Intercity travel OD demand prediction model training method, prediction method and system

Abstract

The present invention provides a training method, prediction method, and system for an intercity travel OD demand prediction model, belonging to the technical field of transportation travel demand prediction. The method establishes an intercity travel database for urban agglomerations based on mobile phone signaling data, weather data, and point of interest (POI) data, analyzes the spatiotemporal characteristics of intercity travel, analyzes the factors influencing intercity travel OD demand, uses a random forest to select highly important variables as input variables for the model, treats each OD pair as a node, models the heterogeneous spatial association relationships between different OD pairs, constructs a multi-graph adjacency matrix consisting of a round-trip relationship graph, a distance relationship graph, and a functional similarity graph, and establishes a spatiotemporal multi-graph convolutional neural network model based on an attention mechanism for intercity travel OD demand prediction. By analyzing the spatiotemporal characteristics of intercity travel and predicting OD demand, the present invention grasps the movement patterns of intercity travel passengers, thereby assisting transportation management departments in pre-perceiving traffic conditions. This model has important practical significance for ensuring stable and efficient operation of regional transportation.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic travel demand prediction, and in particular to a training method, a prediction method and a system for an inter-city travel OD demand prediction model. Background Art

[0002] Intercity travel is becoming an increasingly important mode of transportation. Given the wide temporal and spatial spans, diverse travel purposes, and high levels of heterogeneity in intercity travel, intercity travel demand forecasting can help understand the movement patterns of intercity travelers, thereby assisting traffic management departments in proactively assessing traffic conditions. This has significant practical significance for ensuring stable and efficient regional transportation.

[0003] At present, most of the existing intercity travel research results are based on survey data, AFC, IC, GPS and other incomplete data sets, which cannot accurately describe the intercity travel volume and travel mode, and the credibility of the results is low; the existing intercity travel demand forecasting research has few results on OD demand forecasting between large-scale regions, the spatiotemporal granularity is not fine enough, the characteristic variables in the forecasting model are relatively simple, and the spatiotemporal correlation of different OD demands cannot be fully captured, and the forecast accuracy needs to be improved.

[0004] Mobile phone signaling data, characterized by large data volumes, wide user coverage, high sampling frequency, rapid updates, and diverse attributes, can reflect travel location information across multiple modes of transportation at the scale of user urban clusters and urgently needs to be scientifically, fully, and efficiently mined and utilized. In recent years, artificial intelligence technologies, particularly deep learning, have been widely applied, supported by big data resources and cloud computing. Deep learning can effectively mine the spatiotemporal information implicit in traffic big data, providing new insights for regional and urban intelligent transportation management and smart travel services. Summary of the Invention

[0005] The purpose of the present invention is to provide a training method, a prediction method and a system for an inter-city travel OD demand prediction model to solve at least one technical problem existing in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In one aspect, the present invention provides a method for training an intercity travel OD demand prediction model, comprising:

[0008] Acquire training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix, and OD demand quantities under different date types and weather characteristics within multiple historical time intervals;

[0009] Based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, and OD multi-graph adjacency matrix as input, and based on the OD demand as output, a travel OD demand prediction model is trained; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output layer;

[0010] The input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the functional similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependency relationship of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependency relationship of inter-city travel OD demand; the attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand forecast value.

[0011] Furthermore, the construction of the adjacency matrix of the round-trip relationship graph includes: if two OD pairs are mutually round-trip OD pairs, the intercity travel volume between the mutually round-trip OD pairs is similar, the demand volume of the two OD pairs is similar and there is a tidal phenomenon in opposite directions, indicating whether the intercity travel OD pairs within the two urban agglomerations are mutually round-trip OD pairs as shown in the formula:

[0012]

[0013] Furthermore, the distance relationship graph connection matrix is ​​constructed, including: in inter-city travel, the closer the distance between the starting point and the end point of two OD pairs, the stronger the correlation between the demand of the two OD pairs. Therefore, the reciprocal of the distance is used to represent the weight between the two OD pairs. First, the reciprocal of the distance between the O and the reciprocal of the distance between the D of the two OD pairs are calculated to construct the OD distance relationship graph. The calculation formulas are:

[0014]

[0015]

[0016] In order to make the weight of the two OD pairs with similar starting and ending distances larger, the distance relationship graph weight of the two OD pairs is taken as the minimum value of the inverse of the distance between O and the inverse of the distance between D, that is, the distance matrix weight = min{inverse of the distance between O, inverse of the distance between D}, and the calculation formula is Finally, a distance relationship diagram is obtained that can describe the distance between the starting point and the end point of two OD pairs.

[0017] Furthermore, the construction of the adjacency matrix of the functional similarity graph includes: using the correlation of the POI vectors of the two regions to represent the functional similarity of the two OD pairs, where the dimension of the POI vector is the number of POI categories and the value is the number of POIs in each category; using the cosine similarity method to measure the similarity of the two POI vectors. The smaller the angle between the two vectors, the larger the cosine value, and the greater the similarity.

[0018] Furthermore, the functional similarity between O and D of two OD pairs is calculated to construct the OD distance relationship graph. The calculation formula is:

[0019]

[0020]

[0021] In order to make the weight of the two OD pairs with similar functional relationships at the starting and end points larger, the weight of the functional similarity relationship graph of the two OD pairs takes the minimum value of the functional similarity between O and the functional similarity between D, that is, functional similarity weight = min{functional similarity between O, functional similarity between D}, and finally obtain a relationship graph that can describe the functional similarity between the starting and end points of the two OD pairs.

