial features of time steps->Merging to obtain the picture convolution feature->The time sequence feature extraction module is used for carrying out the picture convolution feature of the time step t->Fusion with the graph convolution feature of the past time step to extract the time feature H of the time step t t The method comprises the steps of carrying out a first treatment on the surface of the The traffic flow prediction module is used for predicting the traffic flow according to the time characteristic H t Traffic flow prediction value +.>
All traffic flow data training samples and merging samples are sent into the traffic flow prediction model for training, and model parameters are updated to be continuously optimized;
and predicting the traffic flow by using the parameter-optimized traffic flow prediction model.
Preferably, the attention filter module includes a band timing normalizationIs used for screening combined samples to acquire traffic flow data X sel The specific process comprises the following steps:
the combined samples are subjected to linear transformation to obtain a query vector Q and a key tensor K,
wherein the query vector Q and the key tensor K belong to Representing the combined samples, linear () is a Linear transform, d h A dimension number representing the hidden representation obtained after the linear transformation;
mapping Q and K to Fourier space with FFT, and mapping the result calculated in Fourier space back to original space by inverse Fourier transform to obtain an FFT-based attention matrix M agg ，M agg Each element of (3)The calculation process of (1) is as follows:
wherein,,representing FFT, & lt>Representing the inverse Fourier transform, " >Indicates the conjugation operation, as followsRepresenting Hadamard product, Q i Represents the ith query vector, K j Represents the j-th key vector,>and->Respectively represent Q i And K j The value obtained after FFT transformation, dF represents the number of dimensions of hidden representation in Fourier space,/I>Is Q i And K j Calculated attention matrix, M ij Averaging in node dimension and feature dimension to obtain a representative Q i And K j Attention value of the correlation between +.>
Attention moment array based on FFTScreening tau relevant time steps for each time step, and obtaining corresponding screening indexes and weights based on the tau relevant time steps, wherein the screening indexes and weights are respectively +.>And->Obtaining relevant traffic flow data according to the index>Expressed by the formula:
I sel ，W sel ＝Topτ(M agg )。
preferably, the dynamic space diagram construction module is used for embedding the node into E N And time step tThe fusion is specifically as follows:
wherein the time embedding is expressed asNode embedding is denoted +.> The representation will->And E is connected with N Added to each row of the table to obtain a node embedded representation E at time step t t ；
Will E t Space diagram for constructing time step tThe method comprises the following steps:
wherein softmax () is a normalization function used to embed the node of time step t into the representation E t Normalization processing is carried out to obtain the space dependence among all nodes in the time step t
Preferably, the dynamic time connection graph construction module is used for being based on a space graphDiagonal cornerA diagonal matrix obtained from the values on the line and the associated time step weight for time step t>Multiplication to construct multiple temporal join graphs of time step tThe method comprises the following steps:
taking outValues on diagonal are diagonals matrix +.>Associating it with each relevant time step weight of time step t +.>Multiplying, constructing a plurality of time-connected graphs of time step t, wherein for each time-connected graph construction:
wherein,,representing the ith time connection diagram, and the time steps t are represented as tau time connection diagrams in totalt 1 ，t 2 ，...，t τ For index corresponding to τ time steps related to time step t, +.>To model the time dependence of each node on its own node at time steps t and τ related time steps.
Preferably, the dynamic space-time interleaving diagram construction module is used forTo map spaceConnect with multiple time charts->Combining to obtain a spatio-temporal interleaving pattern of time step t>Modeling the spatial dependence and the time dependence of each node at the time step t;
space diagram of time step tA plurality of time connection diagrams of time step t are arranged on the diagonal line of the space-time staggered diagram +.>According to the rule that the time step index is large and the index is small and the rule is placed on the right side, the rule is placed on the upper triangle part of the space-time staggered graph to model directed time dependence, and the space graph and the time connection graph on the diagonal are added to perform fusion of space dependence and time dependence, specifically:
Wherein,,is a spatio-temporal interleaving pattern of time steps t, and t 1 ＜t 2 ＜...＜t τ 。
Preferably, the graph convolution module comprises a space graph convolution module, a space-time interleaving graph convolution module and a feature fusion module, wherein the space-time interleaving graph convolution module is used for correlating the traffic flow data X with the time step t sel And time step t space-time interleaving diagramPerforming convolution operation to obtain the space-time interleaving feature +.>The method comprises the following steps: node embedding representation E using time step t t To generate the graph convolution parameters corresponding to the time step t,
wherein,,generating a kernel function of the convolution weights of the spatio-temporal interleaving map, < >>Generating a kernel function of the space-time interleaving pattern bias, d i And d o Is the input and output dimensions of the space-time interleaving map convolution, which is based on the message passing theory:
wherein,,for the input values of the spatiotemporal cross-map convolution, for the first layer of map convolution, +.>Is-> Is an identity matrix>A space-time interleaving feature for time step t;
the space diagram convolution module is used for calculating traffic flow data X of time step t t And a space diagram of time step tPerforming convolution operation to obtain spatial feature +.>The method comprises the following steps: node embedding representation E using time step t t To generate the graph convolution parameters corresponding to the time step t,
Wherein,,generating a kernel function of the convolution weights of the spatio-temporal interleaving map, < >>Generating a kernel function of the space-time interleaving pattern bias, d i And d o Is the input and output dimensions of the space-diagram convolution, based on the space-time diagram convolution of the message passing theory:
wherein,,for the input values of the spatiotemporal cross-map convolution, for the first layer of map convolution, +.>Is-> Is an identity matrix>A spatiotemporal feature that is time step t;
the feature fusion module is used for interleaving the time step t space-time featuresAnd spatial characteristics of time step t->Merging to obtain the convolution feature->The time-space interleaving characteristic of time step t is adopted by adopting a cyclic neural network GRU>And spatial characteristics->Merging, namely inputting the merged characteristics into the GRU, wherein the GRU comprises the following steps:
wherein AvgPooling () is an average pooling operation, which is a spatio-temporal interleaving feature of time step tAveraging pooling to reduce the number of scales of features for easy calculation, averaging the feature dimensions after pooling with +.>The dimensions of (2) remain the same, willAnd->After merging, linear () Linear transformation operation is carried out to obtain the picture convolution feature of the space-time interlacing feature and the space feature which are fused with the time step t>
Preferably, the time sequence feature extraction module adopts a cyclic neural network GRU to roll up the graphInputting two layers of GRU networks, and extracting time characteristic according to time dependence in traffic flow data >
Preferably, the traffic flow prediction module extracts time features, space features and time-space features of traffic flow data of T time steps, takes hidden representation of the last time step, and inputs the hidden representation into a layer of convolutional neural network to obtain traffic flow prediction values of H time steps in the future
Preferably, the average absolute error function MAE is used to calculate the error between the predicted true value and the predicted value output by the actual model and update the model parameters during training, so as to optimize continuously.
