Improved complementary integrated empirical mode decomposition based on spatiotemporal convolution short-term traffic flow prediction method

Abstract

The present invention discloses a short-term traffic flow prediction method of spatiotemporal convolution based on improved complementary integrated empirical modulus decomposition, including: obtaining original traffic flow data and performing data preprocessing; inputting traffic flow data into a pre-constructed CEEMD-GAT-Attention model for short-term traffic flow prediction. The constructed prediction model includes: using improved CEEMD to decompose traffic flow data into relatively stable time series signals, using GAT to fully mine the spatial features of the data, and using Attention to calculate the attention weight of each spatiotemporal feature, thereby improving the model's ability to capture spatiotemporal features and prediction accuracy. The present invention can further improve the accuracy of short-term traffic flow prediction and provide favorable reference value for the transportation department.

Description

Technical Field

[0001] The invention relates to a spatiotemporal convolution short-term traffic flow prediction method based on improved complementary integrated empirical modulus decomposition, and belongs to the technical field of intelligent transportation. Background Art

[0002] Only by obtaining the traffic status of the road ahead in advance can we make proactive decisions. The effects of traffic control and guidance usually take a period of accumulation to show up. According to different time scales, traffic flow prediction is divided into long-term and short-term, which are used to predict and analyze the number of vehicles passing through a certain section of the road within 5-30 minutes.

[0003] How to improve the accuracy of traffic flow prediction has become one of the important challenges in traffic flow research at this stage. Accurate short-term traffic flow prediction is currently the most fundamental and effective measure to solve urban road traffic congestion and reduce environmental pollution. With the continuous accumulation of traffic flow data and the continuous development of related technologies, prediction methods are gradually evolving from the original classical statistical methods to data-driven perspectives. Traffic flow sequence has the characteristics of complex non-stationarity, spatial correlation and time dependence. The noisy components in the traffic flow signal, the spatial structure of the road network, the time dependence and the external factors in the road system will have a certain impact on traffic flow prediction. At present, there is a lack of a short-term traffic flow prediction method with high accuracy and good stability. Summary of the invention

[0004] In view of the defects in the above-mentioned background technology, the present invention provides a spatiotemporal convolution short-term traffic flow prediction method based on improved complementary integrated empirical modulus decomposition, aiming at improving the accuracy of short-term traffic flow prediction.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] The present invention provides a time-space convolution short-term traffic flow prediction method based on improved complementary integrated empirical mode decomposition, comprising:

[0007] Firstly, the original traffic flow data is obtained and preprocessed to generate a data set;

[0008] The traffic flow dataset is input into the pre-built CEEMD-GAT-Attention model for short-term traffic flow prediction;

[0009] The prediction model construction process based on the improved CEEMD-GAT-Attention includes:

[0010] Mathematical modeling of CEEME: CEEMD is a decomposition algorithm based on EMD and improved on EEMD to process nonlinear signals according to noise frequency. By introducing positive and negative paired Gaussian white noise into the traffic flow data set, the mixed signal is decomposed by EMD at the same time.

[0011] Mathematical modeling of the improved CEEMD: After CEEMD decomposition, the data will represent each IMF component from high to low according to the frequency. The high-frequency IMF component has a large spectrum value and a small vibration frequency. Noise is the dominant mode, which will directly affect the prediction accuracy of the model. Therefore, the highest frequency component after CEEMD decomposition is discarded, and the sample entropy of the remaining IMF component and RES component is calculated to reconstruct the decomposed signal.

[0012] Mathematical modeling of GAT: Calculate the hidden value between each node and adjacent nodes for the input traffic flow data set, assign different weights, and combine the multi-head attention mechanism to obtain more node feature information and improve the learning ability of the model;

[0013] Mathematical modeling of the improved CEEMD-GAT-Attention prediction model is carried out: the improved CEEMD is used to decompose and reconstruct the traffic flow data signal to mine the temporal characteristics of the data; the reconstructed signal is input into the GAT model to capture the spatial characteristics of the data; the Attention mechanism is used to calculate the attention weights of the spatiotemporal features, extract key information for learning, and perform feature fusion through the fully connected layer to output the predicted value.

