CN114220271B - Traffic flow prediction method, equipment and storage medium based on dynamic space-time diagram convolution circulation network - Google Patents

Traffic flow prediction method, equipment and storage medium based on dynamic space-time diagram convolution circulation network Download PDF

Info

Publication number
CN114220271B
CN114220271B CN202111575360.0A CN202111575360A CN114220271B CN 114220271 B CN114220271 B CN 114220271B CN 202111575360 A CN202111575360 A CN 202111575360A CN 114220271 B CN114220271 B CN 114220271B
Authority
CN
China
Prior art keywords
time
traffic flow
dynamic space
space
dynamic
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
Application number
CN202111575360.0A
Other languages
Chinese (zh)
Other versions
CN114220271A (en
Inventor
潘卫鹏
郭唐仪
唐坤
陆奕
蒋继扬
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.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
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 Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202111575360.0A priority Critical patent/CN114220271B/en
Publication of CN114220271A publication Critical patent/CN114220271A/en
Application granted granted Critical
Publication of CN114220271B publication Critical patent/CN114220271B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • 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
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/12—Discovery or management of network topologies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • 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

  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a traffic flow prediction method based on a dynamic space-time diagram convolution circulation network, which comprises the following steps: constructing a dynamic space-time relationship diagram of the road network; constructing a dynamic space-time diagram convolution circulation module; establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network; preprocessing traffic flow data and dividing a data set; training a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network; and carrying out traffic flow prediction based on the trained model and evaluating the prediction performance of the model. The invention can effectively capture the dynamic time-space dependence of traffic flow data, improve the accuracy of traffic flow prediction and provide scientific traffic guidance basis for traffic management departments. The method has extremely important significance for relieving urban traffic jams, improving running efficiency and safety and relieving traffic environment load. The invention is simple and practical, and has strong operability and effectiveness.