In order to achieve the aim of the invention, the invention also provides a traffic flow prediction device constructed based on the dynamic space-time staggered graph, which is characterized by comprising a data acquisition unit, a model construction unit, a training unit and an application unit,
the data acquisition unit is used for dividing traffic flow data into training sets, randomly taking a batch of traffic flow data from the training sets as training samples to perform time sequence standardization operation, and combining the standardized samples with the training samples to acquire combined samples;
the model building unit is used for
The traffic flow prediction model is constructed by a attention screening module, a dynamic space diagram construction module, a dynamic time connection diagram construction module, a dynamic space-time staggered diagram construction module, a diagram convolution module, a time sequence feature extraction module and a traffic flow prediction module, wherein the attention screening module is used for screening the combined samples to obtain related time step weights W sel Corresponding related traffic flow data X sel The method comprises the steps of carrying out a first treatment on the surface of the The dynamic space diagram construction module is used for embedding the nodes into the representation E N And time step tFusion is carried out to obtain E t Will E t Space diagram for constructing time step t +.>The dynamic time connection diagram construction module is used for constructing a space diagram-based +.>The resulting diagonal matrix and the associated time step weight of time step t +.>Multiplication to construct a plurality of time-connected graphs of time step t +.>The dynamic space-time interleaving diagram construction module is used for constructing a space diagram +.>With multiple time connectivity graphsCombining to obtain a spatio-temporal interleaving pattern of time step t>The graph convolution module comprises a space graph convolution module, a space-time staggered graph convolution module and a feature fusion module, wherein the space graph convolution module is used for carrying out time step t on traffic flow data X t And a space diagram of time step t->Performing convolution operation to obtain spatial feature +.>The space-time interleaving chart convolution module is used for relating the time step t to the traffic flow data +.>And time step t +.>Performing convolution operation to obtain the space-time interleaving feature +.>The feature fusion module is used for interleaving the time step t space-time characteristics +.>And spatial features of time steps->Merging to obtain the picture convolution feature- >The time sequence feature extraction module is used for carrying out the picture convolution feature of the time step t->Fusion with the graph convolution feature of the past time step to extract the time feature H of the time step t t The method comprises the steps of carrying out a first treatment on the surface of the The traffic flow prediction module is used for predicting the traffic flow according to the time characteristic H t Traffic flow prediction value +.>
The training unit is used for sending all traffic flow data training samples and combined samples into the traffic flow prediction model for training, and continuously optimizing by updating model parameters;
the application unit is used for predicting the traffic flow by using the traffic flow prediction model with optimized parameters.
Compared with the prior art, the invention has the technical effects that at least:
the invention introduces an FFT attention screening device, screens relevant time steps for each time step according to traffic flow data of each time step, and further limits the number of relevant time steps to reduce calculation complexity; the method has the advantages that the concept of space-time combined graphs is introduced, the combined graphs are used for representing the dependence of cross space and time dimensions, the dynamic space-time staggered graph construction module is introduced, the dynamic space graph and the dynamic time connection graph are constructed in a data driving mode, and the dynamic space-time staggered graph is constructed by combining the dynamic space graph and the dynamic time connection graph, so that the dynamic space dependence among all sensors and the dynamic time dependence of all sensors are simultaneously modeled, and the problem of modeling of the dynamic space-time staggered dependence is effectively solved. Compared with the traditional prediction method, the dynamic space-time staggered graph can predict traffic flow more accurately and efficiently, can improve the efficiency and safety of an intelligent traffic system, relieves traffic jams, and provides reliable suggestions for daily commute of citizens. Meanwhile, the technology has a certain pushing effect on the development of the graph neural network.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a traffic flow prediction method constructed based on a dynamic space-time interleaving diagram provided by an embodiment of the invention;
FIG. 2 is a general training flow chart of a traffic flow prediction model constructed based on a dynamic space-time interleaving diagram provided by an embodiment of the invention;
FIG. 3 is a diagram of a traffic flow prediction model constructed based on a dynamic space-time interleaving diagram according to an embodiment of the present invention;
FIG. 4 is a block diagram of an FFT-based attention filter with time sequence normalization in a traffic flow prediction model constructed based on dynamic space-time interleaving patterns, provided by an embodiment of the invention;
fig. 5 is a block diagram of a traffic flow prediction device constructed based on a dynamic space-time interleaving diagram according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the detailed description is presented by way of example only and is not intended to limit the scope of the invention.