[0014] Furthermore, the linear interpolation method is used to process missing values ​​in the data, the data is reaggregated into a data format with one record every 20 minutes, and the min-max method is used to normalize the traffic flow data.

[0015] Furthermore, a traffic flow prediction model is constructed, and the traffic flow data set is input into the traffic flow prediction model. The traffic flow prediction model is trained according to the input traffic flow data set and outputs a predicted value, the predicted value is compared with the corresponding true value, and the loss value between the predicted value and the true value is calculated. Iterative training is performed according to the loss value until a preset maximum number of iterations is reached or the loss value converges.

[0016] Furthermore, the traffic flow dataset is decomposed by CEEMD. By introducing positive and negative paired Gaussian white noise into the traffic flow dataset, it is decomposed into several IMF components and a residual component:

[0017]

[0018] Where M1 represents the original sequence after adding positive noise, M2 represents the original sequence after adding negative noise, S represents the traffic flow dataset, and N represents the added Gaussian white noise.

[0019] Perform EMD decomposition on M1 and M2 to obtain 2n groups of IMF components, and denote each group as Find the average of each group of IMF components:

[0020]

[0021] in, It represents the jth IMF component after adding positive noise to the i-th signal; It represents the jth IMF component after adding negative noise to the i-th signal; IMF j It represents the jth IMF component obtained after the signal is decomposed by CEEMD. In summary, the traffic flow dataset S is finally decomposed into the data form of multiple IMF components and a residual component res.

[0022]

[0023] Furthermore, the sample entropy is used to calculate the sample entropy value of each component after CEEMD decomposition, and the highest frequency IMF component with large spectrum value, small vibration frequency and strong noise interference is discarded, and the remaining IMF components are reconstructed to regenerate the random component HF, characteristic component MF and trend component LF, and retain the original RES as the residual component;

[0024]

[0025] In the formula, m is the given pattern dimension, r is the given threshold, and B m (r) is the ratio of the number of m-dimensional data with a distance less than r to the total number of distances, B m+1 (r) is the ratio of the number of m+1-dimensional data whose distance is less than r to the total number of distances;

[0026]

[0027] Furthermore, to capture the spatial characteristics of the traffic flow dataset, GAT is used to calculate the attention coefficient e between node i and node j. ij , and use the softmax function to normalize all adjacent nodes, and use the attention coefficient, adopt the multi-head attention mechanism, and use the weighted summation method to update the feature vector of each node;

[0028] e ij =α(Wh i ,Wh j )

[0029]

[0030]

[0031] In the formula, α represents the function used to calculate the correlation, W represents a parameter matrix that linearly transforms all node features to convert the input features into high-dimensional features, σ represents the activation function, and N i represents the neighboring nodes directly connected to node i, || represents the connection operation, K represents the number of attention heads, represents the attention coefficient under the kth attention head, W k Represents the linear transformation matrix under the kth attention head.

[0032] Furthermore, the attention mechanism is used to assign different weights to the input spatiotemporal features, from which more important and critical information is selected for learning, thus improving the model's operational efficiency and prediction accuracy;

[0033] e t =tanh(w e h t +b e )

[0034]

[0035]

[0036] In the formula, e t Represents the input feature vector h at time t t The attention value, w e and b e are weights and bias terms respectively; use the softmax function to t Perform normalization to obtain the weight coefficient α at time t t ; Perform weighted summation on the feature vectors to obtain the output value p of the attention mechanism t .

[0037] Furthermore, the evaluation index for measuring the loss value between the predicted value and the true value includes:

[0038]

[0039]

[0040]

[0041] Where N is the number of traffic flow data, y i is the experimental prediction value, x i is the actual value of traffic flow data, is the mean of the actual values ​​of the traffic flow data.