Description

Traffic flow prediction method, equipment and storage medium based on dynamic space-time diagram convolution circulation network
Technical Field
The invention belongs to the field of intelligent traffic, and particularly relates to a traffic flow prediction method, equipment and a storage medium based on a dynamic space-time diagram convolution circulation network.
Background
Road traffic is an artery for social and economic activities, and plays an extremely important role in sustainable development of urban economy and improvement of living standard of people. In recent years, with the continuous promotion of the urban process, a large number of people are rushed into urban work and life, and the requirements on the maintenance quantity and travel of motor vehicles are also increased, so that the problem of urban traffic jam is more and more serious. In order to relieve the traffic pressure of cities, various countries aim at intelligent transportation systems (ITS, intelligent Transportation System). Traffic flow prediction is a hotspot of ITS research, and the predicted result can provide scientific traffic guidance basis for traffic management departments. The traffic system is a complex system, and the traffic flow change process is a real-time, nonlinear, high-dimensional, non-stationary random process. The traditional traffic flow prediction method relies on data driving, only considers the time sequence characteristics of traffic flow data, ignores the space topology of a road network, and breaks the time correlation and the space correlation of traffic flow. In recent years, graph roll-up networks have been applied to traffic flow prediction to enable efficient mining of complex topology of road networks. However, it simply builds a static graph and assigns a fixed weight to represent the associations between nodes, ignoring that the associations between nodes are dynamically changing over time. Therefore, a traffic flow prediction method based on a dynamic space-time diagram convolution circulation network is urgently needed at present, different dynamic space-time relation diagrams are constructed along with time change, dynamic space-time dependence of traffic flow is fully captured, accuracy of traffic flow prediction results is improved, reliable information support is provided for traffic management departments, and positive significance is provided for formulating corresponding traffic control measures.
Disclosure of Invention
The invention aims to solve the problems in the prior art and provide a traffic flow prediction method based on a dynamic space-time diagram convolution circulation network, wherein a dynamic space-time diagram convolution circulation module is constructed by constructing a dynamic space-time relation diagram of a road network, and a traffic flow prediction model based on the dynamic space-time diagram convolution circulation network is established, so that the dynamic space-time dependence of traffic flow is fully captured, the capacity of traffic flow prediction tasks is improved, and a scientific traffic guidance basis is provided for traffic management departments.
The technical solution for realizing the purpose of the invention is as follows: a traffic flow prediction method based on a dynamic space-time diagram convolutional loop network, the method comprising the steps of:
step 1, constructing a dynamic space-time relation diagram of a road network;
step 2, constructing a dynamic space-time diagram convolution circulation module;
step 3, establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
step 4, preprocessing traffic flow data and dividing data sets;
step 5, training a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
and 6, predicting the traffic flow based on the trained model.
Preferably, the step 1 builds a dynamic space-time relation diagram of the road network, and the specific steps include:
step 1-1, constructing a static diagram based on space distance:
constructing a weighted topological relation diagram of the road network according to the geographic position information of the traffic flow data sensor, wherein the sensor is used as a node of the diagram, and the connection relation of the sensor is used as an edge of the diagram;
constructing an adjacency matrix to describe the spatial relationship of the sensor, wherein the weight of the adjacency matrix is calculated by the spatial distance of the sensor in the road network, and specifically comprises the following steps:
Figure BDA0003424671310000021
wherein A is ij Dist The connection weights of node i and node j in the adjacency matrix representing the static graph,
Figure BDA0003424671310000022
sigma, the spatial distance in the road network between node i and node j 1 Epsilon is the standard deviation of the distance between nodes 1 Is a control threshold;
step 1-2, constructing a dynamic diagram based on flow correlation:
the pearson correlation coefficient is used for calculating the correlation of the flow between adjacent nodes, and the method specifically comprises the following steps:
suppose node V i The flow rate of (2) is
Figure BDA0003424671310000023
Neighbor node V j The flow rate of (2) is
Figure BDA0003424671310000024
T is the current time, T is a time sliding time window, and node V i And V is equal to j The correlation of the inter-flow is expressed as:
Figure BDA0003424671310000025
nodeCorrelation Sim (V) i ,V j ) The method comprises the following steps:
Figure BDA0003424671310000031
constructing an adjacency matrix to describe a dynamic space-time topological graph of the correlation among the nodes, wherein the weight of the adjacency matrix is calculated by the correlation among the nodes, and specifically comprises the following steps:
Figure BDA0003424671310000032
wherein A is ij Sim The connection weights of node i and node j in the adjacency matrix representing the dynamic graph,
Figure BDA0003424671310000033
sigma, the spatial distance in the road network between node i and node j 2 Is the standard deviation, epsilon, of the correlation of traffic between nodes 2 Is a control threshold;
step 1-3, stacking a static diagram based on space distance and a dynamic diagram based on flow correlation, and constructing a dynamic space-time relation diagram of a road network:
Figure BDA0003424671310000034
wherein A is ij The connection weights of node i and node j in the adjacency matrix representing the dynamic space-time relationship graph,
Figure BDA0003424671310000035
sigma, the spatial distance in the road network between node i and node j 1 Sigma is the standard deviation of the distance between nodes 2 The standard deviation of the correlation of the flow between the nodes is shown, alpha and beta are weight coefficients, and epsilon is a control threshold.
Preferably, the dynamic space-time diagram convolution loop module comprises a dynamic space-time diagram convolution layer and a loop layer;
the dynamic space-time diagram convolution layer takes the dynamic space-time relation diagram as the diagram input of the network, and continuously updates the dynamic space-time relation diagram along with the movement of the sliding time window; and inputting traffic flow data into a dynamic space-time diagram convolution layer, capturing dynamic space dependence of traffic flow, and outputting time sequence information with space characteristics, wherein the time sequence information is shown in the following formula:
Figure BDA0003424671310000036
wherein X is an input traffic flow time sequence matrix, X' is an output new time sequence matrix with space characteristics,
Figure BDA0003424671310000037
representing a processing process, wherein W is a weight matrix, sigma represents a sigmoid activation function, and GCN (·) represents a graph convolution operation; />
Figure BDA0003424671310000038
And->
Figure BDA0003424671310000039
I N Is a unitary matrix, A represents an adjacency matrix, +.>
Figure BDA00034246713100000310
Is->
Figure BDA00034246713100000311
Elements of (a) and (b);
the time sequence information with space characteristics output by the dynamic space-time diagram convolution layer is input to the circulation layer, the dynamic time dependence of traffic flow is captured, and the hidden layer information with space-time characteristics is output, wherein the hidden layer information with space-time characteristics is shown in the following formula:
X″＝LSTM(X′)
the specific combination process is as follows:
f t ＝σ(W f [h t-1 ，X′ t ]+b f )
i t ＝σ(W i [h t-1 ，X′ t ]+b i )
Figure BDA0003424671310000041
Figure BDA0003424671310000042
o t ＝σ(W o [h t-1 ，X′ t ]+b o )
h t ＝o t ⊙tanh(C t )
where LSTM (·) represents the loop layer operation, X' represents the input of the loop layer, and X "represents the output of the loop layer. X is X t ' represents the input at time t, h t - 1 H is the hidden state of the last moment t Is an updated hidden state; f (f) t 、i t 、o t Indicating a forget gate, an input gate, and an output gate;
Figure BDA0003424671310000043
C t - 1 、C t is the state of the memory cell; w (W) f 、W i 、W o And b f 、b i 、b o Indicating the weight matrix and bias term in the training process, and the letter indicates the corresponding multiplication of matrix elements.
Preferably, step 3 establishes a traffic flow prediction model based on a dynamic space-time diagram convolution loop network, and specifically includes:
step 3-1, integrating dynamic space-time diagram convolution circulating layers through residual connection by a plurality of dynamic space-time diagram convolution circulating modules, extracting space-time characteristics from shallow layer information for aggregation, and generating deeper hidden layer information:
DSTGCRL(X)＝DSTGCRM(ReLU(…ReLU(DSTGCRM(X))))
wherein DSTGCRL (·) represents dynamic space-time diagram convolution loop layer operation, and ReLU (·) represents an activation function;
step 3-2, sequentially connecting 2 dynamic space-time diagram convolution circulating layers, 1 attention mechanism layer and 1 full-connection layer, and establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulating network;
the dynamic space-time diagram convolution circulating layer is used for outputting hidden layer information with dynamic space-time characteristics, the attention mechanism layer distributes different weights for nodes of the hidden layer, and the full-connection layer outputs the result:
f(X,A)＝FC(LeakyReLU(Attention(DSTGCRL(ReLU(DSTGCRL(X))))))
in the formula, f (·) represents model output, attention (·) represents Attention mechanism, FC (·) represents fully connected layer, and LeakyReLU (·) and ReLU (·) are activation functions.
Preferably, the traffic flow data preprocessing and data set dividing in the step 4 specifically includes:
step 4-1, selecting a real-world traffic flow data set comprising geographic position information of a sensor and traffic flow data acquired by the sensor;
step 4-2, data cleaning is carried out on the traffic flow data set, and a linear interpolation method is adopted to fill missing data;
step 4-3, carrying out standardization processing on the data set by adopting a Z-Score method;
and 4-4, dividing the data set into a training set, a testing set and a verification set by adopting a leave-out method.
Preferably, the specific method for training the traffic flow prediction model based on the dynamic space-time diagram convolution circulation network is as follows:
step 5-1, selecting a mean square error MSE as a loss function, selecting an Adam algorithm as a parameter optimizer, setting a model parameter learning rate and setting a Batch size;
and 5-2, randomly selecting data with the Batch size from the training set, sending the data into a model for training, and continuously updating model parameters until the training stopping condition is met, so as to obtain a trained traffic flow prediction model.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the computer program:
step 1, constructing a dynamic space-time relation diagram of a road network;
step 2, constructing a dynamic space-time diagram convolution circulation module;
step 3, establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
step 4, preprocessing traffic flow data and dividing data sets;
step 5, training a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
and 6, predicting the traffic flow based on the trained model.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
step 1, constructing a dynamic space-time relation diagram of a road network;
step 2, constructing a dynamic space-time diagram convolution circulation module;
step 3, establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
step 4, preprocessing traffic flow data and dividing data sets;
step 5, training a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
and 6, predicting the traffic flow based on the trained model.
Compared with the prior art, the invention has the remarkable advantages that:
1) The invention constructs a dynamic time-space relation graph which changes along with time, and gets rid of the limitation of the existing static graph method; calculating the relevance among the nodes by using the pearson relevance coefficient of the flow among the nodes, and continuously updating the relevance among the nodes along with the movement of a sliding time window; stacking a static diagram based on distance and a dynamic diagram based on flow correlation among nodes to construct a space-time relation diagram of a road network; the space-time relation diagram of the road network is dynamically changed along with the input value, and the dynamic space-time dependence of the traffic flow can be effectively explained.
2) The invention constructs a dynamic space-time diagram convolution circulation module, fuses a dynamic space-time relation diagram, a diagram convolution network and a long-short-period memory network, gets rid of the limitation that the traditional deep learning method only depends on data driving, takes account of the space characteristics and the time characteristics of the dynamic change of the traffic flow, and effectively captures the dynamic space-time dependence of the traffic flow.
3) The invention establishes a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network, data is firstly input into the dynamic space-time diagram convolution circulation network layer, dynamic space-time characteristics of traffic flow are extracted, and then different weights are distributed by introducing an attention mechanism, so that differentiated information aggregation is carried out, and finally, a result is output by a full connection layer, thereby improving the prediction performance of the model.