In order to solve the problems in the prior art, the embodiment provides a traffic flow prediction method and a traffic flow prediction device constructed based on a dynamic space-time staggered graph.
As shown in fig. 1, the traffic flow prediction method constructed based on the dynamic space-time interleaving diagram provided by the embodiment includes the following steps:
s110, dividing the traffic flow data into training sets, taking a batch of traffic flow data randomly from the training sets as training samples to perform time sequence standardization operation, and combining the standardized samples with the training samples to obtain combined samples.
In the embodiment, abnormal value elimination processing and standardization processing are carried out on given traffic flow data, and the processed data are divided by utilizing a sliding window to obtain a training set.
Removing abnormal values and invalid values (such as values exceeding a normal range) in a given traffic flow, and performing z-score standardization processing on the processed traffic flow, wherein the z-score standardization processing specifically comprises the following steps:
wherein X is i，raw Original traffic flow data for the ith node, μ i，raw Sigma, which is the average value in the original traffic flow data of the ith node i，raw Variance, X, in the original traffic flow data for the ith node i Traffic flow data after normalization for the ith node.
And according to the empirical human set time window size T, dividing the standardized data by utilizing the sliding step length with fixed length to obtain a training set.
In an embodiment, as shown in fig. 2, the training set is batched according to a fixed batch size, and a batch of training samples is randomly selected from the training set, wherein the data of each sample includes traffic flow X of N nodes T time steps on the traffic road network 1：T 。
In an embodiment, each batch of training samples is subjected to time sequence normalization processing, and the normalized samples and the original training samples are combined in a feature dimension.
Processing the batch training samples by time sequence standardization for extracting high-frequency components in the traffic flow dataThe method comprises the following steps:
wherein mu batch Sum sigma batch Respectively the mean and variance of a batch of samples, gamma high And beta high Is used for estimating mu batch Sum sigma batch E is a minimum,i.e., a sample of the time series normalization. />Comprises N nodes and high frequency components of T time steps, and is input with the original input X 1：T Combining according to the feature dimension to obtain traffic flow data after expanding time sequence feature>The method comprises the following steps:
i.e., the combined samples.
S120, constructing a traffic flow prediction model, wherein the traffic flow prediction model comprises an attention filter module, a random initialization module, a dynamic space diagram construction module, a dynamic time connection diagram construction module, a combination module, a diagram rolling module, a space-time staggered diagram convolution module, a feature fusion extraction module and a traffic flow prediction module.
In an embodiment, as shown in fig. 2, the overall training flow is that the combined samples are input into an attention filter based on FFT, and the most relevant τ time steps are obtained by filtering for each time step, and the corresponding filtering indexes and weights are respectively I sel And W is sel And obtain the related traffic flow data X according to the index sel 。
In an embodiment, FFT-basedThe structure diagram of the attention filter is shown in fig. 4, and the combined samples are input into the attention filter based on FFT to obtain the filtered traffic flow data X sel The specific process comprises the following steps:
the combined samples are subjected to linear transformation to obtain a query vector Q and a key tensor K,
wherein the query vector Q and the key tensor K belong to Representing the combined samples, linear () is a Linear transform, d h Is the dimension number of the hidden representation obtained after the linear transformation;
mapping Q and K to Fourier space with FFT, and mapping the result calculated in Fourier space back to original space by inverse Fourier transform to obtain an FFT-based attention matrix M agg ，M agg Each element of (3)The calculation process of (1) is as follows:
wherein,,representing FFT, & lt>Representing the inverse Fourier transform, ">Indicating the conjugation operation, ++TableShow Hadamard product, Q i Represents the ith query vector, K j Represents the j-th key vector,>and->Respectively represent Q i And K j The value obtained after FFT conversion, d F Is the number of dimensions of the hidden representation in fourier space, < >>Is Q i And K j Calculated attention matrix, M ij Averaging in node dimension and feature dimension to obtain a representative Q i And K j Attention value of the correlation between +.>
Attention moment array based on FFTScreening tau relevant time steps for each time step, and obtaining corresponding screening indexes and weights based on the tau relevant time steps, wherein the screening indexes and weights are respectively +.>And->Obtaining relevant traffic flow data according to the index>Expressed by the formula:
I sel ，W sel ＝Topτ(M agg )。
as shown in FIG. 3, the dynamic space diagram construction module is used to embed nodes into E N And time step tThe fusion is specifically as follows:
wherein the time embedding is expressed asNode embedding is denoted +.> The representation will->And E is connected with N Added to each row of the table to obtain a node embedded representation E at time step t t ；
Will E t Space diagram for constructing time step tThe method comprises the following steps:
wherein softmax () is a normalization function used to embed the node of time step t into the representation E t Normalization processing is carried out to obtain the space dependence among all nodes in the time step t
As shown in fig. 3, the dynamic time connection graph construction module is used for being based on the space graph A diagonal matrix obtained from the values on the diagonal and the associated time step weight for time step t>Multiple time connection diagrams multiplied to build time t +.>The method comprises the following steps:
taking outValues on diagonal are diagonals matrix +.>Associating it with each relevant time step weight of time step t +.>Multiplying, constructing a plurality of time-connected graphs of time step t, wherein for each time-connected graph construction:
wherein,,representing the ith time connection diagram, and the time steps t are represented as tau time connection diagrams in totalt 1 ，t 2 ，...，t τ For index corresponding to τ time steps related to time step t, +.>To model the time dependence of each node on its own node at time steps t and τ related time steps.
