Rail transit short-time passenger flow prediction method and device based on multi-scale convolution and storage medium

Abstract

The invention relates to the technical field of rail transit passenger flow analysis, in particular to a rail transit short-time passenger flow prediction method and device based on multi-scale convolution and a storage medium. Forming a multi-source data set through historical time sequence passenger flow data, station network topology data, external environment data and the like, and extracting fusion feature data; extracting multi-scale time characteristics of the passenger flow by using a multi-scale convolution module, and reserving bottom information in combination with residual connection; the over-smoothing problem of the deep network is relieved through the improved spatial dependency relationship between the residual image convolutional network modeling sites; and dynamically fusing the spatial-temporal features and the external environment features by adopting an attention mechanism, and generating a prediction result through a long-short term memory network. According to the method, the passenger flow prediction precision and the model robustness are remarkably improved, the method can adapt to complex external environment changes, and high-reliability decision support is provided for rail transit dispatching management.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit passenger flow analysis, and in particular to a short-term rail transit passenger flow prediction method, device and storage medium based on multi-scale convolution. Background Art

[0002] Under the background of accelerating urbanization, the growth of urban rail transit demand has significantly exacerbated traffic congestion, posing a challenge to the sustainable development of social economy. In response to this problem, many cities have vigorously promoted the development of rail transit infrastructure, including subways and light rails, effectively expanding the network scale and improving the operation decision-making and service efficiency. Timely and accurate prediction of short-term rail passenger flow is the key to improving the operation efficiency and management level of the rail system. The short-term rail passenger flow prediction methods can be roughly divided into three categories: the first category is the prediction method based on statistical theory, the second category is the prediction method based on deep learning, and the third category is the prediction method based on model combination, which can give full play to the advantages of individual models and achieve good prediction results in short-term rail passenger flow prediction.

[0003] Specifically, the first category of prediction methods based on statistical theory includes linear regression, Kalman filter, gradient boosting decision tree, etc. Although it is relatively simple in operation, it has very poor effects when dealing with passenger flow data with large volatility and non-linearity.

[0004] The second category of prediction methods based on machine learning, due to its powerful ability to process non-linear data, is widely used in short-term passenger flow prediction. Such as decision tree, random forest, multi-layer perceptron, backpropagation (BP) neural network, etc. When making predictions using the above methods, since rail passenger flow data depends on spatio-temporal characteristics, that is, there is spatio-temporal correlation between different stations, and only predicting one or several stations separately is difficult to fully extract the spatio-temporal characteristics of passenger flow, so the final prediction will be inaccurate.

[0005] The third category is prediction methods represented by deep learning models. In recent years, with the rapid development of deep learning, the model performance in the field of traffic prediction has been continuously improved. Models such as recurrent neural network (RNN), convolutional neural network (CNN), long short-term memory network (LSTM), and graph convolutional neural network (GCN) have been gradually applied to short-term passenger flow prediction of rail transit. In order to improve the accuracy of prediction results and the robustness of the model, hybrid prediction frameworks using multi-source data have been proposed successively in recent years. For example, a ResLSTM deep learning architecture is proposed based on ResNet, GCN, and LSTM networks, considering the influence of air quality factors to complete short-term passenger flow prediction at the urban rail transit network level. According to the fusion stage of multi-source data, some models complete the fusion at the feature level, while some other studies consider the fusion at the result level, and the latter cannot fully utilize the advantages of multi-source data to complete complementarity. Moreover, during the operations of convolution, pooling, dimensionality reduction, and fusion in deep learning models, information loss is inevitable, and the internal correlation between multi-source data is often ignored.

[0006] To sum up, the existing short-term passenger flow prediction models for rail transit mainly have the following problems:

[0007] (1) Incompleteness of feature factors.

[0008] The existing research on short-term passenger flow prediction of rail transit is limited to specific stations or lines, or simply classifies stations, ignoring the significant spatial correlation of the entire network, failing to comprehensively reflect the spatial topological characteristics of the entire network, resulting in insufficient prediction of the overall passenger flow dynamics of the network. Therefore, the research scope needs to shift from a single station or line to the entire rail transit network to comprehensively consider and capture the impact of its spatial structure on passenger flow patterns. At the same time, most scholars mainly rely on historical passenger flow data to build prediction models, ignoring external factors such as weather and ground traffic congestion, which makes the model less adaptable to complex scenarios and the prediction results lack accuracy.