[0022] In a second aspect, the present invention provides an intercity travel OD demand prediction model training system, comprising:

[0023] An acquisition module is used to acquire training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix, and OD demand quantities under different date types and weather characteristics within multiple historical time intervals;

[0024] A training module for training a travel OD demand prediction model based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, and OD multi-graph adjacency matrix as input and the OD demand as output; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output layer;

[0025] The input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the functional similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependency relationship of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependency relationship of inter-city travel OD demand; the attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand forecast value.

[0026] In a third aspect, the present invention provides a method for predicting inter-city travel OD demand, comprising:

[0027] Obtain the date type and weather characteristic data of the time point to be predicted;

[0028] The acquired date type and weather characteristic data are input into a travel OD demand prediction model for processing to obtain travel OD demand; wherein the travel OD demand prediction model is trained by the model training method described above.

[0029] In a fourth aspect, the present invention provides an inter-city travel OD demand forecasting system, comprising:

[0030] An acquisition module is used to obtain the date type and weather characteristic data of the time point to be predicted;

[0031] The prediction module is used to input the acquired date type and weather characteristic data into the travel OD demand prediction model for processing to obtain the travel OD demand; wherein, the travel OD demand prediction model is trained by the model training method described above.

[0032] In a fifth aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the inter-city travel OD demand prediction method as described above is implemented.

[0033] In a sixth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed on one or more processors, is used to implement the inter-city travel OD demand prediction method as described above.

[0034] In the seventh aspect, the present invention provides an electronic device, comprising: a processor, a memory and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the intercity travel OD demand prediction method as described above.

[0035] The beneficial effects of the present invention are as follows: it takes into account the impact of the spatial structure of urban agglomerations and urban attributes on intercity travel demand, deepens the understanding of each city in the urban agglomeration, provides strong support for the refined management of intercity travel in the urban agglomeration, and has good prediction effect and high reliability.

[0036] Additional advantages of the present invention will be more clearly given in the following description or learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is a flow chart of the method for predicting short-term OD demand in urban agglomerations based on deep learning according to an embodiment of the present invention.

[0039] Figure 2 This is a schematic diagram of the OD-ST-AMGCN model structure described in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention.

[0041] Those skilled in the art will understand that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.

[0042] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0043] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0044] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.

[0045] Example 1

[0046] In this embodiment 1, a training system for an intercity travel OD demand prediction model is first provided, comprising: an acquisition module for acquiring training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, an OD multi-graph adjacency matrix, and OD demand quantities under different date types and weather characteristics within multiple historical time intervals;

[0047] A training module for training a travel OD demand prediction model based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, and OD multi-graph adjacency matrix as input and the OD demand as output; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output layer;

[0048] The input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the functional similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependency relationship of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependency relationship of inter-city travel OD demand; the attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand forecast value.

[0049] In this embodiment 1, a model training method is implemented using the above-mentioned system to obtain an inter-city travel OD demand prediction model, and the training method includes: obtaining training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix within multiple historical time intervals, and OD demand under different date types and weather characteristics; based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix as input, and based on OD demand as output, the travel OD demand prediction model is trained; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output Layer; wherein, the input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the function similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependency relationship of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependency relationship of inter-city travel OD demand; the attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand forecast value.

[0050] The construction of the adjacency matrix of the round-trip relationship graph includes: if two OD pairs are mutually round-trip OD pairs, the intercity travel volume between the mutually round-trip OD pairs is similar, the demand volume of the two OD pairs is similar and there is a tidal phenomenon in opposite directions, indicating whether the intercity travel OD pairs within the two urban agglomerations are mutually round-trip OD pairs as shown in the formula:

[0051]

[0052] The construction of the distance relationship diagram connection matrix includes: in inter-city travel, the closer the distance between the starting point and the end point of two OD pairs, the stronger the correlation between the two OD pairs' demand. Therefore, the reciprocal of the distance is used to represent the weight between the two OD pairs. First, the reciprocal of the distance between the O and the reciprocal of the distance between the D of the two OD pairs are calculated to construct the OD distance relationship diagram. The calculation formulas are:

[0053]

[0054]

[0055] In order to make the weight of the two OD pairs with similar starting and ending distances larger, the distance relationship graph weight of the two OD pairs is taken as the minimum value of the inverse of the distance between O and the inverse of the distance between D, that is, the distance matrix weight = min{inverse of the distance between O, inverse of the distance between D}, and the calculation formula is Finally, a distance relationship diagram is obtained that can describe the distance between the starting point and the end point of two OD pairs.

[0056] The construction of the adjacency matrix of the functional similarity graph includes: using the correlation of the POI vectors of the two regions to represent the functional similarity of the two OD pairs, where the dimension of the POI vector is the number of POI categories and the value is the number of POIs in each category; using the cosine similarity method to measure the similarity of the two POI vectors. The smaller the angle between the two vectors, the larger the cosine value, and the greater the similarity.