The invention is described in further detail below with reference to the accompanying drawings.
Drawings
FIG. 1 is a flow chart of a traffic flow prediction method based on a dynamic space-time diagram convolutional loop network in one embodiment.
FIG. 2 is a flow diagram of the construction of a dynamic spatio-temporal relationship diagram of a road network in one embodiment.
FIG. 3 is a block diagram of a dynamic space-time diagram convolution loop module in one embodiment.
FIG. 4 is a block diagram of a traffic flow prediction model based on a dynamic space-time convolutional loop network in one embodiment.
Fig. 5 is a schematic diagram of a PeMS04 dataset traffic flow prediction result in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
In addition, it should be noted that, if there is a description of "first", "second", etc. in the embodiments of the present invention, the description of "first", "second", etc. is only for descriptive purposes, and is not to be construed as indicating or implying relative importance or implying that the number of technical features indicated is indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In addition, the technical solutions of the embodiments may be combined with each other, but it is necessary to base that the technical solutions can be realized by those skilled in the art, and when the technical solutions are contradictory or cannot be realized, the combination of the technical solutions should be considered to be absent and not within the scope of protection claimed in the present invention.
In one embodiment, in conjunction with fig. 1, there is provided a traffic flow prediction method based on a dynamic space-time graph convolutional loop network, the method comprising the steps of:
step 1, constructing a dynamic space-time relation diagram of a road network;
step 2, constructing a dynamic space-time diagram convolution circulation module;
step 3, establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
step 4, preprocessing traffic flow data and dividing data sets;
step 5, training a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
and 6, carrying out traffic flow prediction based on the trained model and evaluating the prediction performance of the model.
Further, the Dynamic space-time relationship Graph (DST-Graph) of constructing the road network in the step 1 includes:
step 1-1, constructing a static diagram based on space distance:
and constructing a weighted topological relation diagram of the road network according to the geographic position information of the traffic flow data sensor. The sensor is used as a node of the graph, and the connection relation of the sensor is used as an edge of the graph. Constructing an adjacency matrix to describe the spatial relationship of the sensors, wherein the weight of the adjacency matrix is calculated by the spatial distance of the sensors in the road network. The weight of the adjacency matrix can be obtained by a threshold Gaussian kernel function:
Figure BDA0003424671310000071
wherein A is ij Dist Representing silenceThe connection weights of node i and node j in the adjacency matrix of the state diagram,
Figure BDA0003424671310000072
sigma, the spatial distance in the road network between node i and node j 1 Epsilon is the standard deviation of the distance between nodes 1 Is a control threshold;
step 1-2, constructing a dynamic diagram based on flow correlation:
if the interval of each time slice is short enough, at a moment, the traffic of the adjacent nodes can be considered to have linear correlation, that is, the traffic states of two adjacent nodes in the road network at the same moment have similarity. The pearson correlation coefficient (Pearson Correlation Coefficient) is used to calculate the correlation of traffic between neighboring nodes.
Suppose node V i The flow rate of (2) is
Figure BDA0003424671310000073
Neighbor node V j The flow rate of (2) is
Figure BDA0003424671310000074
T is the current time and T is the sliding time window. Node V i And V is equal to j The correlation of the inter-flow can be expressed as:
Figure BDA0003424671310000081
due to r (X) i ,X j ) The value is [ -1,1]I.e. there is a positive and a negative correlation of the inter-node traffic. Only consider the correlation of adjacent nodes, when the nodes are connected without edges, the correlation Sim (V) between the nodes is obtained after taking the absolute value i ,V j ) I.e.
Figure BDA0003424671310000082
Constructing an adjacency matrix to describe a dynamic space-time topological graph of the correlation among the nodes, wherein the weight of the adjacency matrix is calculated by the correlation among the nodes. The weight of the adjacency matrix can be obtained by a threshold Gaussian kernel function:
Figure BDA0003424671310000083
wherein A is ij Sim The connection weights of node i and node j in the adjacency matrix representing the dynamic graph,
Figure BDA0003424671310000084
sigma, the spatial distance in the road network between node i and node j 2 Is the standard deviation, epsilon, of the correlation of traffic between nodes 2 Is a control threshold;
step 1-3, stacking a static diagram based on space distance and a dynamic diagram based on flow correlation, and constructing a dynamic space-time relation diagram of a road network:
Figure BDA0003424671310000085
wherein A is ij The connection weights of node i and node j in the adjacency matrix representing the dynamic space-time relationship graph,
Figure BDA0003424671310000087
sigma, the spatial distance in the road network between node i and node j 1 Sigma is the standard deviation of the distance between nodes 2 The standard deviation of the correlation of the flow between the nodes is shown, alpha and beta are weight coefficients, and epsilon is a control threshold.
Traffic input to the road network node within a sliding time window T
Figure BDA0003424671310000086
Expressed as:
Figure BDA0003424671310000091
with the movement of the sliding time window T, each time an input is madeThe node traffic is continuously changed, so that the correlation Sim (V i ,V j ) And also constantly changing. Thus, A of the adjacency matrix ij Dynamically changing along with the input value, thereby constructing a dynamic space-time relation diagram of the road network. The method for constructing the map gives consideration to the correlation of the space distance and the flow, so that the dynamic time-space dependence of traffic flow can be explained.
Further, in one embodiment, in conjunction with fig. 3, the constructing a dynamic space-time diagram convolution loop module in step 2 specifically includes:
step 2-1, building a dynamic space-time diagram convolution layer based on a dynamic space-time relation diagram and a diagram convolution network:
according to the sequence of time steps, traffic flow data of different sliding time windows are sequentially input, and a dynamic time-space relation diagram of the current time step is constructed according to the step 1. Based on a graph convolution network (GraphConvolutionalNetwork, GCN) architecture of the first-order chebyshev approximation proposed by Kipf and Welling, a dynamic space-time relationship graph is used as a graph input of the network, and a dynamic space-time graph convolution layer is built. The basic algorithm of GCN is as follows:
in atlas theory, a graph may be represented by its corresponding laplacian matrix, defined as l=d-a, normalized to
Figure BDA0003424671310000092
Wherein A represents an adjacency matrix, I N Is an identity matrix, the degree matrix D is a diagonal matrix D ii ＝∑ j A ij 。
In the graph signal processing, the graph signal
Figure BDA0003424671310000093
Is the feature vector of the node on the graph. According to the spectrum convolution theory, the convolution operation of the input signal x and the convolution kernel g in the time domain can be converted into an inner product form of the frequency domain, namely:
Figure BDA0003424671310000094
in the method, in the process of the invention,
Figure BDA0003424671310000095
for fourier transform, ++Hadamard product, G is a graph convolution operation.
Writing the fourier transform of the convolution kernel g into a diagonal matrix, i.e., g θ ＝diag(U T g) The graph convolution can be reduced to
x* G g θ ＝Ug θ U T x
Due to g θ The (Λ) is relatively complex to calculate, taking into account the first order chebyshev polynomial approximation, i.e
Figure BDA0003424671310000101
Then the first time period of the first time period,
Figure BDA0003424671310000102
further simplify and make beta 0 ＝-β 1 X is x =θ G g θ ＝θ(I N +D -1/2 AD -1/2 )x
To avoid gradient explosions or gradient vanishing, there are
Figure BDA0003424671310000103
Wherein->
Figure BDA0003424671310000104
And is also provided with
Figure BDA0003424671310000105
Thus, the final formula for the graph convolution network is derived:
Figure BDA0003424671310000106
generating a dynamically-changed road network space-time relation graph along with the movement of the sliding time window T, and continuously updating the above formula
Figure BDA0003424671310000109
Thereby building a dynamic space-time diagram convolution layer. And inputting traffic flow data into a dynamic space-time diagram convolution layer, capturing dynamic space dependence of traffic flow, and outputting time sequence information with space characteristics, namely:
Figure BDA0003424671310000107
wherein X is an input traffic flow time sequence matrix, X' is an output new time sequence matrix with space characteristics,
Figure BDA0003424671310000108
representing a processing process, wherein W is a weight matrix, sigma represents a sigmoid activation function, and GCN (·) represents a graph convolution operation;
step 2-2, building a circulating layer based on a long-term and short-term memory network:
based on a Long short-term memory (LSTM) architecture proposed by Hochrite and Schmidhuber, a loop layer is built. LSTM uses memory cells to replace hidden layers of the recurrent neural network, with the memory cells controlled by input gates, forget gates, and output gates. And (3) inputting the time sequence information with the spatial characteristics output in the step (2-1) into a circulating layer, capturing the dynamic time dependence of traffic flow, and outputting hidden layer information with the spatial and temporal characteristics.
X″＝LSTM(X′)
The specific combination process is as follows:
f t ＝σ(W f [h t-1 ,X′ t ]+b f )
i t ＝σ(W i [h t-1 ,X′ t ]+b i )
Figure BDA0003424671310000111
Figure BDA0003424671310000112
o t ＝σ(W o [h t-1 ,X′ t ]+b o )
h t ＝o t ⊙tanh(C t )
where LSTM (·) represents the loop layer operation, X' represents the input of the loop layer, and X "represents the output of the loop layer. X is X t ' represents the input at time t, h t-1 H is the hidden state of the last moment t Is an updated hidden state; f (f) t 、i t 、o t Indicating a forget gate, an input gate, and an output gate;
Figure BDA0003424671310000113
C t-1 、C t is the state of the memory cell; w (W) f 、W i 、W o And b f 、b i 、b o Indicating the weight matrix and bias term in the training process, and the letter indicates the corresponding multiplication of matrix elements.
Therefore, the Dynamic space-time diagram convolution loop module (Dynamic space-temporal Graph Convolutional RecurrentModule, DSTGCRM) has the following operation process:
DSTGCRM(X)＝X″＝LSTM(X′)＝LSTM(GCN(X))
where DSTGCRM (·) represents the dynamic space-time diagram convolution loop module operation.
Further, in one embodiment, referring to fig. 4, the building a traffic flow prediction model based on a dynamic space-time diagram convolutional loop network in step 3 specifically includes:
and 3-1, integrating a Dynamic space-time diagram convolution loop layer (Dynamic space-temporal Graph Convolutional Recurrent Layers) by a plurality of Dynamic space-time diagram convolution loop modules through residual connection (residual block). And increasing the network depth by stacking the dynamic space-time diagram convolution circulation modules, extracting space-time characteristics from shallow information, and aggregating to generate deeper hidden layer information.
DSTGCRL(X)＝DSTGCRM(ReLU(…ReLU(DSTGCRM(X))))
Wherein DSTGCRL (·) represents dynamic space-time diagram convolution loop layer operation and ReLU (·) represents an activation function.
And 3-2, sequentially connecting 2 Dynamic space-time diagram convolution circulating Layers, 1 attention mechanism layer (Attention Mechanism) and 1 full-connection layer (full-connected Layers), and establishing a traffic flow prediction model based on a Dynamic space-time diagram convolution circulating network (Dynamic space-temporal Graph Convolutional Recurrent Network, DSTGCRN). The activation function of the dynamic space-time diagram convolution loop layer is a ReLU function, and the activation function of the attention mechanism layer is a LeakyReLU function. And outputting hidden layer information with dynamic space-time characteristics through a dynamic space-time diagram convolution circulating layer, distributing different weights for nodes of the hidden layer by a attention introducing mechanism, carrying out differentiated information aggregation, and finally outputting a result through a full-connection layer.
f(X,A)＝FC(LeakyReLU(Attention(DSTGCRL(ReLU(DSTGCRL(X))))))
In the formula, f (·) represents model output, attention (·) represents Attention mechanism, FC (·) represents fully connected layer, and LeakyReLU (·) and ReLU (·) are activation functions.
Further, in one embodiment, the traffic flow data preprocessing and data set partitioning in step 4 specifically includes:
step 4-1, selecting a real-world traffic flow data set comprising geographic position information of a sensor and traffic flow data acquired by the sensor;
and 4-2, cleaning the data of the traffic flow data set, and filling the missing data by adopting a linear interpolation method.
Step 4-3, adopting a Z-Score method to perform standardization treatment on the data set, wherein the average value of the treated data is 0, and the standard deviation is 1, namely
Figure BDA0003424671310000121
Wherein x is traffic flow data acquired by a sensor, mu is the mean value of the original data, sigma is the standard deviation of the original data, and x' is standardized data;
and 4-4, dividing the data set into a training set, a test set and a verification set by adopting a leave-out method, wherein the principle is similar to hierarchical sampling, the data set is divided in proportion for a plurality of times in order to ensure randomness, and then the average value of the division results is taken.