[0009] (2) Traditional short-term passenger flow prediction models for rail transit do not fully mine time features.

[0010] Traditional long short-term memory network (LSTM) can effectively solve the long-term dependence problem of time series data, but the changes in rail transit passenger flow have obvious multi-scale characteristics, such as the coexistence of short-term violent fluctuations and long-term overall trends. However, existing methods usually use a single convolution scale for modeling, ignoring the detailed features and context information under different convolution ranges, which makes it difficult for the model to accurately capture the complex dynamic changes of passenger flow data. Summary of the Invention

[0011] To solve the above technical problems, the present invention provides a short-term passenger flow prediction method for rail transit based on multi-scale convolution, which is characterized by including the following steps:

[0012] S1. Obtain historical time series passenger flow data, station network topology data, and external environment data, and perform preprocessing on each respectively to obtain a multi-source data set;

[0013] S2. Respectively extract features from the historical time series passenger flow data, station network topology data, and external environment data in the multi-source data set to obtain feature data for each data source, and perform feature weighted fusion on the feature data through the Hadamard weighting method to obtain fused feature data;

[0014] S3. Construct a passenger flow prediction model based on the fused feature data, input the fused feature data into the passenger flow prediction model to obtain a short-term passenger flow prediction result;

[0015] S4. Use the mean square error function as the loss function, and evaluate the short-term passenger flow prediction result through the root mean square error, mean absolute error, and weighted mean absolute percentage error until the function converges to obtain the final short-term passenger flow prediction result.

[0016] Furthermore, the historical time series passenger flow data includes historical time step in-out passenger flow tensors, weekly cycle in-out passenger flow tensors, and daily cycle in-out passenger flow tensors;

[0017] The weekly cycle in-out passenger flow tensor can be expressed as:

[0018]

[0019] where, is the weekly cycle in-out passenger flow tensor, T is the current time slice, w is the previous week, and TS w is the passenger flow data at the same time period of the previous week, and N is the number of stations;

[0020] The daily cycle in-out passenger flow tensor can be expressed as:

[0021]

[0022] where, is the daily cycle in-out passenger flow tensor, d is the previous day, and TS d is the passenger flow with a time step of TS at the same time period of the previous day;

[0023] Concatenate the weekly cycle in-out passenger flow tensor, the daily cycle in-out passenger flow tensor, and the historical time step in-out passenger flow tensor to obtain an in-out passenger flow cycle tensor, which can be expressed as:

[0024]

[0025] Among them, is the in-out passenger flow cycle tensor.

[0026] Furthermore, the feature extraction of the site network topology data in step 2 includes the following steps:

[0027] S201-A. Obtain the urban rail transit network data and represent it in a graph structure as:

[0028] G = (N, E, A)

[0029] A ∈ R N*N

[0030] Among them, G is the urban rail transit network data, N is the number of stations, E is the number of edges between stations, and A is the adjacency matrix representing whether stations are adjacent;

[0031] S201-B. Obtain the historical passenger flow feature matrix, which can be represented as:

[0032] F = (X t , X t -1, X t -2, …, X t -m+1)

[0033] Among them, F is the feature matrix, X t is the historical passenger flow at time t, and m is the time interval;

[0034] S201-C. Construct a GCN convolution function and calculate the topological relationship between stations in the urban rail transit network data. The GCN convolution function can be represented as:

[0035]

[0036] H ∈ R N*t , H′ ∈ R N*t

[0037] Among them, I is the identity matrix, is the sum of the adjacency matrix A and the identity matrix I, is 's diagonal node degree matrix, l is the number of convolution layers, W is the parameter matrix of the l-th layer, H is the feature matrix, N is the number of stations, t is the feature number of each node, H’ is the feature matrix containing topological information, and σ is the activation function;

[0038] S201-D. Based on the residual function, transform the GCN convolution function, construct a residual graph convolutional network, and extract the site network topology data features. The residual convolutional network can be represented as:

[0039]

[0040] ln ∈ R N*M

[0041] Among them, is the symmetric normalized Laplacian operator, ln is the model input, N is the number of stations, M is the number of historical time steps used for each station, and ln' is the output feature matrix after being transformed by the symmetric normalized Laplacian operator.