[0057] Calculate the functional similarity between O and D of two OD pairs to construct the OD distance relationship graph The calculation formula is:

[0058]

[0059]

[0060] In order to make the weight of the two OD pairs with similar functional relationships at the starting and end points larger, the weight of the functional similarity relationship graph of the two OD pairs takes the minimum value of the functional similarity between O and the functional similarity between D, that is, functional similarity weight = min{functional similarity between O, functional similarity between D}, and finally obtain a relationship graph that can describe the functional similarity between the starting and end points of the two OD pairs.

[0061] In this embodiment 1, OD multi-graph adjacency matrix fusion includes: A is fused based on the parameter matrix c ,A d ,A f The fusion is done by combining different images by weighted sum of element-wise products. W1', W2', and W3' are adjustable and can be used to control the importance weights of different images. In order to standardize these parameters, the parameter matrix is ​​normalized using the softmax function. Since the values ​​of different images can vary greatly, each image is first normalized, and then the weight matrix is ​​weighted summed by element-wise product with the image to obtain the final fusion image A. fusion , the calculation formula is W1',W2',W3'=softmax(W1,W2,W3), and

[0062] Example 2

[0063] In this second embodiment, a system for predicting intercity travel OD demand is provided, comprising: an acquisition module for acquiring date type and weather characteristic data at a time point to be predicted; and a prediction module for inputting the acquired date type and weather characteristic data into a travel OD demand prediction model for processing to obtain travel OD demand. The system implements travel OD demand prediction, comprising: acquiring date type and weather characteristic data at a time point to be predicted; and inputting the acquired date type and weather characteristic data into the travel OD demand prediction model for processing to obtain travel OD demand.

[0064] In this embodiment 2, the training method of the intercity travel OD demand prediction model includes: obtaining training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix within multiple historical time intervals, and OD demand under different date types and weather characteristics;

[0065] Based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, and OD multi-graph adjacency matrix as input, and based on the OD demand as output, a travel OD demand prediction model is trained; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output layer;

[0066] The input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the functional similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependency relationship of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependency relationship of inter-city travel OD demand; the attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand forecast value.

[0067] The construction of the adjacency matrix of the round-trip relationship graph includes: if two OD pairs are mutually round-trip OD pairs, the inter-city travel volume between the mutually round-trip OD pairs is similar, the demand volume of the two OD pairs is similar and there is a tidal phenomenon in opposite directions, indicating whether the inter-city travel OD pairs within the two urban agglomerations are mutually round-trip OD pairs as shown in the formula:

[0068]

[0069] The construction of the distance relationship diagram connection matrix includes: in inter-city travel, the closer the distance between the starting point and the end point of two OD pairs, the stronger the correlation between the two OD pairs' demand. Therefore, the reciprocal of the distance is used to represent the weight between the two OD pairs. First, the reciprocal of the distance between the O and the reciprocal of the distance between the D of the two OD pairs are calculated to construct the OD distance relationship diagram. The calculation formulas are:

[0070]

[0071]

[0072] In order to make the weight of the two OD pairs with similar starting and ending distances larger, the distance relationship graph weight of the two OD pairs is taken as the minimum value of the inverse of the distance between O and the inverse of the distance between D, that is, the distance matrix weight = min{inverse of the distance between O, inverse of the distance between D}, and the calculation formula is Finally, a distance relationship diagram is obtained that can describe the distance between the starting point and the end point of two OD pairs.

[0073] The construction of the adjacency matrix of the functional similarity graph includes: using the correlation of the POI vectors of the two regions to represent the functional similarity of the two OD pairs, where the dimension of the POI vector is the number of POI categories and the value is the number of POIs in each category; using the cosine similarity method to measure the similarity of the two POI vectors. The smaller the angle between the two vectors, the larger the cosine value, and the greater the similarity.

[0074] Calculate the functional similarity between O and D of two OD pairs to construct the OD distance relationship graph The calculation formula is:

[0075]

[0076]

[0077] In order to make the weight of the two OD pairs with similar functional relationships at the starting and end points larger, the weight of the functional similarity relationship graph of the two OD pairs takes the minimum value of the functional similarity between O and the functional similarity between D, that is, functional similarity weight = min{functional similarity between O, functional similarity between D}, and finally obtain a relationship graph that can describe the functional similarity between the starting and end points of the two OD pairs.

[0078] Example 3

[0079] In this embodiment 3, a method for predicting short-term OD demand of urban agglomerations based on deep learning is provided, in which the problem of predicting short-term OD demand of intercity travel in urban agglomerations is defined, an intercity travel database of urban agglomerations is constructed, and the spatiotemporal characteristics of intercity travel in urban agglomerations are analyzed; input variables for travel demand prediction are selected; an OD multi-graph adjacency matrix is ​​constructed; and an OD demand prediction model for intercity travel in urban agglomerations is constructed.

[0080] In this embodiment, the problem definition for predicting short-term OD demand for intercity travel in urban agglomerations is as follows: the research object is the OD demand for intercity travel between different administrative regions. OD pairs are used as research nodes, and the research area is divided into n subspaces. Travel between subspaces of different cities forms a set V of intercity travel OD pairs. The OD spatial network relationship graph is an important input data for capturing spatial dependencies in the demand forecasting model. Its relationship is an undirected weighted graph G = (V, E, A), where nodes are OD pairs, V is the set of all intercity travel OD pairs, edges are the relationships between OD pairs, E is the set of all edges, and edge weights represent the strength of the correlation between OD pairs, ensuring that the spatial network graph is fully connected. A is a graph matrix describing the size of edge weights.