As shown in fig. 3, for the dynamic space-time interleaving diagram construction moduleTo map spaceConnect with multiple time charts->Combining to obtain a spatio-temporal interleaving pattern of time step t>Modeling the spatial dependence and the time dependence of each node at the time step t;
space diagram of time step tA plurality of time connection diagrams of time step t are arranged on the diagonal line of the space-time staggered diagram +.>According to the rule that the time step index is large and the index is small and the rule is placed on the right side, the rule is placed on the upper triangle part of the space-time staggered graph to model directed time dependence, and the space graph and the time connection graph on the diagonal are added to perform fusion of space dependence and time dependence, specifically:
Wherein,,is a spatio-temporal interleaving pattern of time steps t, and t 1 ＜t 2 ＜...＜t τ 。
As shown in fig. 3, the graph convolution module includes a space graph convolution module, a space-time interleaving graph convolution module, and a feature fusion module, where the space-time interleaving graph convolution module is configured to correlate traffic flow data X with time step t sel And time step t space-time interleaving diagramPerforming convolution operation to obtain the space-time interleaving feature +.>The method comprises the following steps: node embedding representation E using time step t t To generate the graph convolution parameters corresponding to the time step t,
wherein,,generating a kernel function of the convolution weights of the spatio-temporal interleaving map, < >>Generating a kernel function of the space-time interleaving pattern bias, d i And d o Is the input and output dimensions of the space-time interleaving map convolution, which is based on the message passing theory:
wherein,,for the input values of the spatiotemporal cross-map convolution, for the first layer of map convolution, +.>Is-> Is an identity matrix>A space-time interleaving feature for time step t;
the space diagram convolution module is used for calculating traffic flow data X of time step t t And a space diagram of time step tPerforming convolution operation to obtain spatial feature +.>The method comprises the following steps: node embedding representation E using time step t t To generate the graph convolution parameters corresponding to the time step t,
Wherein,,generating a kernel function of the convolution weights of the spatio-temporal interleaving map, < >>Generating a kernel function of the space-time interleaving pattern bias, d i And d o Is the input and output dimensions of the space-diagram convolution, based on the space-time diagram convolution of the message passing theory:
wherein,,for the input values of the spatiotemporal cross-map convolution, for the first layer of map convolution, +.>Is-> Is an identity matrix>A spatiotemporal feature that is time step t; />
The feature fusion module is used for interleaving the time step t space-time featuresAnd spatial characteristics of time step t->Merging to obtain the convolution feature->The time-space interleaving characteristic of time step t is adopted by adopting a cyclic neural network GRU>And spatial characteristics->Merging, namely inputting the merged characteristics into the GRU, wherein the GRU comprises the following steps:
wherein AvgPooling () is an average pooling operation, which is a spatio-temporal interleaving feature of time step tAveraging pooling to reduce the number of scales of features for easy calculation, averaging the feature dimensions after pooling with +.>The dimensions of (2) remain the same, willAnd->After merging, linear () Linear transformation operation is carried out to obtain the picture convolution feature of the space-time interlacing feature and the space feature which are fused with the time step t>
As shown in FIG. 3, the time sequence feature extraction module uses a recurrent neural network GRU to roll up the graph to featureInputting two layers of GRU networks, and extracting time characteristic according to time dependence in traffic flow data >
As shown in fig. 2, for time step t=1, 2,...
As shown in fig. 3, the traffic flow prediction module extracts time features, space features and time-space features of traffic flow data of T time steps, takes hidden representation of the last time step, and inputs the hidden representation into a layer of convolutional neural network to obtain predicted values of traffic flow of H time steps in the future
And S130, sending all traffic flow data training samples and combined samples into a traffic flow prediction model constructed by the dynamic space-time staggered graph for training, and continuously optimizing by updating model parameters.
As shown in FIG. 2, a predicted true value X corresponding to each training sample is calculated T+1：T：H And predicted values output from the actual modelError between->
In an embodiment, the average absolute error is used as the prediction lossI.e. the predicted true value X corresponding to the training sample T +1：T：H And predicted value of actual model output +. >The error between them, in particular,
and adjusting network parameters in the whole model according to errors of all samples in the batch.
According to the predicted lossCalculation formula, calculate loss of all samples in batch +.>The formula is as follows,
wherein the method comprises the steps ofFor the loss of sample B in the batch, B is the number of samples in each batch. According to the loss->Updating the network parameter theta in the whole model, wherein the formula is as follows:
/>
where η is the learning rate.
As shown in FIG. 2, the training sample timing normalization operation is repeated to calculate the loss of all samples in the batchUntil all batches participated in model training.
And repeating the training sample time sequence standardization operation until all batches participate in model training until the designated iteration times are reached.
And S140, carrying out traffic flow prediction by using the traffic flow prediction model with optimized parameters.
In an embodiment, the traffic flow prediction is performed using a parameter-optimized traffic flow prediction model. And outputting a traffic flow prediction result.