[0042] Furthermore, the external environmental data includes weather data and traffic congestion data; the weather data at least includes temperature data, wind speed data, air pressure data, humidity data, and rainfall data; the traffic congestion data at least includes a congestion index, the length of congested sections, and the average vehicle speed; the congestion index is the ratio of the actual formation time to the unobstructed travel time, and the larger the ratio, the higher the degree of congestion; the external environmental data can be represented by a matrix as:

[0043]

[0044] where W is the external environmental data, x is the eigenvalue of the external environmental data, t is the current moment, and l is the length of the time window.

[0045] Furthermore, in step S2, the feature data of each data source is subjected to feature fusion by the Hadamard weighting method to obtain fused feature data, which can be expressed as:

[0046] Fusion = W 1 ⊙ O 1 + W 2 ⊙ O 2 + W 3 ⊙ O 3 + W 4 ⊙ O 4

[0047] where Fusion is the fused feature data, O 1 is the historical time series passenger flow feature data, W 1 is the weight vector of the historical time series passenger flow feature data, O 2 is the station network topology feature data, W 2 is the weight vector of the station network topology data, O 3 is the weather feature data, W 3 is the weight vector of the weather feature data, O 4 is the traffic congestion feature data, W 4 is the weight vector of the traffic congestion data, and ⊙ is the Hadamard product.

[0048] Furthermore, the passenger flow prediction model can be expressed as:

[0049] [Y t+1 , …, Y t+k = f(P t , P t ′, G, W t )

[0050] Among them, P t represents the inbound passenger flow time series matrix at time step t, P t ′ represents the outbound passenger flow time series matrix at time step t, G represents the adjacency matrix of the station topological structure, and W t represents the weather information and traffic congestion index feature matrix. [Y t+1 , …, Y t+k represents the short-term passenger flow prediction values for k time steps from Y t+1 to Y t+k k, and f represents the passenger flow prediction model.

[0051] Furthermore, the passenger flow prediction model includes a multi-scale residual feature extraction module, a residual graph convolutional neural network module, and an attention time series module;

[0052] The multi-scale residual feature extraction module is used to extract the multi-scale characteristics in the historical time series passenger flow data to obtain the spatio-temporal feature map of the passenger flow;

[0053] The residual graph convolutional neural network module is used to extract the topological dependence relationship of the station network topological data to obtain the network topological feature map;

[0054] The attention time series module is used to fuse the external environment data features through the attention long short-term neural network layer and extract the time features of the fused feature data to obtain the time features of the fused feature data.

[0055] Furthermore, the multi-scale residual feature extraction module includes 1*1 convolutional kernel, 3*3 convolutional kernel, 5*5 convolutional kernel, and 7*7 convolutional kernel;

[0056] The 1x1 convolutional kernel is used to perform information weighted combination on the channel dimension of the input feature map, thereby realizing the reduction or increase of the number of channels;

[0057] The 3x3 convolutional kernel is used to capture the rapid changes in data within a local spatio-temporal range and is suitable for capturing the passenger flow fluctuations on a short time scale, such as the passenger flow changes within 15 minutes to 1 hour;

[0058] The 5x5 convolutional kernel is used to extract the data trends within a longer time span and can identify periodic or trend changes, such as long-term patterns during morning and evening rush hours or seasonal fluctuations;

[0059] The 7x7 convolutional kernel is used to capture spatio-temporal patterns spanning multiple time steps and can identify long-term dependencies, such as the passenger flow change trend within multiple hours or a full working day.

[0060] Furthermore, the present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the method for predicting short-term passenger flow of rail transit based on multi-scale convolution described in any one of the above are implemented.

[0061] Furthermore, the present invention also provides a storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for predicting short-term passenger flow of rail transit based on multi-scale convolution described in any one of the above are implemented.

[0062] The beneficial effects of the present invention are as follows: Through the deep learning framework based on the multi-scale convolutional neural network and the long short-term memory neural network, while fully capturing global and local features, the expression ability of multi-source data features is enhanced, thereby improving the prediction accuracy of the model.

[0063] (1) Aiming at the incompleteness of feature factors, by comprehensively introducing and utilizing external data, weather data and traffic congestion index data, the comprehensiveness and expression ability of features for short-term rail passenger flow prediction are improved, the adaptability of the model to external complex situations is enhanced, the prediction results are more accurate. At the same time, integrating traffic and weather data enhances the robustness of the model and significantly reduces the prediction error.