[0081] In this embodiment, the construction of the intercity travel database for urban agglomerations involves preprocessing and integrating mobile phone signaling data, POI data, and weather data. This requires multi-source data matching and data preprocessing at varying levels of accuracy across administrative regions.

[0082] The input variable types constructed for the demand forecast feature set are selected from the following perspectives: in terms of time distribution characteristics, the intercity travel distribution patterns and hourly distribution of different date types are analyzed; in terms of spatial distribution characteristics, the inflow and outflow of intercity travel on different date types and the OD spatial distribution patterns are analyzed.

[0083] The characteristic variable importance analysis and input variable selection process include: The temporal characteristics of intercity OD volume indicate that intercity OD volume exhibits similar daily and weekly variations. The factors influencing historical OD volume can be analyzed from three perspectives: adjacent time slices, trending time slices, and weekly time slices. Using a random forest algorithm to calculate the importance of each feature, the OD demand volume for appropriate time slices was selected as the historical time slice characteristic variables. These, along with weather type and date type variables, serve as input variables for the OD demand forecasting model.

[0084] The construction of the adjacency matrix includes the construction of the adjacency matrix of the round-trip relationship graph, the construction of the adjacency matrix of the distance relationship graph and the construction of the adjacency matrix of the functional similarity graph.

[0085] The construction of the round-trip relationship graph connection matrix includes: if two OD pairs are mutually round-trip OD pairs, and the intercity travel volume between the mutually round-trip OD pairs is similar, then the demand of the two OD pairs is similar. The construction is to indicate whether the two OD pairs are mutually round-trip OD pairs.

[0086] The construction of the distance relationship graph connection matrix includes: in inter-city travel, the closer the distance between the starting point and the end point of two OD pairs, the stronger the correlation between the demand of the two OD pairs. Therefore, the reciprocal of the distance is used to represent the weight between the two OD pairs. First, the reciprocal of the distance between O and the reciprocal of the distance between D of the two OD pairs are calculated respectively to construct the OD distance relationship graph. In order to make the weight of the two OD pairs with similar starting point and end point distances greater, the weight of the distance relationship graph of the two OD pairs takes the minimum value of the reciprocal of the distance between O and the reciprocal of the distance between D, that is, the distance matrix weight = min{reciprocal of the distance between O, reciprocal of the distance between D}.

[0087] The construction of the adjacency matrix of the functional similarity graph includes the following steps: From a semantic point of view, the demand for two locations with similar functional structures tends to have a strong relationship. This embodiment uses the cosine similarity method to measure the similarity of two POI vectors. The smaller the angle between the two vectors, the larger the cosine value, and the greater the similarity. First, the functional similarity between O and D of the two OD pairs is calculated to construct the OD distance relationship graph. Then, in order to make the weight of the functional relationship between the starting point and the end point of the two OD pairs similar, the weight of the functional similarity relationship graph of the two OD pairs is the minimum value of the functional similarity between O and the functional similarity between D, that is, the functional similarity weight = min{functional similarity between O, functional similarity between D}. Finally, a relationship graph that can describe the functional similarity between the starting point and the end point of the two OD pairs is obtained.

[0088] The three multi-graph adjacency matrices are then fused, including: A parameter matrix-based fusion method fuses the graphs into a single overall graph, combining the different graphs by performing a weighted sum of element-wise multiplications. Each graph is first normalized, and then a weighted sum of element-wise multiplications of the weight matrix and the graph is performed to obtain the final fused graph.

[0089] In this embodiment, after the above-mentioned basic work is completed, a short-time OD demand prediction model for intercity travel in an urban agglomeration is trained, including: a spatiotemporal multi-graph convolutional neural network model proposed for the problem of intercity travel OD demand prediction, hereinafter referred to as the OD-ST-AMGCN model, and the model structure consists of five parts: input layer, spatial convolution layer, temporal recurrent layer, attention mechanism module, and output layer. The spatial convolution layer is composed of a graph convolutional neural network GCN, which is an effective method for capturing spatial correlation. The present invention uses a graph convolutional neural network GCN to capture the spatial dependencies of different OD pairs. The temporal recurrent layer is composed of two layers of superimposed gated recurrent neural networks GRU, which are used to capture the temporal dependencies in feature variables. The attention mechanism module is composed of a soft attention mechanism model (Soft-Attention), which can select important information from a large amount of information for integrating and capturing global traffic status information.

[0090] In this embodiment 3, the inter-city travel OD demand between districts and counties is taken as the research object, and the short-term OD demand prediction method of urban agglomerations based on deep learning includes: unlike regional traffic prediction, a region is used as a node, OD pairs are used as research nodes, and the research area is divided into n subspaces. The travel between subspaces of different cities forms an inter-city travel OD pair set V. represents the inter-city travel OD demand of all OD pairs at the tth time interval on the dth day, In order to predict the inter-city travel OD demand of all OD pairs in the tth time interval on the dth day All OD demand at the historical time point before the tth time interval on the dth day and other related factor variables can be used as features to predict The OD spatial network relationship graph is an important input data for capturing spatial dependencies in demand forecasting models. The spatial network relationship between ODs is defined as an undirected weighted graph G = (V, E, A), where nodes are OD pairs, V is the set of all intercity OD pairs, edges are relationships between OD pairs, and E is the set of all edges. Edge weights represent the strength of the correlation between OD pairs, and the spatial network graph is fully connected. A is a graph matrix describing the edge weights.