Further, in one embodiment, the training of the traffic flow prediction model based on the dynamic space-time diagram convolutional loop network in step 5 specifically includes:
step 5-1, selecting a mean square error MSE as a loss function, selecting an Adam algorithm as a parameter optimizer, setting a model parameter learning rate and setting a Batch size;
and 5-2, randomly selecting data with the Batch size from the training set, sending the data into a model for training, and continuously updating model parameters until the training stopping condition is met, so as to obtain a trained traffic flow prediction model.
Step 6, traffic flow prediction is carried out based on the trained model, a traffic flow prediction result is output, and a baseline model is selected for comparison and reference;
for performance evaluation of the predictive model, it is essential to compare the actual value with the predicted value. And selecting three error functions of average absolute error, average absolute percentage error and root mean square error as indexes for evaluating the prediction effect of the model.
The modules in the traffic flow prediction method based on the dynamic space-time diagram convolution loop network can be fully or partially realized by software, hardware and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the steps of when executing the computer program:
step 1, constructing a dynamic space-time relation diagram of a road network;
step 2, constructing a dynamic space-time diagram convolution circulation module;
step 3, establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
step 4, preprocessing traffic flow data and dividing data sets;
step 5, training a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
and 6, carrying out traffic flow prediction based on the trained model and evaluating the prediction performance of the model.
For specific limitations on each step, reference may be made to the above limitation on the traffic flow prediction method based on the dynamic space-time diagram convolutional loop network, and no further description is given here.
In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of:
step 1, constructing a dynamic space-time relation diagram of a road network;
step 2, constructing a dynamic space-time diagram convolution circulation module;
step 3, establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
step 4, preprocessing traffic flow data and dividing data sets;
step 5, training a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
and 6, carrying out traffic flow prediction based on the trained model and evaluating the prediction performance of the model.
For specific limitations on each step, reference may be made to the above limitation on the traffic flow prediction method based on the dynamic space-time diagram convolutional loop network, and no further description is given here.
As a specific example, in one embodiment, the traffic flow prediction method based on the dynamic space-time diagram convolution loop network of the present invention is further verified and described.
In this example, the PeMS04 dataset collected by the State of California traffic agency Performance measurement System (Caltrans Performance Measurement System, peMS) in the United states is taken as an example. PeMS04 was collected in the san Francisco Bay area and contained 307 detectors at time intervals ranging from 1 month 1 day to 2 months 28 days in 2018 at 5 minutes intervals for 16992 samples. The PeMS04 includes the distance between sensors and traffic flow data, where the traffic flow data is flow, speed and density data collected by the sensors, and in this embodiment, the flow is selected for predictive analysis.
And constructing a dynamic time-space relation diagram of the road network according to the distance between the sensors and traffic flow data and combining the pearson correlation coefficient. And (3) integrating the dynamic space-time relation diagram, the diagram convolution network and the long-short-term memory network to construct a dynamic space-time diagram convolution circulation module, thereby establishing a traffic flow prediction model based on the dynamic space-time diagram convolution circulation network.
The samples were divided into training, test and validation sets in a ratio of about 7:2:1. The sliding time window is defined to be 7 in size, the first 6 time sequence data are input values, the last 1 time sequence data are labels, and a training sample is formed, namely, the traffic state of 5 minutes in the future is predicted by using the traffic data of the first 30 minutes. Before training, the data set is subjected to Z-Score normalization treatment, and the final result is obtained through inverse normalization during prediction. VAR, SVR, LSTM, STGCN is selected as a baseline model and is compared with a DSTGCRN model for analysis. Constructing a network model by using a Pytorch deep neural network framework, setting a mean square error MSE as a loss function, selecting an Adam algorithm as a parameter optimizer, and setting the model parameter learning rate to be 10 -3 The Batch size is 64. Experiments were performed on an Intel (R) Xeon (R) CPU E5-2660 v2 processor and NVIDIA GeForce GTX Titan X graphics card.
Referring to fig. 5, the performance evaluation index of the method of the present invention is shown in table 1:
TABLE 1 Performance indicators of different models on datasets
Figure BDA0003424671310000141
The prediction result obtained by the prediction flow of the method can be seen, and the method can realize traffic flow prediction of the PeMS04 data set. As shown in table 1, the DSTGCRN model showed significant improvement in performance at 3 performance metrics of MAE, MAPE, and RMSE, 19.15, 12.89%, and 28.88, respectively, as compared to the baseline model. Referring to fig. 5, the predicted result of the DSTGCRN model in one day is closest to the data real value group trunk, which indicates that the DSTGCRN model obtains the best predicted performance through training.
In conclusion, the traffic flow prediction method based on the dynamic space-time diagram convolution circulation network provided by the invention can effectively capture the dynamic space-time dependence of traffic flow, and has the advantages of higher prediction precision and strong interpretation. The predicted result can provide reliable information support for traffic management departments, and has positive significance for formulating corresponding traffic control measures.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including 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 will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
While the foregoing is directed to embodiments of the present invention, other and further details of the invention may be had by the present invention, it should be understood that the foregoing description is merely illustrative of the present invention and that no limitations are intended to the scope of the invention, except insofar as modifications, equivalents, improvements or modifications are within the spirit and principles of the invention.