[0064] (2) Aiming at the deficiencies in time feature mining, the multi-scale characteristics in the time series are extracted through the multi-scale residual feature extraction module (Multi-Scale Residual Feature Extractor, MSRFE). At the same time, the depth and stability of feature extraction are improved through the residual structure, and the comprehensive characteristics of short-term and long-term changes in the time series are captured, improving the adaptability of the model to complex dynamic changes. The residual connection structure effectively avoids information loss during the feature extraction process and enhances the ability of deep feature expression, making the model perform better when processing long-time-span data and further improving the prediction accuracy of short-term rail passenger flow. Description of the Drawings

[0065] Figure 1 、Flowchart of the method for predicting short-term passenger flow of rail transit based on multi-scale convolution of the present invention.

[0066] Figure 2 、Schematic diagram of the passenger flow periodicity of the embodiment of the present invention.

[0067] Figure 3 、Schematic diagram of the GCN graph convolutional neural network of the embodiment of the present invention.

[0068] Figure 4 、Schematic diagram of the multi-scale convolution model structure according to an embodiment of the present invention.

[0069] Figure 5 、Schematic diagram of the multi-scale residual feature extraction module structure according to an embodiment of the present invention.

[0070] Figure 6 、Schematic diagram of a computer device structure according to the present invention.

[0071] Figure 7 、Schematic diagram of a storage medium structure according to the present invention.

[0072] In the figure, 200 - terminal device, 210 - memory, 211 - RAM, 212 - cache memory, 213 - ROM, 214 - program / utilities, 215 - program modules, 220 - processor, 230 - bus, 240 - external device, 250 - I / O interface, 260 - network adapter, 300 - program product. Detailed implementation manners

[0073] In order for those skilled in the art to better understand the content of the present invention, and to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with embodiments and the accompanying drawings. The illustrative embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to further limit the present invention.

[0074] Embodiment 1:

[0075] As Figure 1 shown, an embodiment of the present invention provides a short-term passenger flow prediction method for rail transit based on multi-scale convolution, including the following steps:

[0076] S1. Obtain historical time series passenger flow data, site network topology data, and external environment data, and perform preprocessing on each of them to obtain a multi-source data set;

[0077] Specifically, historical time series passenger flow data often has obvious periodicity. Therefore, two types of periodic patterns hidden in the passenger flow data need to be considered, namely weekly periodicity and daily periodicity. Weekly periodicity refers to the similar characteristics of passenger flow at the same time of each week, mainly caused by regular travel activities, such as commuting to and from work or school. Daily periodicity refers to the passenger flow characteristics at the same time of adjacent two days. For example, the passenger flow of adjacent two working days has a similar peak trend. To accurately describe the periodic characteristics of the passenger flow, in addition to the passenger flow data of the historical TS time steps, the following two types of data are introduced in this embodiment.

[0078] 1. Weekly cycle tensor

[0079] To capture the weekly periodicity, in this embodiment, the passenger flow data with a time step of TS in the same period of last week is considered, which can be expressed as In this embodiment, the selected time period is 16.5 hours from 6:30 to 23:00. Assuming the time interval (such as 10 minutes, 15 minutes, etc.) is tl, the time step is ts, and the current time slice is t, the weekly cycle is defined as:

[0080]

[0081] Daily cycle tensor Similarly, to capture the daily periodicity, we consider the passenger flow with a time step of TS in the same period of yesterday and obtain the daily cycle tensor Its definition is as follows:

[0082]

[0083] After obtaining the weekly cycle tensor and the daily cycle tensor , the passenger flow tensor of the historical TS time steps is concatenated with them to obtain a new passenger flow tensor as the input of the model. As Figure 2 shown, the representations of three different passenger flow cycle tensors on the time axis.