[0091] The mobile phone signaling data, weather data, and POI data required for intercity travel in urban agglomerations are obtained, and multi-source data are fused and preprocessed to construct an intercity OD travel dataset for urban agglomerations.

[0092] Analyze the spatiotemporal characteristics of intercity travel. This primarily involves analyzing the daily and hourly distribution of intercity travel on holidays and non-holidays, the total volume of intercity travel and its inflow and outflow patterns, the travel distance distribution, and the spatial distribution of intercity travel OD patterns. Different intercity travel modes are identified based on geographic location and travel speed.

[0093] The influencing factors of intercity travel OD demand are analyzed, and random forest is used to select variables with high importance as input variables of the model. At the same time, weather factors and date types are considered to constitute the input variables of the intercity travel short-term OD demand prediction model.

[0094] In order to capture the spatial dependency between different OD pairs, each OD pair is regarded as a research node, and the heterogeneous spatial association relationship between different OD pairs is modeled. From the perspectives of geography and semantics, a round-trip relationship graph, a distance relationship graph, and a functional similarity graph are constructed to form a multi-graph adjacency matrix.

[0095] The OD-ST-AMGCN model, a spatiotemporal multi-graph convolutional neural network model, was proposed to predict intercity travel out-of-bounds (OD) demand. The model focuses on out-of-bounds (OD) pairs within district- and county-level administrative units within urban agglomerations. The model uses a GCN to capture spatial relationships based on adjacency matrices, including round-trip graphs, distance graphs, and functional similarity graphs. The GRU module captures the temporal dependencies of input feature variables, including historical data and weather conditions. Furthermore, an attention mechanism is used to select important information from a large amount of data, integrating and capturing global traffic status information. Ultimately, this allows for accurate prediction of OD demand.

[0096] The intercity OD travel prediction method for urban agglomerations provided in this embodiment has good prediction effect and high reliability. It takes into account the impact of the spatial structure and city attributes of the urban agglomeration on intercity travel demand, deepens the understanding of each city in the urban agglomeration, and provides strong support for the refined management of intercity travel in the urban agglomeration.

[0097] Example 4

[0098] In this implementation 4, a method for predicting short-term OD demand of urban agglomerations based on deep learning is provided, such as Figure 1 As shown, the method includes the following steps:

[0099] Step 1: Obtain multi-source data such as mobile phone signaling data, weather data, and POI data to establish an intercity travel database for urban agglomerations. At the same time, perform data preprocessing tasks including district-level information fusion, coordinate system conversion, and cleaning of unreasonable data.

[0100] Step 2: Analyze the spatiotemporal distribution characteristics of intercity O / D travel within the urban agglomeration. Regarding temporal distribution characteristics, the first step is to analyze the daily distribution of intercity travel volume and compare the differences in intercity travel volume distribution during holidays and non-holiday periods. The second step is to describe the hourly distribution of intercity travel volume, analyze the peak hours of intercity travel during holidays, compare the differences in peak travel hours during different holidays, and study the temporal impact of the holiday effect on intercity travel volume. Regarding spatial distribution characteristics, the main focus is to examine the differences in intercity travel inflows and outflows between different districts and counties during holidays and non-holiday periods to analyze the characteristics of intercity travel in different regions.

[0101] Step 3: Select input variables for demand forecasting and construct an adjacency matrix. Based on the conclusions from the spatiotemporal feature analysis, we preliminarily select the main spatiotemporal factors affecting intercity travel within the urban agglomeration. We then use a random forest algorithm to select factors that are most important for predicting the target travel (OD) demand. These factors are then input into the prediction model as feature variables. Furthermore, we model the spatial dependencies between all OD pairs from geographic and semantic perspectives, which serve as the adjacency matrix for the multi-graph convolutional neural network.

[0102] Step 4: Use the OD-ST-AMGCN model to predict intercity OD travel volume. By integrating the spatial attributes of the multi-graph fusion model and the attributes of the counties themselves, and using GRU to process temporal variables, the model uses an attention mechanism to allocate prediction weights, ultimately achieving intercity OD travel prediction for urban agglomerations.

[0103] Step 5: Analyze the model prediction results. Evaluate the model prediction effect through various model evaluation indicators and compare it with many benchmark models for time series prediction. On this basis, use ablation experiments to study the effects of different adjacency matrices to verify the effectiveness and reliability of this patented model.

[0104] In the selection of input variables for demand forecasting in step 3, it can be seen from the analysis of the time characteristics of inter-city travel OD volume that the inter-city travel OD volume has similar daily and weekly variation patterns. The factors affecting the historical time OD travel volume can be analyzed from three aspects: adjacent time slices, trend time slices, and weekly time slices. Inputting all the travel volumes of these time slices into the model will take up a lot of computing resources, so it is necessary to select time slices that are highly important to the model prediction effect, use random forest to calculate the importance of these time slices, and select time slices with high importance as the historical time OD travel volume characteristic variables of the model. After using random forest to calculate the importance of the OD demand volume of the historical time slices, the OD demand volume of the time slices with greater influence is selected as the historical time slice characteristic variable, together with the weather type and date type variables as the input variables of the OD demand forecasting model.