Claims (6)

1. A traffic flow prediction method based on a dynamic space-time diagram convolutional loop network, the method comprising the steps of:
step 1, constructing a dynamic space-time relation diagram of a road network;
step 2, constructing a dynamic space-time diagram convolution circulation module, wherein the dynamic space-time diagram convolution circulation module comprises a dynamic space-time diagram convolution layer and a circulation layer;
the dynamic space-time diagram convolution layer takes the dynamic space-time relation diagram as the diagram input of the network, and continuously updates the dynamic space-time relation diagram along with the movement of the sliding time window; and inputting traffic flow data into a dynamic space-time diagram convolution layer, capturing dynamic space dependence of traffic flow, and outputting time sequence information with space characteristics, wherein the time sequence information is shown in the following formula:
Figure FDA0004146237800000011
wherein X is an input traffic flow time sequence matrix, X' is an output new time sequence matrix with space characteristics,
Figure FDA0004146237800000012
representing a processing process, wherein W is a weight matrix, sigma represents a sigmoid activation function, and GCN (·) represents a graph convolution operation; />
Figure FDA0004146237800000013
And->
Figure FDA0004146237800000014
I N Is a unitary matrix, A represents an adjacency matrix, +.>
Figure FDA0004146237800000015
Is->
Figure FDA0004146237800000016
Elements of (a) and (b);
the time sequence information with space characteristics output by the dynamic space-time diagram convolution layer is input to the circulation layer, the dynamic time dependence of traffic flow is captured, and the hidden layer information with space-time characteristics is output, wherein the hidden layer information with space-time characteristics is shown in the following formula:
X″＝LSTM(X′)
the specific combination process is as follows:
f t ＝σ(W f [h t-1 ,X′ t ]+b f )
i t ＝σ(W i [h t-1 ,X′ t ]+b i )
Figure FDA0004146237800000017
Figure FDA0004146237800000018
o t ＝σ(W o [h t-1 ,X t ′]+b o )
h t ＝o t ⊙tanh(C t )
wherein LSTM (·) represents a loop layer operation, X' represents an input of the loop layer, X "represents an output of the loop layer, X t ' represents the input at time t, h t-1 H is the hidden state of the last moment t Is an updated hidden state; f (f) t Indicating forgetful door, i t Represents an input gate, o t Representing an output gate;
Figure FDA0004146237800000021
C t-1 、C t is the state of the memory cell; w (W) f 、W i 、W o Representing a weight matrix during training, b f 、b i 、b o Representing bias items in the training process, and e represents corresponding multiplication of matrix elements;
step 3, establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network, which specifically comprises the following steps:
step 3-1, integrating dynamic space-time diagram convolution circulating layers through residual connection by a plurality of dynamic space-time diagram convolution circulating modules, extracting space-time characteristics from shallow layer information for aggregation, and generating deeper hidden layer information:
DSTGCRL(X)＝DSTGCRM(ReLU(…ReLU(DSTGCRM(X))))
wherein DSTGCRL (·) represents dynamic space-time diagram convolution loop layer operation, and ReLU (·) represents an activation function;
step 3-2, sequentially connecting 2 dynamic space-time diagram convolution circulating layers, 1 attention mechanism layer and 1 full-connection layer, and establishing a traffic flow prediction model based on a dynamic space-time diagram convolution circulating network;
the dynamic space-time diagram convolution circulating layer is used for outputting hidden layer information with dynamic space-time characteristics, the attention mechanism layer distributes different weights for nodes of the hidden layer, and the full-connection layer outputs the result:
f(X,A)＝FC(LeakyReLU(Attention(DSTGCRL(ReLU(DSTGCRL(X))))))
in the formula, f (·) represents model output, attention (·) represents Attention mechanism, FC (·) represents fully connected layer, and LeakyReLU (·) and ReLU (·) are activation functions;
step 4, preprocessing traffic flow data and dividing data sets;
step 5, training a traffic flow prediction model based on a dynamic space-time diagram convolution circulation network;
and 6, predicting the traffic flow based on the trained model.
2. The traffic flow prediction method based on the dynamic space-time diagram convolution circulation network according to claim 1, wherein the step 1 is to construct a dynamic space-time relation diagram of a road network, and the specific steps include:
step 1-1, constructing a static diagram based on space distance:
constructing a weighted topological relation diagram of the road network according to the geographic position information of the traffic flow data sensor, wherein the sensor is used as a node of the diagram, and the connection relation of the sensor is used as an edge of the diagram;
constructing an adjacency matrix to describe the spatial relationship of the sensor, wherein the weight of the adjacency matrix is calculated by the spatial distance of the sensor in the road network, and specifically comprises the following steps:
Figure FDA0004146237800000031
wherein A is ij Dist The connection weights of node i and node j in the adjacency matrix representing the static graph,
Figure FDA0004146237800000032
sigma, the spatial distance in the road network between node i and node j 1 Epsilon is the standard deviation of the distance between nodes 1 Is a control threshold;
step 1-2, constructing a dynamic diagram based on flow correlation:
the pearson correlation coefficient is used for calculating the correlation of the flow between adjacent nodes, and the method specifically comprises the following steps:
suppose node V i The flow rate of (2) is
Figure FDA0004146237800000033
Neighbor node V j The flow rate of (2) is
Figure FDA0004146237800000034
T is the current time, T is a time sliding time window, and node V i And V is equal to j The correlation of the inter-flow is expressed as:
Figure FDA0004146237800000035
correlation Sim between nodes (V i ,V j ) The method comprises the following steps:
Sim(V i ,V j )＝|r(X i ,X j )|∈[0,1],
Figure FDA0004146237800000036
constructing an adjacency matrix to describe a dynamic space-time topological graph of the correlation among the nodes, wherein the weight of the adjacency matrix is calculated by the correlation among the nodes, and specifically comprises the following steps:
Figure FDA0004146237800000037
wherein A is ij Sim Connection weight value sigma of node i and node j in adjacency matrix representing dynamic graph 2 Is the standard deviation, epsilon, of the correlation of traffic between nodes 2 Is a control threshold;
step 1-3, stacking a static diagram based on space distance and a dynamic diagram based on flow correlation, and constructing a dynamic space-time relation diagram of a road network:
Figure FDA0004146237800000041
wherein A is ij Connection weight value sigma of node i and node j in adjacency matrix representing dynamic space-time relation diagram 1 Sigma is the standard deviation of the distance between nodes 2 The standard deviation of the correlation of the flow between the nodes is shown, alpha and beta are weight coefficients, and epsilon is a control threshold.
3. The traffic flow prediction method based on the dynamic space-time diagram convolutional loop network according to claim 1, wherein the traffic flow data preprocessing and data set dividing in step 4 specifically comprises:
step 4-1, selecting a real-world traffic flow data set comprising geographic position information of a sensor and traffic flow data acquired by the sensor;
step 4-2, data cleaning is carried out on the traffic flow data set, and a linear interpolation method is adopted to fill missing data;
step 4-3, carrying out standardization processing on the data set by adopting a Z-Score method;
and 4-4, dividing the data set into a training set, a testing set and a verification set by adopting a leave-out method.
4. The traffic flow prediction method based on the dynamic space-time diagram convolutional cyclic network according to claim 1, wherein the specific method for training the traffic flow prediction model based on the dynamic space-time diagram convolutional cyclic network is as follows:
step 5-1, selecting a mean square error MSE as a loss function, selecting an Adam algorithm as a parameter optimizer, setting a model parameter learning rate and setting a Batch size;
and 5-2, randomly selecting data with the Batch size from the training set, sending the data into a model for training, and continuously updating model parameters until the training stopping condition is met, so as to obtain a trained traffic flow prediction model.
5. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any one of claims 1 to 4 when the computer program is executed.
6. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 4.
CN202111575360.0A 2021-12-21 2021-12-21 Traffic flow prediction method, equipment and storage medium based on dynamic space-time diagram convolution circulation network Active CN114220271B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111575360.0A CN114220271B (en) 2021-12-21 2021-12-21 Traffic flow prediction method, equipment and storage medium based on dynamic space-time diagram convolution circulation network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111575360.0A CN114220271B (en) 2021-12-21 2021-12-21 Traffic flow prediction method, equipment and storage medium based on dynamic space-time diagram convolution circulation network