[0084] Specifically, when obtaining the site network topology data, the urban rail transit network is characterized by the graph structure G=(V, E, A), and the feature matrix F=(X t , X t-1 , X t-2 ,…X t-m+1 ) at time t represents the historical passenger flow situation in the previous m time intervals. Here, V represents the number of subway stations, there are E edges between stations, A∈R n*n is the adjacency matrix characterizing whether stations are adjacent. The dimension of F is n×m, where n is the number of stations. As Figure 3 shown, the GCN graph convolutional neural network is used to depict the topological dependence relationship between stations in the subway network. The GCN convolution operation is defined as follows:

[0085]

[0086] H∈R N*M , H′∈R N*M

[0087] where I is the identity matrix, is the sum of the adjacency matrix A and the identity matrix I, is The diagonal node degree matrix, l is the number of convolutional layers, W is the parameter matrix of the l-th layer, H is the feature matrix, N is the number of stations, t is the number of features of each node, H’ is the feature matrix containing topological information, and σ is the activation function;

[0088] However, as the number of stacked layers of GCN increases, the performance of the model will deteriorate. Stacking more GCN layers not only leads to higher complexity in the backpropagation process but also causes problems such as vanishing gradients, thus reducing the performance of deep GCN. In addition, the serious "over-smoothing" problem, that is, as the number of layers increases, multiple features of the same vertex converge to the same value, is also a common problem existing in deep GCN. Therefore, residuals are introduced to weaken the deficiencies of GCN, that is, the input matrix In is treated as a graph signal and then processed according to the following formula.

[0089]

[0090] ln ∈ R N*M

[0091] Among them, is the symmetric normalized Laplacian operator, ln is the model input, N is the number of stations, M is the number of historical time steps used for each station, and ln’ is the output feature matrix after being transformed by the symmetric normalized Laplacian operator.

[0092] The transformed ln′ and ln have the same dimension and shape, but ln′ contains rich network topology structure information and is used as the input of ResNet subsequently. At the same time, the GCN used in this embodiment is not the original graph convolution, but the part that depicts the network topology information in the original graph convolution is introduced into the model input, so that the model input contains topological information and can weaken the deficiencies of using the original deep graph convolution.

[0093] Specifically, the external environment data includes weather data and traffic congestion data. Weather information is obtained by crawling weather websites, such as: temperature (°C), wind speed (m / s), air pressure (hPa), humidity (%). Since the obtained weather conditions are recorded every half hour, when performing short-term passenger flow prediction at a 10-minute granularity, the weather conditions from 6:30 to 6:40 are shared with the real data from 6:30 to 7:00, that is, the same set of data is used.

[0094] Traffic congestion information is obtained by crawling the traffic big data platform, such as: congestion index, length of severely congested sections (km), average speed (km / h). Among them, the congestion index is the ratio of the actual travel time to the unobstructed travel time, and the larger the congestion index, the higher the degree of congestion. The finally obtained model input data can be represented by a matrix as:

[0095]

[0096] Among them, W is the external environment data, x is the eigenvalue of the external environment data, x 1 , x 2 , …, x 7 are the eigenvalues after normalization of temperature, wind speed, … average speed respectively, t is the current time, and l is the length of the time window.

[0097] S2. Take the obtained time series passenger flow, station network topology information, weather information, and traffic congestion index data as multi-source data, and perform feature weighted fusion on the feature data through the Hadamard weighted method (Hadamard), which can be expressed as:

[0098] Fusion = W 1 ⊙O 1 +W 2 ⊙O 2 +W 3 ⊙O 3 +W 4 ⊙O 4

[0099] Among them, Fusion is the fused feature data, O 1 is the passenger flow feature data of the historical time series, W 1 is the weight vector of the passenger flow feature data of the historical time series, O 2 is the station network topology feature data, W 2 is the weight vector of the station network topology data, O 3 is the weather feature data, W 3 is the weight vector of the weather feature data, O 4 is the traffic congestion feature data, W 4 is the weight vector of the traffic congestion data, ⊙ is the Hadamard product, and its dimension is the same as the dimension of each branch output vector, which is randomly initialized and its weights are adjusted through backpropagation.

[0100] S3. Construct a passenger flow prediction model based on the fused feature data, input the fused feature data into the passenger flow prediction model, and obtain the short-term passenger flow prediction result;

[0101] Specifically, the weighted features after multi-source data fusion are further used to capture the temporal features between features through an attention long short-term memory network. Its output is flattened and input into a fully connected layer containing 243 neurons for the output of the final prediction result, which is expressed as:

[0102] [Y t+1 , …, Y t+k = f(P t , P t ′, G, Wt )

[0103] Among them, P t represents the inbound passenger flow time series matrix at time step t, and P t ' represents the outbound passenger flow time series matrix at time step t. G represents the adjacency matrix of the station topological structure, and W t represents the weather information and traffic congestion index feature matrix. [Y t+1 , …, Y t+k represents the short-term passenger flow prediction values for k time steps from Y t+1 to Y t+k k. f represents the passenger flow prediction model.