[0105] The construction of the adjacency matrix of the round-trip relationship graph in step 3 includes: if two OD pairs are mutually round-trip OD pairs, the intercity travel volume between the two OD pairs is similar, and the demand of the two OD pairs is similar and there may be a tidal phenomenon in opposite directions. It is constructed to indicate whether the intercity travel OD pairs within the two urban agglomerations are mutually round-trip OD pairs, as shown in the formula:

[0106]

[0107] The construction of the adjacency matrix of the distance relationship graph in step 3 includes: in inter-city travel, the closer the distance between the starting point and the end point of two OD pairs, the stronger the correlation between the demand of the two OD pairs. Therefore, the reciprocal of the distance is used to represent the weight between the two OD pairs. First, the reciprocal of the distance between the O and the reciprocal of the distance between the D of the two OD pairs are calculated to construct the OD distance relationship graph. The calculation formula is:

[0108]

[0109]

[0110] Then, in order to make the weight of the two OD pairs with similar starting and ending distances larger, the distance relationship graph weight of the two OD pairs is taken as the minimum value of the inverse of the distance between O and the inverse of the distance between D, that is, the distance matrix weight = min{inverse of the distance between O, inverse of the distance between D}, and the calculation formula is Finally, a distance relationship diagram is obtained that can describe the distance between the starting point and the end point of two OD pairs.

[0111] The construction of the adjacency matrix of the functional similarity graph in step 3 includes: From a semantic point of view, the demand for two locations with similar functional structures tends to have a strong relationship. Studies have shown that POI data is highly feasible for measuring the functionality of a city. POI data can reflect the functional attributes of a city. The correlation between the POI vectors of two regions is used to represent the functional similarity of two OD pairs, where the dimension of the POI vector is the number of POI categories and the value is the number of POIs in each category. The cosine similarity method is used to measure the similarity of two POI vectors. The smaller the angle between the two vectors, the larger the cosine value, and the greater the similarity. First, the functional similarity between the O and D of the two OD pairs is calculated to construct the OD distance relationship graph. The calculation formula is:

[0112]

[0113]

[0114] Then, in order to make the weight of the similar functional relationship between the starting point and the end point of the two OD pairs larger, the weight of the functional similarity relationship graph of the two OD pairs takes the minimum value of the functional similarity between O and the functional similarity between D, that is, functional similarity weight = min{functional similarity between O, functional similarity between D}, and finally obtains a relationship graph that can describe the functional similarity between the starting point and the end point of the two OD pairs.

[0115] In step 3, the OD multi-graph adjacency matrix is ​​fused. The fusion method based on the parameter matrix is ​​used to combine A c ,A d ,A fThe fusion is done by combining different images by weighted sum of element-wise products. W1', W2', and W3' are adjustable and can be used to control the importance weights of different images. In order to standardize these parameters, the parameter matrix is ​​normalized using the softmax function. Since the values ​​of different images can vary greatly, each image is first normalized, and then the weight matrix is ​​weighted summed by element-wise product with the image to obtain the final fusion image A. fusion , the calculation formula is:

[0116] W1',W2',W3'=softmax(W1,W2,W3)

[0117]

[0118]

[0119] The process of OD-ST-AMGCN model training to predict intercity travel OD volume in step 4 is as follows: Figure 2 As shown, it includes: Step 1: First input the historical time OD demand sequence characteristic variable X M×N , Date type variable D M×N , weather characteristic variables W M×N , OD multi-graph adjacency matrix A c ,A d ,A f , where M is the length of the historical time series, N is the number of OD pairs, and the input layer selects a sliding window of appropriate size according to the feature variables selected by the random forest in the previous article to process X M×N 、D M×N 、W M×N , we will get K segments of feature variable sequences, which are divided into training set and test set; at the same time, we use the parameter matrix fusion method to transform A c ,A d ,A f Merge into a general map A fusion Step 2: The spatial convolution layer uses the graph convolutional neural network GCN to fusion graph A fusion Convolution operations are performed to generate a spatial feature matrix to extract the spatial dependencies between different intercity travel O / D pairs. Step 3: The temporal recurrent layer uses a two-layer gated recurrent neural network (GRU) to process the feature variable sequence to obtain feature information to extract the temporal dependencies between intercity travel O / D demands. Step 4: The attention module uses an attention model to calculate the weight coefficients of the feature information at each time step, learning the important features from a large amount of feature information. Step 5: Finally, the output layer fuses the feature information from all time steps, and a fully connected layer is used to obtain the prediction results.

[0120] In this embodiment, the intercity travel short-time OD demand prediction model is a deep neural network with multiple hyperparameters, including learning rate, number of hidden layer neurons, etc. By comparing the model effect evaluation indicators, the hyperparameter values ​​that make the model effect optimal are selected. The learning rate is set to 0.001 by default, the number of hidden layer neurons is 10, and the batch size is 64.

[0121] In order to describe the prediction effect of the inter-city travel short-time OD demand prediction model (OD-ST-AMGCN model) proposed in this embodiment, the mean absolute error (MAE), root mean square error (RMSE), accuracy (ACC) and goodness of fit (R 2 ) is used as an evaluation indicator to evaluate the difference between the true value and the predicted value. The calculation formula is:

[0122]

[0123]

[0124]

[0125]

[0126] Where y i , Y represent the true values ​​respectively; Represent the predicted values ​​respectively.

[0127] The model of this embodiment was experimentally compared with five other existing models on four indicators on an intercity OD travel dataset of a certain regional urban agglomeration. The results are shown in Table 1.