[0042] Furthermore, a device includes:

[0043] one or more processors;

[0044] A memory for storing one or more programs;

[0045] When one or more of the programs are executed by one or more of the processors, the one or more processors implement the spatiotemporal convolution short-term traffic flow prediction method based on complementary integrated empirical mode decomposition as described above.

[0046] Furthermore, a storage medium containing computer executable instructions, which, when executed by a computer processor, are used to execute the spatiotemporal convolution short-term traffic flow prediction method based on complementary integrated empirical mode decomposition as described above.

[0047] Beneficial effects of the present invention:

[0048] The present invention uses an improved CEEMD module to decompose traffic flow data according to noise frequency, and introduces positive and negative paired Gaussian white noise into the traffic flow data set to overcome the modal aliasing phenomenon existing in the EMD module; the sample entropy value of each component after decomposition is calculated and reconstructed, which can not only reduce the calculation time, but also because complementary noise is added, the noise residual amount in the reconstructed data can be almost ignored;

[0049] The present invention adopts the graph attention network layer GAT and the attention mechanism Attention to establish a prediction model. GAT calculates the attention coefficient between adjacent nodes and allows adjacent nodes to obtain different weights according to their own characteristics, so as to better capture the spatial characteristics of the traffic flow data set. Attention can help the model assign different weights to the input spatiotemporal features, select more important and critical information for learning, and improve the model's operational efficiency and prediction accuracy.

[0050] The present invention provides a spatiotemporal convolution short-term traffic flow prediction method based on improved complementary integrated empirical modulus decomposition; accurate short-term traffic flow prediction has important practical significance in the field of intelligent transportation, can provide travelers with more accurate road information, improve road traffic efficiency, solve traffic congestion and other problems, strengthen the connection between vehicles and roads, and is an important part of building a smart city. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The present invention will be further described below in conjunction with the accompanying drawings.

[0052] Figure 1 A CEEMD decomposition flow chart provided in Example 1 of the present invention;

[0053] Figure 2 A schematic diagram of the GAT structure provided in Embodiment 1 of the present invention;

[0054] Figure 3 This is a schematic diagram of the Attention structure provided in the first embodiment of the present invention;

[0055] Figure 4 This is an overall flow chart provided for Embodiment 1 of the present invention; DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Embodiment 1:

[0058] The embodiment of the present invention provides a method for predicting short-term traffic flow using space-time convolution based on improved complementary integrated empirical mode decomposition, comprising the following steps:

[0059] 1. First, obtain the original traffic flow data and perform data preprocessing to generate a data set;

[0060] 2. Input the traffic flow dataset into the pre-built CEEMD-GAT-Attention model for short-term traffic flow prediction;

[0061] The following mainly introduces the construction process of the improved CEEMD-GAT-Attention model:

[0062] Mathematical modeling of the improved CEEMD: CEEMD decomposition is performed on the traffic flow dataset. By introducing positive and negative paired Gaussian white noise into the traffic flow dataset, it is decomposed into several IMF components and a residual component:

[0063]

[0064] Where M1 represents the original sequence after adding positive noise, M2 represents the original sequence after adding negative noise, S represents the traffic flow dataset, and N represents the added Gaussian white noise.

[0065] Perform EMD decomposition on M1 and M2 to obtain 2n groups of IMF components, and denote each group as Find the average of each group of IMF components:

[0066]

[0067] in, It represents the jth IMF component after adding positive noise to the i-th signal; It represents the jth IMF component after adding negative noise to the i-th signal; IMF j It represents the jth IMF component obtained after the signal is decomposed by CEEMD. In summary, the traffic flow dataset S is finally decomposed into the data form of multiple IMF components and a residual component res.