[0104] Furthermore, the passenger flow prediction model is provided with a multi-scale residual feature extraction module, a residual graph convolutional neural network module, and an attention time series module, where:

[0105] The multi-scale residual feature extraction module is used to effectively extract the multi-scale characteristics in the time series, and at the same time improve the depth and stability of feature extraction through the residual structure to obtain the spatio-temporal feature map of the passenger flow;

[0106] The residual graph convolutional neural network module is used to depict the topological dependence relationship between stations in the rail network, and use the residual to weaken the deficiencies existing in the graph convolutional neural network to obtain the network topological feature map;

[0107] The attention time series module uses the attention long short-term neural network layer to fuse the weather data and traffic congestion data features, and performs feature weighted fusion on the spatio-temporal feature map, the network feature map, and the weather and traffic congestion index time feature information processed by two layers of long short-term memory networks. The features after fusion further capture the time features between features through the attention long short-term neural network layer. The multi-scale convolutional model framework is as Figure 4 shown.

[0108] Furthermore, the multi-scale residual feature extraction module processes the inbound flow graph and the outbound flow graph respectively to extract the multi-scale time and space features of the passenger flow. The multi-scale residual feature extraction module includes 1*1 convolution kernels, 3*3 convolution kernels, 5*5 convolution kernels, and 7*7 convolution kernels, where:

[0109] The 1x1 convolution kernel is used to perform information weighted combination on the channel dimension of the input feature map, so as to realize the reduction or increase of the channels;

[0110] The 3x3 convolution kernel is used to capture the rapid changes in the data within the local spatio-temporal range, and is suitable for capturing the passenger flow fluctuations on short time scales, such as the passenger flow changes within 15 minutes to 1 hour;

[0111] The 5x5 convolution kernel is used to extract data trends over a relatively long time span and can identify periodic or trend changes, such as long-term patterns during morning and evening rush hours or seasonal fluctuations.

[0112] The 7x7 convolution kernel is used to capture spatio-temporal patterns spanning multiple time steps and can identify long-term dependency relationships, such as the passenger flow change trends over multiple hours or a full working day.

[0113] Furthermore, prior art indicates that appropriately increasing the number of network layers can extract rich network features. However, due to problems such as vanishing gradients and exploding gradients during model training, the deeper the network layers are not necessarily the better. Therefore, the present invention adds residual connections between layers to largely solve problems such as vanishing gradients or exploding gradients, and can directly retain the input information to ensure that important basic features are not lost during model training. The multi-scale residual feature extraction module is as Figure 5 shown.

[0114] S4. By using the mean square error (MSE) as the loss function and using the root mean square error (RMSE), mean absolute error (MAE), and weighted mean absolute percentage error (WMAPE) as evaluation metrics for the model prediction effect, the calculation formulas can be expressed as:

[0115]

[0116] where Xi is the true value, n is the total number of samples, and i is the i-th sample.

[0117] Embodiment 2

[0118] As Figure 6 shown, based on Embodiment 1, this embodiment proposes a terminal device for a rail transit short-term passenger flow prediction method based on multi-scale convolution. The terminal device 200 includes at least one memory 210, at least one processor 220, and a bus 230 connecting different platform systems.

[0119] The memory 210 may include a readable medium in the form of volatile memory, such as RAM 211 and / or cache memory 212, and may further include ROM 213.

[0120] Among them, the memory 210 also stores a computer program, which can be executed by the processor 220, so that the processor 220 executes any one of the above-mentioned short-term passenger flow prediction methods for rail transit based on multi-scale convolution. The specific implementation manner is consistent with the implementation manners and the achieved technical effects described in the embodiments of the above method, and some contents will not be repeated. The memory 210 may also include a program / utilities 214 having a set (at least one) of program modules 215. Such program modules include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment.

[0121] Correspondingly, the processor 220 can execute the above computer program and can also execute the program / utilities 214.

[0122] The bus 230 can represent one or more of several types of bus structures, including a memory bus or a memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any bus structure in a variety of bus structures.