[0128] Table 1 Comparison of evaluation indicators with baseline models

[0129]

[0130]

[0131] In Table 1: (1) OD-ST-AGCN-c: This model only uses the round-trip relationship graph for graph convolution operations, does not use the distance relationship graph, and does not use functional similarity. Other conditions remain unchanged. (2) OD-ST-AGCN-d: This model only uses the distance relationship graph for graph convolution operations, does not use the round-trip relationship graph, and does not use functional similarity. Other conditions remain unchanged. (3) OD-ST-AGCN-f: This model only uses functional similarity for graph convolution operations, does not use the round-trip relationship graph, and does not use the distance relationship graph. Other conditions remain unchanged. (4) OD-ST-AGCN-cd: This model uses the round-trip relationship graph and the distance relationship graph for graph convolution operations, does not use functional similarity, and other conditions remain unchanged. (5) OD-ST-AGCN-cf: This model uses the round-trip relationship graph and functional similarity for graph convolution operations, does not use the distance relationship graph, and other conditions remain unchanged. (6) OD-ST-AGCN-df: This model uses the distance relationship graph and functional similarity for graph convolution operations, does not use the round-trip relationship graph, and other conditions remain unchanged. (7) OD-ST-AMGCN: This is the OD demand prediction model trained in this embodiment.

[0132] The prediction effect evaluation indicators of the baseline model and the OD-ST-AMGCN model of this embodiment on the intercity travel OD dataset of the urban agglomeration in this region are shown in Table 1. It can be seen that among all the models, the evaluation indicators of the OD-ST-AMGCN model show the best model prediction effect, indicating that the feature variable selection and model construction of this embodiment are effective.

[0133] In this example, the OD-ST-AMGCN model proposed for intercity short-term OD demand prediction includes an OD multi-graph fusion module. The multi-graphs include a round-trip relationship graph, a distance relationship graph, and a functional similarity graph. Ablation experiments were conducted to analyze the contribution of these three types of spatial relationship graphs to the model's prediction performance. Table 2 shows the evaluation metrics for the model's prediction performance.

[0134] Table 2 Comparison of evaluation indicators of ablation experiments

[0135]

[0136]

[0137] As shown in Table 2, firstly, the OD-ST-AMGCN model proposed in this embodiment, which includes the fusion of multiple OD graphs, performs better than other variant models. This means that the round-trip relationship graph A c , distance relationship diagram A d , functional similarity A f The three OD spatial relationship graphs all contribute to the final prediction results of the model, indicating that the present invention is effective in extracting spatial dependencies by modeling three types of OD spatial graphs from geographical and semantic perspectives.

[0138] The deep learning-based short-term OD demand forecasting method for urban agglomerations provided in this embodiment is highly reliable, takes into account the impact of factors such as the spatial distribution of urban agglomerations, the land use characteristics of each district and county, and urban functions on intercity travel demand, and has higher calculation accuracy; the OD demand forecasting method described in this embodiment provides important theoretical support for the implementation of refined management of intercity travel in urban agglomerations.

[0139] Example 5

[0140] This embodiment 5 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the method for predicting short-term OD demand of urban agglomerations based on deep learning is implemented. The method includes:

[0141] Obtain the date type and weather characteristic data of the time point to be predicted;

[0142] The acquired date type and weather characteristic data are input into a travel OD demand prediction model for processing to obtain travel OD demand; wherein the travel OD demand prediction model training method includes:

[0143] Acquire training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix, and OD demand quantities under different date types and weather characteristics within multiple historical time intervals;

[0144] Based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, and OD multi-graph adjacency matrix as input, and based on the OD demand as output, a travel OD demand prediction model is trained; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output layer;

[0145] The input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the functional similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependency relationship of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependency relationship of inter-city travel OD demand; the attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand forecast value.

[0146] Example 6

[0147] This embodiment 6 provides a computer program product, including a computer program. When the computer program is executed on one or more processors, the computer program is used to implement the above-mentioned method for predicting short-term OD demand of urban agglomerations based on deep learning. The method includes:

[0148] Obtain the date type and weather characteristic data of the time point to be predicted;

[0149] The acquired date type and weather characteristic data are input into a travel OD demand prediction model for processing to obtain travel OD demand; wherein the travel OD demand prediction model training method includes:

[0150] Acquire training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix, and OD demand quantities under different date types and weather characteristics within multiple historical time intervals;

[0151] Based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, and OD multi-graph adjacency matrix as input, and based on the OD demand as output, a travel OD demand prediction model is trained; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output layer;

[0152] The input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the functional similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependency relationship of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependency relationship of inter-city travel OD demand; the attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand forecast value.

[0153] Example 7

[0154] This embodiment 7 provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the above-mentioned method for predicting short-term OD demand of urban agglomerations based on deep learning, and the method includes:

[0155] Obtain the date type and weather characteristic data of the time point to be predicted;

[0156] The acquired date type and weather characteristic data are input into a travel OD demand prediction model for processing to obtain travel OD demand; wherein the travel OD demand prediction model training method includes:

[0157] Acquire training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix, and OD demand quantities under different date types and weather characteristics within multiple historical time intervals;

[0158] Based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, and OD multi-graph adjacency matrix as input, and based on the OD demand as output, a travel OD demand prediction model is trained; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output layer;

[0159] The input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the functional similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependency relationship of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependency relationship of inter-city travel OD demand; the attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand forecast value.