Publications (2)

Publication Number Publication Date
CN114220271A CN114220271A (en) 2022-03-22
CN114220271B true CN114220271B (en) 2023-06-30

Family

ID=80704891

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111575360.0A Active CN114220271B (en) 2021-12-21 2021-12-21 Traffic flow prediction method, equipment and storage medium based on dynamic space-time diagram convolution circulation network

Country Status (1)

Country Link
CN (1) CN114220271B (en)

Families Citing this family (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114004152B (en) * 2021-10-29 2022-11-22 河海大学 Multi-wind-field wind speed space-time prediction method based on graph convolution and recurrent neural network
CN114694379B (en) * 2022-03-29 2024-05-03 中山大学 Traffic flow prediction method and system based on self-adaptive dynamic graph convolution
CN114723127B (en) * 2022-03-31 2024-07-26 河钢数字技术股份有限公司 Steel mill air quality prediction method based on correlation diagram convolutional network
CN114724386B (en) * 2022-03-31 2023-10-27 温州大学 Short-time traffic flow prediction method and system under intelligent traffic and electronic equipment
CN114664090B (en) * 2022-04-14 2023-07-04 山东大学 Traffic data filling method and system based on cyclic neural network
CN114781609B (en) * 2022-04-19 2023-04-25 华东交通大学 Traffic flow prediction method based on multi-mode dynamic residual map convolution network
CN114973653B (en) * 2022-04-27 2023-12-19 中国计量大学 Traffic flow prediction method based on space-time diagram convolutional network
CN115482656B (en) * 2022-05-23 2023-09-26 汕头大学 Traffic flow prediction method by using space dynamic graph convolutional network
CN115205306A (en) * 2022-08-02 2022-10-18 吉林建筑大学 Medical image segmentation method based on graph convolution
CN115376317B (en) * 2022-08-22 2023-08-11 重庆邮电大学 Traffic flow prediction method based on dynamic graph convolution and time sequence convolution network
CN115512545B (en) * 2022-09-30 2023-06-27 浙江财经大学 Traffic speed prediction method based on space-time dynamic graph convolution network
CN115755219B (en) * 2022-10-18 2024-04-02 长江水利委员会水文局 STGCN-based flood forecast error real-time correction method and system
CN116363874B (en) * 2023-03-20 2024-04-23 南京理工大学 Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation
CN116363878B (en) * 2023-05-26 2023-08-11 云南大学 Traffic flow prediction system and method based on continuous dynamic ordinary differential equation
CN118298639A (en) * 2024-04-28 2024-07-05 南京理工大学 Traffic flow prediction method based on multi-span feature extraction

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103700255B (en) * 2013-12-30 2015-10-07 复旦大学 A kind of traffic flow forecasting method based on spacetime correlation data mining
KR101969064B1 (en) * 2018-10-24 2019-04-15 주식회사 블루시그널 Method of predicting road congestion based on deep learning and controlling signal and server performing the same
CN109461311B (en) * 2018-12-19 2020-11-10 哈尔滨工业大学（深圳） Road network traffic flow space-time prediction method for intelligent traffic and intelligent driving
CN110491129A (en) * 2019-09-24 2019-11-22 重庆城市管理职业学院 The traffic flow forecasting method of divergent convolution Recognition with Recurrent Neural Network based on space-time diagram
CN110827544B (en) * 2019-11-11 2022-09-02 重庆邮电大学 Short-term traffic flow control method based on graph convolution recurrent neural network
CN112216108B (en) * 2020-10-12 2021-06-29 中南大学 Traffic prediction method based on attribute-enhanced space-time graph convolution model
CN113487088B (en) * 2021-07-06 2024-09-13 哈尔滨工业大学（深圳） Traffic prediction method and device based on dynamic space-time diagram convolution attention model

Also Published As

Publication number Publication date
CN114220271A (en) 2022-03-22

Similar Documents

Publication Publication Date Title
CN114220271B (en) Traffic flow prediction method, equipment and storage medium based on dynamic space-time diagram convolution circulation network
Liang et al. A novel wind speed prediction strategy based on Bi-LSTM, MOOFADA and transfer learning for centralized control centers
Ma et al. Improving air quality prediction accuracy at larger temporal resolutions using deep learning and transfer learning techniques
Shahriari et al. Ensemble of ARIMA: combining parametric and bootstrapping technique for traffic flow prediction
Wang et al. Multiple convolutional neural networks for multivariate time series prediction
Bi et al. A hybrid prediction method for realistic network traffic with temporal convolutional network and LSTM
Khan et al. Development and evaluation of recurrent neural network-based models for hourly traffic volume and annual average daily traffic prediction
Wang et al. A compound framework for wind speed forecasting based on comprehensive feature selection, quantile regression incorporated into convolutional simplified long short-term memory network and residual error correction
CN113673769B (en) Traffic flow prediction method of graph neural network based on multivariate time sequence interpolation
Sun et al. Hourly PM2. 5 concentration forecasting based on mode decomposition-recombination technique and ensemble learning approach in severe haze episodes of China
Jia et al. Missing data imputation for traffic congestion data based on joint matrix factorization
CN110570035B (en) People flow prediction system for simultaneously modeling space-time dependency and daily flow dependency
CN113570859B (en) Traffic flow prediction method based on asynchronous space-time expansion graph convolution network
Bi et al. Multi-indicator water quality prediction with attention-assisted bidirectional LSTM and encoder-decoder
He et al. MTAD‐TF: Multivariate Time Series Anomaly Detection Using the Combination of Temporal Pattern and Feature Pattern
CN112862177B (en) Urban area aggregation degree prediction method, device and medium based on deep neural network
Yan et al. Small watershed stream-flow forecasting based on LSTM
Xia et al. SW-BiLSTM: a Spark-based weighted BiLSTM model for traffic flow forecasting
CN112598165A (en) Private car data-based urban functional area transfer flow prediction method and device
Wang et al. Deep temporal multi-graph convolutional network for crime prediction
Ayus et al. Comparison of machine learning and deep learning techniques for the prediction of air pollution: a case study from China
CN116108984A (en) Urban flow prediction method based on flow-POI causal relationship reasoning
Liu et al. Multi-stage residual life prediction of aero-engine based on real-time clustering and combined prediction model
Feng et al. Stream-flow forecasting based on dynamic spatio-temporal attention
Huo et al. Traffic anomaly detection method based on improved GRU and EFMS-Kmeans clustering

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
GR01 Patent grant