[0123] The terminal device 200 can also communicate with one or more external devices 240, such as a keyboard, a pointing device, a Bluetooth device, etc., and can also communicate with one or more devices capable of interacting with the terminal device 200, and / or communicate with any device (such as a router, a modem, etc.) that enables the terminal device 200 to communicate with one or more other computing devices. Such communication can be carried out through the I / O interface 250. And, the terminal device 200 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through the network adapter 260. The network adapter 260 can communicate with other modules of the terminal device 200 through the bus 230. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in combination with the terminal device 200, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms, etc.

[0124] Embodiment 3

[0125] As Figure 7 shown, on the basis of Embodiment 1, this embodiment proposes a computer-readable storage medium for a short-term passenger flow prediction method for rail transit based on multi-scale convolution. An instruction is stored on the computer-readable storage medium, and when the instruction is executed by a processor, it implements any one of the above-mentioned short-term passenger flow prediction methods for rail transit based on multi-scale convolution. The specific implementation manner is consistent with the implementation manners and the achieved technical effects described in the embodiments of the above method, and some contents will not be repeated.

[0126] Figure 7 Shown is a program product 300 provided by this embodiment for implementing the above method. It can be a portable compact disc read-only memory (CD-ROM), include program code, and can run on a terminal device, such as a personal computer. However, the program product 300 of the present invention is not limited thereto. In this embodiment, the readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in conjunction with an instruction execution system, apparatus, or device. The program product 300 can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection with one or more wires, a portable 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0127] The computer-readable storage medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries the readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium can also be any readable medium other than the readable storage medium, and this readable medium can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted by any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination of the above. The program code for performing the operations of the present invention can be written in any combination of one or more programming languages. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, executed as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, by using an Internet service provider to connect through the Internet).

[0128] The present invention is described from the viewpoints of purpose of use, efficacy, progressiveness, and novelty. It has practical progressiveness and meets the functional enhancement and use requirements emphasized by the patent law. The above description and drawings of this application are only preferred embodiments of this application and do not limit this application. Therefore, all those similar or identical to the structure, device, features, etc. of this application, that is, all equivalent substitutions or modifications made according to the scope of the patent application of this application, shall fall within the scope of patent application protection of this application.

[0129] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above is only the specific embodiment of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A rail transit short-term passenger flow prediction method based on multi-scale convolution, characterized in that: The following steps are involved: S1. Obtain historical time series passenger flow data, site network topology data, and external environment data, and preprocess them separately to obtain a multi-source data set; S2. Extract features from the historical time series passenger flow data, site network topology data and external environment data in the multi-source data set, obtain feature data of each data source, and perform feature weighted fusion on the feature data by using the Hadamard weighting method to obtain fused feature data; S3. Build a passenger flow prediction model based on the fused feature data, input the fused feature data into the passenger flow prediction model, and obtain a short-term passenger flow prediction result; S4. The mean square error function is used as the loss function, and the short-term passenger flow prediction results are evaluated by the root mean square error, mean absolute error and weighted mean absolute percentage error until the function converges to obtain the final short-term passenger flow prediction results.

2. According to claim 1, a method for predicting short-term passenger flow in rail transit based on multi-scale convolution is characterized in that: The historical time series passenger flow data includes the historical time step outbound passenger flow tensor, the weekly cycle inbound and outbound passenger flow tensor and the daily cycle inbound and outbound passenger flow tensor; The weekly inbound and outbound passenger flow tensor can be expressed as: in, is the weekly passenger flow tensor for inbound and outbound stations, T is the current time slice, w is the current week, TS is the time step, and TS W is the passenger flow data in the same time period last week, and N is the number of stations; The daily cycle inbound and outbound passenger flow tensor can be expressed as: in, is the daily period in-and-out passenger flow tensor, d is the current day, TS d is the passenger flow with a time step of TS in the same time period of the previous day; The weekly inbound and outbound passenger flow tensor, the daily inbound and outbound passenger flow tensor and the historical time step inbound and outbound passenger flow tensor are concatenated to obtain the inbound and outbound passenger flow cycle tensor, which can be expressed as: in, is the passenger flow cycle tensor of inbound and outbound stations.