[0160] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0161] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0162] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0163] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide the functions for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0164] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solutions disclosed in the present invention without the need for creative work should be included in the scope of protection of the present invention.

Claims

1. A method for training an intercity travel OD demand prediction model, characterized by: include: Acquire training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix, and OD demand quantities under different date types and weather characteristics within multiple historical time intervals; Based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, and OD multi-graph adjacency matrix as input, and based on the OD demand as output, a travel OD demand prediction model is trained; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output layer; The input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the functional similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependencies of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependencies of inter-city travel OD demands; The attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand prediction value.

2. The intercity travel OD demand prediction model training method according to claim 1 is characterized in that: The construction of the adjacency matrix of the round-trip relationship graph includes: if two OD pairs are mutually round-trip OD pairs, the intercity travel volume between the mutually round-trip OD pairs is similar, the demand volume of the two OD pairs is similar and there is a tidal phenomenon in opposite directions, indicating whether the intercity travel OD pairs within the two urban agglomerations are mutually round-trip OD pairs as shown in the formula:

3. The intercity travel OD demand prediction model training method according to claim 1 is characterized in that: The construction of the distance relationship diagram connection matrix includes: in inter-city travel, the closer the distance between the starting point and the end point of two OD pairs, the stronger the correlation between the two OD pairs' demand. Therefore, the reciprocal of the distance is used to represent the weight between the two OD pairs. First, the reciprocal of the distance between the O and the reciprocal of the distance between the D of the two OD pairs are calculated to construct the OD distance relationship diagram. The calculation formulas are: In order to make the weight of the two OD pairs with similar starting and ending distances larger, the distance relationship graph weight of the two OD pairs is taken as the minimum value of the inverse of the distance between O and the inverse of the distance between D, that is, the distance matrix weight = min{inverse of the distance between O, inverse of the distance between D}, and the calculation formula is Finally, a distance relationship diagram is obtained that can describe the distance between the starting point and the end point of two OD pairs.

4. The intercity travel OD demand prediction model training method according to claim 1 is characterized in that: The construction of the adjacency matrix of the functional similarity graph includes: using the correlation of the POI vectors of the two regions to represent the functional similarity of the two OD pairs, where the dimension of the POI vector is the number of POI categories and the value is the number of POIs in each category; using the cosine similarity method to measure the similarity of the two POI vectors. The smaller the angle between the two vectors, the larger the cosine value, and the greater the similarity.

5. The intercity travel OD demand prediction model training method according to claim 4 is characterized in that: Calculate the functional similarity between O and D of two OD pairs to construct the OD distance relationship graph The calculation formula is: In order to make the weight of the two OD pairs with similar functional relationships at the starting and end points larger, the weight of the functional similarity relationship graph of the two OD pairs takes the minimum value of the functional similarity between O and the functional similarity between D, that is, functional similarity weight = min{functional similarity between O, functional similarity between D}, and finally obtain a relationship graph that can describe the functional similarity between the starting and end points of the two OD pairs.

6. A training system for predicting intercity travel OD demand, characterized by: include: An acquisition module is used to acquire training data; the training data includes OD demand sequence characteristic variables, date type variables, weather characteristic variables, OD multi-graph adjacency matrix, and OD demand quantities under different date types and weather characteristics within multiple historical time intervals; A training module for training a travel OD demand prediction model based on the historical time OD demand sequence characteristic variables, date type variables, weather characteristic variables, and OD multi-graph adjacency matrix as input and the OD demand as output; the travel OD demand prediction model includes an input layer, a spatial convolution layer, a temporal recurrence layer, an attention layer, and an output layer; The input layer is used to select a sliding window of appropriate size to process the OD demand sequence feature variables, date type variables, and weather characteristic variables to obtain a multi-segment feature variable sequence, and fuse the round-trip relationship graph adjacency matrix, the distance relationship graph connection matrix, and the functional similarity graph adjacency matrix into a total graph matrix; the spatial convolution layer is used to perform a convolution operation on the total graph matrix to extract the spatial dependencies of different inter-city travel OD pairs; the temporal recurrence layer is used to process the feature variable sequence to extract the temporal dependencies of inter-city travel OD demands; The attention layer is used to calculate the weight coefficient of the feature information of each time step; the output layer is used to fuse the feature information of all time steps and use the fully connected layer to obtain the OD demand prediction value.

7. A method for predicting inter-city travel OD demand, characterized by: include: Obtain the date type and weather characteristic data of the time point to be predicted; The acquired date type and weather characteristic data are input into a travel OD demand prediction model for processing to obtain travel OD demand; wherein the travel OD demand prediction model is trained by the model training method according to any one of claims 1 to 5.

8. An inter-city travel OD demand forecasting system, characterized by: include: An acquisition module is used to obtain the date type and weather characteristic data of the time point to be predicted; The prediction module is used to input the acquired date type and weather characteristic data into the travel OD demand prediction model for processing to obtain the travel OD demand; wherein the travel OD demand prediction model is trained by the model training method described in any one of claims 1-5.

9. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by the processor, the inter-city travel OD demand prediction method according to claim 7 is implemented.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions for implementing the intercity travel OD demand prediction method as described in claim 7.

Images