[0068]

[0069] like Figure 1 As shown, it is the CEEMD decomposition flow chart;

[0070] The sample entropy is used to calculate the sample entropy value of each component after CEEMD decomposition, and the highest frequency IMF component with large spectrum value, small vibration frequency and strong noise interference is discarded, and the remaining IMF components are reconstructed to regenerate the random component HF, characteristic component MF and trend component LF, and retain the original RES as the residual component;

[0071]

[0072] In the formula, m is the given pattern dimension, r is the given threshold, and B m (r) is the ratio of the number of m-dimensional data with a distance less than r to the total number of distances, B m+1 (r) is the ratio of the number of m+1-dimensional data whose distance is less than r to the total number of distances;

[0073]

[0074] Mathematical modeling of GAT: Calculate the hidden value between each node and adjacent nodes for the input traffic flow data set, assign different weights, and combine the multi-head attention mechanism to obtain more node feature information and improve the learning ability of the model;

[0075] like Figure 2 As shown, it is the structural diagram of the GAT network;

[0076] GAT is used to calculate the attention coefficient e between node i and node j ij , and use the softmax function to normalize all adjacent nodes, and use the attention coefficient, adopt the multi-head attention mechanism, and use the weighted summation method to update the feature vector of each node;

[0077] e ij =α(Wh i ,Wh j )

[0078]

[0079]

[0080] In the formula, α represents the function used to calculate the correlation, W represents a parameter matrix that linearly transforms all node features to convert the input features into high-dimensional features, σ represents the activation function, and N i represents the neighboring nodes directly connected to node i, || represents the connection operation, K represents the number of attention heads, represents the attention coefficient under the kth attention head, W k Represents the linear transformation matrix under the kth attention head.

[0081] Mathematical modeling of Attention: The attention mechanism is used to assign different weights to the input spatiotemporal features, and more important and critical information is selected for learning, thereby improving the model's operational efficiency and prediction accuracy.

[0082] e t =tanh(w e h t +b e )

[0083]

[0084]

[0085] In the formula, e t Represents the input feature vector h at time t t The attention value, w e and b e are weights and bias terms respectively; use the softmax function to t Perform normalization to obtain the weight coefficient α at time t t ; Perform weighted summation on the feature vectors to obtain the output value p of the attention mechanism t ;

[0086] like Figure 3 As shown, this is the structural diagram of the Attention mechanism;

[0087] The improved CEEMD-GAT-Attention model is mathematically modeled: CEEMD is used to decompose traffic flow data signals, mine the temporal characteristics of the data, and calculate the sample entropy values ​​of each component after decomposition for reconstruction; the reconstructed signal is input into the GAT model to capture the spatial characteristics of the data; the Attention mechanism is used to calculate the attention weights of the spatiotemporal characteristics, extract key information for learning, and perform feature fusion through the fully connected layer to output the predicted value.

[0088] A traffic flow prediction model is constructed, and a traffic flow data set is input into the traffic flow prediction model. The traffic flow prediction model is trained according to the input traffic flow data set and outputs a prediction value, the prediction value is compared with the corresponding true value, and the loss value between the prediction value and the true value is calculated. Iterative training is performed according to the loss value until a preset maximum number of iterations is reached or the loss value converges.

[0089] like Figure 4 As shown, it is the overall flow chart of SE-CEEMD-GAT-Attention;

[0090] The missing values ​​of the data are processed by linear interpolation method, the data are reaggregated into a data format with one record every 20 minutes, and the traffic flow data are normalized by min-max method.

[0091] The dataset in this paper comes from the traffic flow data and weather data of a highway in a certain province provided by KDD CUP 2017. The traffic flow data comes from three toll stations, of which toll stations 1 and 3 allow vehicles to travel in both directions, and toll station 2 only allows vehicles to enter but not exit. Toll station 1 is selected as the central node, and toll stations 2 and 3 are selected as neighbor nodes. The weather data contains seven features: air pressure, sea water pressure, wind direction, wind speed, temperature, humidity, and rainfall. The data frequency is recorded every 3 hours. The time period of traffic flow data and weather data is from September 19, 2016 to October 17, 2016. According to the characteristics of the data, the first 90% is used as the training set, the last 10% is used as the test set, and the sliding window is 7.