For the problems of complex data calculation and low prediction efficiency, an attention filter of the FFT is introduced, and relevant time steps are filtered for each time step according to traffic flow data of each time step, so that the number of relevant time steps is limited to reduce the calculation complexity.
For the problem of dynamic space-time interleaving dependence modeling, a concept of a space-time combined graph is introduced, the dependence of cross space and time dimensions is represented by the combined graph, a dynamic space-time interleaving graph construction module is introduced, a dynamic space graph and a dynamic time connection graph are constructed in a data driving mode, and the dynamic space-time interleaving graph and the dynamic time connection graph are combined to construct the dynamic space-time interleaving graph so as to model the dynamic space dependence among all sensors and the dynamic time dependence of each sensor at the same time, and the problem of dynamic space-time interleaving dependence modeling is effectively solved. Compared with the traditional prediction method, the dynamic space-time staggered graph can more accurately and efficiently predict traffic flow, and can improve the efficiency and safety of the intelligent traffic system.
Based on the same inventive concept, the embodiment also provides a traffic flow prediction device constructed based on the dynamic space-time interleaving diagram, which comprises a memory and a processor, wherein the memory is used for storing a computer program, and when the processor executes the computer program, the steps of the traffic flow prediction method constructed based on the dynamic space-time interleaving diagram provided by the embodiment are realized, and the method comprises the following steps:
s110, dividing traffic flow data into training sets, randomly taking a batch of traffic flow data from the training sets as training samples to perform time sequence standardization operation, and combining the standardized samples with the training samples to obtain combined samples;
S120, constructing a traffic flow prediction model, wherein the traffic flow prediction model comprises an attention filter module, a dynamic space diagram construction module, a dynamic time connection diagram construction module, a dynamic space-time staggered diagram construction module, a diagram convolution module, a time sequence feature extraction module and a traffic flow prediction module;
s130, all traffic flow data training samples and merging samples are sent into a traffic flow prediction model constructed by the dynamic space-time staggered graph for training, and model parameters are updated to be optimized continuously;
and S140, carrying out traffic flow prediction by using the traffic flow prediction model with optimized parameters.
In an embodiment, the memory may be a volatile memory at a near end, such as a RAM, or a nonvolatile memory, such as a ROM, a FLASH, a floppy disk, a mechanical hard disk, or a remote storage cloud. The processor may be a Central Processing Unit (CPU), a Microprocessor (MPU), a Digital Signal Processor (DSP), or a Field Programmable Gate Array (FPGA), i.e., the steps of the traffic flow prediction method constructed based on the dynamic space-time interleaving map may be implemented by these processors.
Based on the same inventive concept, the embodiment also provides a traffic flow prediction device 500 constructed based on the dynamic space-time interleaving map, which comprises a data acquisition unit 510, a model construction unit 520, a training unit 530, an application unit 540,
The data obtaining unit 510 is configured to divide traffic flow data into training sets, randomly take a batch of traffic flow data from the training sets as training samples, perform a time sequence normalization operation, and combine the normalized samples with the training samples to obtain a combined sample.
The model construction unit 520 is configured to construct a traffic flow prediction model, including an attention filter module, a dynamic space diagram construction module, a dynamic time connection diagram construction module, a dynamic space-time interleaving diagram construction module, a diagram convolution module, a time sequence feature extraction module, and a traffic flow prediction module, where the attention filter module is configured to filter the combined samples to obtain a relevant time step weight W sel Corresponding related traffic flow data X sel The method comprises the steps of carrying out a first treatment on the surface of the The dynamic space diagram construction module is used for embedding the nodes into the representation E N And time step tFusion is carried out to obtain E t Will E t Space diagram for constructing time step t +.>The dynamic time connection diagram construction module is used for constructing a space diagram-based +.>The obtained diagonal matrix and the related time step weight of the time step tMultiplication to construct a plurality of time-connected graphs of time step t +.>The dynamic space-time interleaving diagram construction module is used for constructing a space diagramConnect with multiple time charts- >Combining to obtain a spatio-temporal interleaving pattern of time step t>The graph convolution module comprises a space graph convolution module, a space-time staggered graph convolution module and a feature fusion module, wherein the space graph convolution module is used for carrying out time step t on traffic flow data X t And a space diagram of time step t->Performing convolution operation to obtain spatial feature +.>The space-time interleaving chart convolution module is used for relating the time step t to the traffic flow data +.>And time step t +.>Performing convolution operation to obtain the space-time interleaving feature +.>The feature fusion module is used for interleaving the time step t space-time characteristics +.>And spatial characteristics of time step t->MergingObtaining the picture convolution feature->The time sequence feature extraction module is used for carrying out the picture convolution feature of the time step t->Fusion with the graph convolution feature of the past time step to extract the time feature H of the time step t t The method comprises the steps of carrying out a first treatment on the surface of the The traffic flow prediction module is used for predicting the traffic flow according to the time characteristic H t Traffic flow prediction value +.>
The training unit 530 is configured to send all traffic flow data training samples and combined samples into the traffic flow prediction model constructed by the dynamic space-time interleaving map for training, and continuously optimize by updating model parameters;
The application unit 540 is configured to perform traffic flow prediction using the parameter-optimized traffic flow prediction model.