3. The method for predicting short-term passenger flow of rail transit based on multi-scale convolution according to claim 1 is characterized in that: The feature extraction of the site network topology data in step 2 includes the following steps: S201-A. Obtain urban rail transit network data and represent it in a graph structure as follows: G=(N,E,A) A∈R N*N Among them, G is the urban rail transit network data, N is the number of stations, E is the number of edges between stations, and A is the adjacency matrix that represents whether the stations are adjacent; S201-B. Obtain the historical passenger flow feature matrix, which can be expressed as: F=(X t ,X t -1,X t -2,…,X t -m+1) Among them, F is the feature matrix, X t is the historical passenger flow at time t, and m is the time interval; S201-C. Construct a GCN convolution function and calculate the topological relationship between stations in the urban rail transit network data. The GCN convolution function can be expressed as: H∈R N*t ,H′∈R N*t Where I is the identity matrix, is the sum of the adjacency matrix A and the identity matrix I, for The diagonal node degree matrix of , l is the number of convolutional layers, W is the parameter matrix of the lth layer, H is the feature matrix, N is the number of sites, t is the number of features of each node, H ’ is the feature matrix containing topological information, σ is the activation function; S201-D. Based on the residual function, the GCN convolution function is transformed to construct a residual graph convolution network, and the site network topology data features are extracted. The residual convolution network can be expressed as: ln∈R N*M in, is the symmetric normalized Laplace operator, ln is the model input, N is the number of stations, M is the number of historical time steps used by each station, ln ’ is the output feature matrix after symmetric normalized Laplace transform.

4. The method for predicting short-term passenger flow of rail transit based on multi-scale convolution according to claim 1 is characterized in that: The external environment data includes weather data and traffic congestion data; The weather data at least includes temperature data, wind speed data, air pressure data, humidity data and rainfall data; The traffic congestion data at least includes the congestion index, the length of the congested road section and the average vehicle speed; The congestion index is the ratio of the actual formation time to the smooth travel time, and the larger the ratio, the higher the congestion level; The external environment data can be expressed in a matrix as follows: Among them, W is the external environment data, x is the characteristic value of the external environment data, t is the current time, and l is the length of the time window.

5. The method for predicting short-term passenger flow of rail transit based on multi-scale convolution according to claim 4 is characterized in that: Step S2 performs feature fusion on the feature data of each data source by using the Hadamard weighting method to obtain fused feature data, which can be expressed as: Fusion=W1⊙O1+W2⊙O2+W3⊙O3+W4⊙O4 Among them, Fusion is the fusion feature data, O1 is the historical time series passenger flow feature data, W1 is the weight vector of the historical time series passenger flow feature data, O2 is the site network topology feature data, W2 is the site network topology data weight vector, O3 is the weather feature data, W3 is the weather feature data weight vector, O4 is the traffic congestion feature data, W4 is the traffic congestion data weight vector, and ⊙ is the Hadamard product.

6. The method for predicting short-term passenger flow of rail transit based on multi-scale convolution according to claim 1, characterized in that: The passenger flow prediction model can be expressed as: [Y t+1 ,…,Y t+k ]=f(P t ,P′ t ,G,W t ) Among them, P t represents the time series matrix of passenger flow entering the station at time step t, P′ t represents the outbound passenger flow time series matrix at time step t, G represents the adjacency matrix of the station topology structure, and W t represents weather information and traffic congestion index feature matrix, [Y t+1 ,…,Y t+k ] indicates that from Y t+1 To Y t+k The short-term passenger flow prediction value of k time steps, f represents the passenger flow prediction model.

7. The method for predicting short-term passenger flow of rail transit based on multi-scale convolution according to claim 1, characterized in that: The passenger flow prediction model includes a multi-scale residual feature extraction module, a residual graph convolutional neural network module, and an attention timing module; the multi-scale residual feature extraction module is used to extract multi-scale characteristics in historical time series passenger flow data to obtain a spatiotemporal feature graph of the passenger flow; the residual graph convolutional neural network module is used to extract the topological dependency of the site network topology data to obtain a network topology feature graph; the attention timing module is used to fuse the features of the external environment data through the attention long-short time neural network layer, and perform time feature extraction on the fused feature data to obtain the time features of the fused feature data.

8. The method for predicting short-term passenger flow of rail transit based on multi-scale convolution according to claim 7, characterized in that: The multi-scale residual feature extraction module includes a 1*1 convolution kernel, a 3*3 convolution kernel, a 5*5 convolution kernel and a 7*7 convolution kernel.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of a method for predicting short-term passenger flow of rail transit based on multi-scale convolution as described in any one of claims 1 to 8 are implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for predicting short-term passenger flow in rail transit based on multi-scale convolution as described in any one of claims 1 to 8 are implemented.

Images