[0092] In this paper, the PyTorch neural network framework is used in the Python 3.6 environment to complete the construction and training of the model. Auxiliary white noise with an amplitude of 0.2 is selected to perform CEEMD decomposition on the traffic flow dataset; the number of hidden units in the GAT network layer is set to 128, and the number of attention heads K is set to 3; the Adam optimizer is used to optimize the model, and the learning rate is set to 0.001, Batch_size is 32, dropout is 0.2, and the number of iterations is 50.

[0093] In order to verify the prediction effect of the proposed model on traffic flow data, the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and determination coefficient (R 2 ) as the evaluation indicator:

[0094]

[0095]

[0096]

[0097] Where N is the number of traffic flow data, y i is the experimental prediction value, x i is the actual value of traffic flow data, is the mean of the actual values ​​of the traffic flow data.

[0098] The model of the embodiment of the present invention is experimentally compared with other five models on three indicators on a public data set. The results are shown in Table 1.

[0099]

[0100] Table 1

[0101] It can be seen from Table 2 that the traffic flow prediction model proposed in the embodiment of the present invention can achieve good results in the evaluation indicators RMSE, MAE, and MAPE. Compared with other models, the prediction accuracy is significantly improved, and it can more accurately reflect the changes in road traffic flow, providing a certain reference value for travelers.

[0102] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0103] It needs to be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program is executed by a processor to execute the above method. The storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0104] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0105] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. Improved complementary integrated empirical mode decomposition of spatiotemporal convolution short-term traffic flow prediction method, characterized by: The following steps are involved: Acquire original traffic flow data, and perform data preprocessing on the original traffic flow data to generate a traffic flow data set; Input the traffic flow dataset into the pre-built improved CEEMD-GAT-Attention model for training and prediction; The prediction model construction process based on the improved CEEMD-GAT-Attention includes: Mathematical modeling of CEEMD: CEEMD is a decomposition algorithm based on EMD and improved on EEMD to process nonlinear signals according to noise frequency. By introducing positive and negative paired Gaussian white noise into the traffic flow data set, the mixed signal is decomposed by EMD at the same time. Mathematical modeling of the improved CEEMD: After CEEMD decomposition, the data will represent each IMF component from high to low according to the frequency. The high-frequency IMF component has a large spectrum value and a small vibration frequency. Noise is the dominant mode, which will directly affect the prediction accuracy of the model. Therefore, the sample entropy is used to calculate the sample entropy value of each component after CEEMD decomposition, and the highest frequency IMF component with a large spectrum value, a small vibration frequency, and strong noise interference is discarded. The remaining IMF components are reconstructed, and the random component HF, the characteristic component MF, and the trend component LF are regenerated. The original RES is retained as the residual component to reconstruct the decomposed signal. In the formula, m is the given pattern dimension, r is the given threshold, and B m (r) is the ratio of the number of m-dimensional data with a distance less than r to the total number of distances, B m+1 (r) is the ratio of the number of m+1-dimensional data whose distance is less than r to the total number of distances; Mathematical modeling of GAT: Calculate the hidden value between each node and adjacent nodes for the input traffic flow data set, assign different weights, and combine the multi-head attention mechanism to obtain more node feature information and improve the learning ability of the model; Mathematical modeling of the improved CEEMD-GAT-Attention prediction model is carried out: the improved CEEMD is used to decompose and reconstruct the traffic flow data signal to mine the temporal characteristics of the data; the reconstructed signal is input into the GAT model to capture the spatial characteristics of the data; the Attention mechanism is used to calculate the attention weights of the spatiotemporal features, extract key information for learning, and perform feature fusion through the fully connected layer to output the predicted value.

2. The improved complementary integrated empirical mode decomposition spatiotemporal convolution short-term traffic flow prediction method according to claim 1 is characterized in that: The data preprocessing operation specifically includes: using a linear interpolation method to process missing values ​​of the data, re-aggregating the data into a data format with one entry every 20 minutes, and using a min-max method to normalize the traffic flow data.