It should be noted that, when the traffic flow prediction device constructed based on the dynamic space-time interleaving diagram provided in the above embodiment performs traffic flow prediction, the above-mentioned division of each functional unit should be used as an example, and the above-mentioned functional allocation may be performed by different functional units according to the need, that is, the internal structure of the terminal or the server is divided into different functional units, so as to complete all or part of the above-mentioned functions. In addition, the traffic flow prediction device constructed based on the dynamic space-time interleaving map provided in the above embodiment belongs to the same concept as the traffic flow prediction method constructed based on the dynamic space-time interleaving map, and the specific implementation process of the traffic flow prediction device is detailed in the traffic flow prediction method constructed based on the dynamic space-time interleaving map, which is not described herein again.
The foregoing detailed description of the preferred embodiments and advantages of the invention will be appreciated that the foregoing description is merely illustrative of the presently preferred embodiments of the invention, and that no changes, additions, substitutions and equivalents may be made without departing from the scope of the invention.

Claims (10)

1. A traffic flow prediction device constructed based on dynamic spatiotemporal interleaving map, comprising a memory for storing a computer executable program for performing traffic flow prediction constructed based on dynamic spatiotemporal interleaving map, and a processor communicatively coupled to the memory and configured to execute the computer executable program stored by the memory, characterized by:
the processor, when executing the computer executable program, performs the steps of:
dividing traffic flow data into training sets, randomly taking a batch of traffic flow data from the training sets as training samples to perform time sequence standardization operation, and combining the standardized samples with the training samples to obtain combined samples;
the traffic flow prediction model is constructed by a attention screening module, a dynamic space diagram construction module, a dynamic time connection diagram construction module, a dynamic space-time staggered diagram construction module, a diagram convolution module, a time sequence feature extraction module and a traffic flow prediction module, wherein the attention screening module is used for screening the combined samples to obtain related time step weights W sel Corresponding related traffic flow data X sel The method comprises the steps of carrying out a first treatment on the surface of the The dynamic space diagram construction module is used for embedding the nodes into the representation E N And time step tFusion is carried out to obtain E t Will E t Space diagram for constructing time step t +.>The dynamic time connection diagram construction module is used for constructing a space diagram-based +.>The resulting diagonal matrix and the phase of time step tTime step off weight->Multiplication to construct a plurality of time-connected graphs of time step t +.>The dynamic space-time interleaving diagram construction module is used for constructing a space diagram +.>Connect with multiple time charts->Combining to obtain a spatio-temporal interleaving pattern of time step t>The graph convolution module comprises a space graph convolution module, a space-time staggered graph convolution module and a feature fusion module, wherein the space graph convolution module is used for carrying out time step t on traffic flow data X t And a space diagram of time step t->Performing convolution operation to obtain spatial feature +.>The space-time interleaving chart convolution module is used for relating the time step t to the traffic flow data +.>And time step t +.>Performing convolution operation to obtain the space-time interleaving feature +.>The feature fusion module is used for interleaving the time step t space-time characteristics +.>And spatial features of time step tMerging to obtain the picture convolution feature->The time sequence feature extraction module is used for rolling up features of the graphs of the time step tFusion with the graph convolution feature of the past time step to extract the time feature H of the time step t t The method comprises the steps of carrying out a first treatment on the surface of the The traffic flow prediction module is used for predicting the traffic flow according to the time characteristic H t Traffic flow prediction value +.>
All traffic flow data training samples and merging samples are sent into the traffic flow prediction model for training, and model parameters are updated to be continuously optimized;
and predicting the traffic flow by using the parameter-optimized traffic flow prediction model.
2. The traffic flow prediction device according to claim 1, wherein the attention filter module comprises an attention filter with time-series normalization based on FFT algorithm, the attention filter being configured to filter the combined samples to obtain the traffic flow data X sel The specific process comprises the following steps:
the combined samples are subjected to linear transformation to obtain a query vector Q and a key tensor K,
wherein the query vector Q and the key tensor K belong to Representing the combined samples, linear () is a Linear transform, d h A dimension number representing the hidden representation obtained after the linear transformation;
mapping Q and K to Fourier space with FFT, and mapping the result calculated in Fourier space back to original space by inverse Fourier transform to obtain an FFT-based attention matrix M agg ，M agg Each element of (3)The calculation process of (1) is as follows:
wherein,,representing FFT, & lt>Representing the inverse Fourier transform, ">Represents the conjugate operation, ++represents the Hadamard product, Q i Represents the ith query vector, K j Represents the j-th key vector,>and->Respectively represent Q i And K j The value obtained after FFT conversion, d F Representing the number of dimensions of the hidden representation in fourier space, < >>Is Q i And K j Calculated attention matrix, M ij Averaging in node dimension and feature dimension to obtain a representative Q i And K j Attention value of the correlation between +.>
Attention moment array based on FFTScreening tau relevant time steps for each time step, and obtaining corresponding screening indexes and weights based on the tau relevant time steps, wherein the screening indexes and weights are respectively +.>And->Obtaining relevant traffic flow data according to the index>Expressed by the formula:
I sel ，W sel ＝Topσ(M agg )。
3. the traffic flow prediction device based on dynamic space-time interleaving map construction according to claim 1, wherein the dynamic space map construction module is configured to embed nodes into E N And time of time step tInterval embedding representationThe fusion is specifically as follows:
wherein the time embedding is expressed asNode embedding is denoted +.> The representation will->And E is connected with N Added to each row of the table to obtain a node embedded representation E at time step t t ；
Will E t Space diagram for constructing time step tThe method comprises the following steps:
wherein softmax () is a normalization function used to embed the node of time step t into the representation E t Normalization processing is carried out to obtain the space dependence among all nodes in the time step t
4. Root of Chinese characterThe traffic flow prediction device based on dynamic space-time interleaving map construction according to claim 1, wherein the dynamic time connection map construction module is configured to construct a space-based mapA diagonal matrix obtained from the values on the diagonal and the associated time step weight for time step t>Multiplication to construct a plurality of time-connected graphs of time step t +.>The method comprises the following steps:
taking outValues on diagonal are diagonals matrix +.>Correlating it to each associated time step weight of time step tMultiplying, constructing a plurality of time-connected graphs of time step t, wherein for each time-connected graph construction:
wherein,,representing the ith time connection diagram, the time steps t share sigma time connection diagrams, expressed ast 1 ，t 2 ，...，t τ For index corresponding to τ time steps related to time step t, +.>To model the time dependence of each node on its own node at time steps t and sigma.