3. The improved complementary integrated empirical mode decomposition spatiotemporal convolution short-term traffic flow prediction method according to claim 1 is characterized in that: The training process of the traffic flow prediction model includes: constructing a traffic flow prediction model, and inputting a traffic flow data set into the traffic flow prediction model, the traffic flow prediction model is trained according to the input traffic flow data set and outputs a prediction value, the prediction value is compared with the corresponding true value, the loss value between the prediction value and the true value is calculated, and iterative training is performed according to the loss value until a preset maximum number of iterations is reached or the loss value converges.

4. The improved complementary integrated empirical mode decomposition spatiotemporal convolution short-term traffic flow prediction method according to claim 1 is characterized in that: The EMD can decompose a complex signal into a finite number of IMF components, each of which contains a local characteristic signal of the original signal at different time scales, that is, decomposing the original sequence into an eigenmode function and a residual component; Where m is the number of IMF components; c i (t) is the i-th IMF component; r m (t) is the mth residual component; The EEMD adds auxiliary white noise to calculate the aggregate average of the IMF components obtained after multiple EMD decompositions; the CEEMD introduces positive and negative paired Gaussian white noise into the traffic flow data set to decompose it into several IMF components and a residual component; the signal decomposition process includes: Where M1 represents the original sequence after adding positive noise, M2 represents the original sequence after adding negative noise, S represents the traffic flow dataset, and N represents the added Gaussian white noise; Perform EMD decomposition on M1 and M2 to obtain 2n groups of IMF components, and denote each group as Find the average of each group of IMF components: in, It represents the jth IMF component after adding positive noise to the i-th signal; It represents the jth IMF component after adding negative noise to the i-th signal; IMF j represents the jth IMF component obtained after the signal is decomposed by CEEMD; in summary, the traffic flow dataset S is finally decomposed into multiple IMF components and a residual component RES.

5. The improved complementary integrated empirical mode decomposition spatiotemporal convolution short-term traffic flow prediction method according to claim 1 is characterized in that: The GAT calculates the attention coefficient e between nodes i and j. ij , and use the softmax function to normalize all adjacent nodes, and use the attention coefficient, adopt the multi-head attention mechanism, and use the weighted summation method to update the feature vector of each node; e ij =α(Wh i ,Wh j ) In the formula, h i represents the feature set of node i, h j represents the feature set of node j, α represents the function used to calculate the correlation, W represents a parameter matrix for linear transformation of all node features, converting the input features into high-dimensional features, exp represents the exponential function with e as the base, σ represents the activation function, and N i represents the neighbor nodes directly connected to node i, K represents the number of attention heads, represents the attention coefficient under the kth attention head, W k Represents the linear transformation matrix under the kth attention head.

6. The improved complementary integrated empirical mode decomposition spatiotemporal convolution short-term traffic flow prediction method according to claim 1 is characterized in that: The attention mechanism is used to assign different weights to the input spatiotemporal features, and more important and critical information is selected for learning, thereby improving the computational efficiency and prediction accuracy of the model; e t =tanh(w e h t +b e ) In the formula, e t Represents the input feature vector h at time t t The attention value, w e and b e are weights and bias terms respectively; use the softmax function to t Perform normalization to obtain the weight coefficient α at time t t ; Perform weighted summation on the feature vectors to obtain the output value p of the attention mechanism t .

7. The improved complementary integrated empirical mode decomposition spatiotemporal convolution short-term traffic flow prediction method according to claim 3 is characterized in that: The evaluation indicators used to measure the loss between the predicted value and the true value include: Where N is the number of traffic flow data, y i is the experimental prediction value, x i is the actual value of traffic flow data, is the mean of the actual values ​​of the traffic flow data.

8. A device, characterized in that The device comprises: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors execute the spatiotemporal convolution short-term traffic flow prediction method based on improved complementary integrated empirical mode decomposition as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the improved complementary integrated empirical mode decomposition spatiotemporal convolution short-term traffic flow prediction method as described in any one of claims 1 to 7 is implemented.

Images