5. The traffic flow prediction device based on dynamic space-time interleaving map construction according to claim 1, wherein the dynamic space-time interleaving map construction module is configured to construct a space map Connect with multiple time charts->Combining to obtain a spatio-temporal interleaving pattern of time step t>Modeling the spatial dependence and the time dependence of each node at the time step t;
space diagram of time step tA plurality of time connection diagrams of time step t are arranged on the diagonal line of the space-time staggered diagramAccording to the rule that the time step index is large and the index is small and the rule is placed on the right side, the rule is placed on the upper triangle part of the space-time staggered graph to model directed time dependence, and the space graph and the time connection graph on the diagonal are added to perform fusion of space dependence and time dependence, specifically:
wherein,,is a spatio-temporal interleaving pattern of time steps t, and t 1 ＜t 2 ＜...＜t τ 。
6. The traffic flow prediction device constructed based on dynamic space-time interleaving map according to claim 1, wherein the map convolution module comprises a space map convolution module, a space-time interleaving map convolution module and a feature fusion module, wherein the space-time interleaving map convolution module is configured to correlate traffic flow data X with time step t sel And time step t space-time interleaving diagramPerforming convolution operation to obtain the space-time interleaving feature +.>The method comprises the following steps: node embedding representation E using time step t t To generate the graph convolution parameters corresponding to the time step t,
Wherein,,generating a kernel function of the convolution weights of the spatio-temporal interleaving map, < >>Generating a kernel function of the space-time interleaving pattern bias, d i And d o Is the input and output dimensions of the space-time interleaving map convolution, which is based on the message passing theory:
wherein,,for the input values of the spatiotemporal cross-map convolution, for the first layer of map convolution, +.>Is that Is an identity matrix>A space-time interleaving feature for time step t;
the space diagram convolution module is used for calculating traffic flow data X of time step t t And a space diagram of time step tPerforming convolution operation to obtain spatial feature +.>The method comprises the following steps: node embedding representation E using time step t t To generate the graph convolution parameters corresponding to the time step t,
wherein,,generating a kernel function of the convolution weights of the spatio-temporal interleaving map, < >>Generating a kernel function of the space-time interleaving pattern bias, d i And d o Is the input and output dimensions of the space-diagram convolution, based on the space-time diagram convolution of the message passing theory:
wherein,,for the input values of the spatiotemporal cross-map convolution, for the first layer of map convolution, +.>Is that Is an identity matrix>A spatiotemporal feature that is time step t;
the feature fusion module is used for interleaving the time step t space-time featuresAnd spatial features of time step t Merging to obtain the convolution feature->The time-space interleaving characteristic of time step t is adopted by adopting a cyclic neural network GRU>And spatial characteristics->Merging, namely inputting the merged characteristics into the GRU, wherein the GRU comprises the following steps:
wherein AvgPooling () is an average pooling operation, which is a spatio-temporal interleaving feature of time step tAveraging pooling to reduce the number of scales of features for easy calculation, averaging the feature dimensions after pooling with +.>The dimensions of (2) remain the same, willAnd->After merging, linear () Linear transformation operation is carried out to obtain the picture convolution feature of the space-time interlacing feature and the space feature which are fused with the time step t>
7. The traffic flow prediction device based on dynamic space-time interleaving map construction according to claim 1, wherein the timing characteristic extraction module uses a recurrent neural network (GRU) to roll up the mapInputting two layers of GRU networks, and extracting time characteristic according to time dependence in traffic flow data>
8. The traffic flow prediction device constructed based on the dynamic space-time interleaving map according to claim 1, wherein the traffic flow prediction module extracts time features and space features from the traffic flow data of T time steps, takes the hidden representation of the last time step, and inputs the hidden representation into a layer of convolutional neural network to obtain the traffic flow prediction values of H time steps in the future
9. The traffic flow prediction device constructed based on the dynamic space-time interleaving map according to claim 1, wherein an average absolute error function MAE used in training calculates an error between a predicted true value and a predicted value outputted from an actual model and updates model parameters, and the error is optimized continuously.
10. The traffic flow prediction device based on the dynamic space-time staggered graph is characterized by comprising a data acquisition unit, a model construction unit, a training unit and an application unit,
the data acquisition unit is used for dividing traffic flow data into training sets, randomly taking a batch of traffic flow data from the training sets as training samples to perform time sequence standardization operation, and combining the standardized samples with the training samples to acquire combined samples;
the model construction unit is used for constructing a traffic flow prediction model and comprises an attention screening module, a dynamic space diagram construction module, a dynamic time connection diagram construction module, a dynamic space-time staggered diagram construction module, a diagram convolution module, a time sequence feature extraction module and a traffic flow prediction module, wherein the attention screening module is used for screening the combined samples to obtain related time step weights W sel Corresponding related traffic flow data X sel The method comprises the steps of carrying out a first treatment on the surface of the The dynamic space diagram construction module is used for embedding the nodes into the representation E N And time step tFusion is carried out to obtain E t Will E t Space diagram for constructing time step t +.>The dynamic time connection diagram construction module is used for constructing a space diagram-based +.>The obtained diagonal matrix and the related time step weight of the time step tMultiplication to construct a plurality of time-connected graphs of time step t +.>The dynamic space-time interleaving diagram construction module is used for constructing a space diagram +.>Connect with multiple time charts->Combining to obtain a spatio-temporal interleaving pattern of time step t>The graph convolution module comprises a space graph convolution module, a space-time staggered graph convolution module and a feature fusion module, wherein the space graph convolution module is used for carrying out time step t on traffic flow data X t And a space diagram of time step t->Performing convolution operation to obtain spatial feature +.>The space-time interleaving chart convolution module is used for relating the time step t to the traffic flow data +.>And time step t +.>Performing convolution operation to obtain the space-time interleaving feature +.>The feature fusion module is used for interleaving the time step t space-time featuresAnd spatial characteristics of time step t->Merging to obtain the picture convolution feature->The time sequence feature extraction module is used for carrying out the picture convolution feature of the time step t- >Fusion with the graph convolution feature of the past time step to extract the time feature H of the time step t t The method comprises the steps of carrying out a first treatment on the surface of the The traffic flow prediction module is used for predicting the traffic flow according to the time characteristic H t Predicting traffic flow predictions for H time steps in the future
The training unit is used for sending all traffic flow data training samples and combined samples into the traffic flow prediction model for training, and continuously optimizing by updating model parameters;
the application unit is used for predicting the traffic flow by using the traffic flow prediction model with optimized parameters.
CN202310412465.7A 2023-04-18 2023-04-18 Traffic flow prediction device constructed based on dynamic space-time staggered graph Pending CN116524734A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310412465.7A CN116524734A (en) 2023-04-18 2023-04-18 Traffic flow prediction device constructed based on dynamic space-time staggered graph

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310412465.7A CN116524734A (en) 2023-04-18 2023-04-18 Traffic flow prediction device constructed based on dynamic space-time staggered graph

Publications (1)

Publication Number Publication Date
CN116524734A true CN116524734A (en) 2023-08-01

Family

ID=87393380

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310412465.7A Pending CN116524734A (en) 2023-04-18 2023-04-18 Traffic flow prediction device constructed based on dynamic space-time staggered graph

Country Status (1)

Country Link
CN (1) CN116524734A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116978236A (en) * 2023-09-25 2023-10-31 南京隼眼电子科技有限公司 Traffic accident early warning method, device and storage medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116978236A (en) * 2023-09-25 2023-10-31 南京隼眼电子科技有限公司 Traffic accident early warning method, device and storage medium
CN116978236B (en) * 2023-09-25 2023-12-15 南京隼眼电子科技有限公司 Traffic accident early warning method, device and storage medium

Similar Documents

Publication Publication Date Title
CN109492822B (en) Air pollutant concentration time-space domain correlation prediction method
CN111639787A (en) Spatio-temporal data prediction method based on graph convolution network
CN114092832B (en) High-resolution remote sensing image classification method based on parallel hybrid convolutional network
CN113705880A (en) Traffic speed prediction method and device based on space-time attention diagram convolutional network
CN113052333A (en) Method and system for data analysis based on federal learning
CN112437451B (en) Wireless network flow prediction method and device based on generation countermeasure network
CN116524734A (en) Traffic flow prediction device constructed based on dynamic space-time staggered graph
CN114240000A (en) Air quality prediction method based on space-time graph convolution network
CN104732545A (en) Texture image segmentation method combined with sparse neighbor propagation and rapid spectral clustering
CN114201924B (en) Solar irradiance prediction method and system based on transfer learning
CN111090764A (en) Image classification method and device based on multitask learning and graph convolution neural network
CN113887704A (en) Traffic information prediction method, device, equipment and storage medium
CN108921286A (en) A kind of tranquillization state function brain network establishing method for exempting from threshold value setting
CN112986492A (en) Method and device for establishing gas concentration prediction model
CN115146842A (en) Multivariate time series trend prediction method and system based on deep learning
CN113988263A (en) Knowledge distillation-based space-time prediction method in industrial Internet of things edge equipment
CN104268561A (en) Hyperspectral image mixing eliminating method based on structure prior low rank representation
CN117530684B (en) Blood glucose abnormality detection and early warning system and method based on health big data
Proietti et al. Edge Intelligence with Deep Learning in Greenhouse Management.
Zhang et al. A multi-site tide level prediction model based on graph convolutional recurrent networks
US20240029556A1 (en) Short-term traffic flow prediction method based on causal gated-low-pass graph convolutional network
CN115909016B (en) GCN-based fMRI image analysis system, method, electronic equipment and medium
CN110993121A (en) Drug association prediction method based on double-cooperation linear manifold
CN116579468A (en) Typhoon generation prediction method, device, equipment and medium based on cloud system memory
CN116578858A (en) Air compressor fault prediction and health degree evaluation method and system based on graphic neural network

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