Method, device, equipment and medium for predicting regional available transportation capacity by intercity carpooling OD

Abstract

The present invention provides a method, device, equipment and medium for predicting the available transport capacity of the OD for inter-city carpooling. The spatial characteristics are obtained by obtaining the road network information of the starting and ending cities and the data of points of interest to divide the starting and ending cities. The time characteristics are obtained by the historical demand characteristics of the starting and ending cities, the historical inflow characteristics of the region, the historical available transport capacity characteristics of the region, the historical vehicle characteristics of the region, and the demand time information. The training data set is input into the OD for regional available transport capacity prediction model for model training, and the test data set is input into the trained OD for regional available transport capacity prediction model to obtain the OD for regional available transport capacity prediction value. By predicting the available transport capacity of the OD for the region of the starting and ending cities in advance, the relevant transportation agencies can match the passenger orders with the available transport capacity in advance, reduce the waiting time of passengers and the idle mileage of drivers. The problem that the available transport capacity of vehicles in carpooling services is affected by the randomness of order generation, resulting in the uncertainty of the distribution of available transport capacity in the city.

Description

Technical Field

[0001] The present invention relates to the technical field of intercity passenger carpooling planning, and in particular to a method, device, equipment and medium for predicting regional available transport capacity for intercity carpooling OD. Background Art

[0002] In inter-city travel, the demand for passengers to choose carpooling is growing. This trend is mainly driven by factors such as transportation costs, travel efficiency and environmental awareness. Carpooling can not only effectively share travel expenses and reduce passengers' travel costs, but also improve overall travel efficiency by optimizing routes, reducing empty driving and vehicle use. Especially in the scenario of long-distance travel between cities, carpooling can alleviate traffic congestion, reduce carbon emissions, and conform to the trend of sustainable development. Therefore, more and more platforms have begun to incorporate carpooling services into inter-city travel plans to meet the diverse needs of passengers and improve the utilization rate of platform capacity. However, due to the uncertainty of passengers' departure and destination in time and space, this also puts higher requirements on the prediction of available capacity, making it more important to predict the capacity distribution of the region in advance and dynamically match passenger orders and capacity.

[0003] For the forecast of regional available capacity, we can know in advance where the capacity is located and the amount of available capacity in the region in the future time period. This forecast is crucial for the platform because it can help the platform match orders and available capacity in advance, thereby optimizing the scheduling and relocation of capacity. However, in the forecast of available capacity, historical capacity distribution data is usually affected by order distribution. Since the generation of orders has a certain degree of randomness and uncertainty, this increases the difficulty of forecasting. In addition, the complexity of spatiotemporal relationships further exacerbates this challenge. Over time, the spatiotemporal relationships between urban areas will change dynamically, and there is significant spatiotemporal heterogeneity. Therefore, relying solely on static regional relationship representations can no longer meet this dynamic and complex demand. The forecasting model must have the ability to capture dynamic changes in time and space to improve the accuracy of available capacity forecasts.

[0004] In terms of carpooling available capacity prediction, in order to improve the accuracy of regional available capacity prediction and the ability to handle uncertainty, it is necessary to organically combine order data with historical capacity data and fully consider their dynamic changes in time and space. By introducing the encoder-decoder architecture, the prediction model can further improve the prediction accuracy of carpooling available capacity. In this architecture, the encoder part focuses on learning hidden spatiotemporal information from historical sequence data, especially the potential correlation between historical order data and capacity distribution. The encoder uses a recurrent neural network to encode regional capacity and order data over a period of time into fixed-length vector representations that can capture the complex dependencies of spatiotemporal dynamic changes.

[0005] Next, the decoder is responsible for making future predictions based on the hidden states learned by the encoder. The decoder gradually generates the regional available capacity distribution at future moments, combined with the representation of the dynamic adjacency matrix, to predict the capacity flow and spatiotemporal evolution trends between different regions. This process can be performed at multiple time steps, thereby achieving multi-step predictions of available capacity at multiple moments in the future. The decoder does not simply continue the past capacity distribution trend, but makes more targeted and flexible predictions based on the deep information learned from the historical sequence, thereby more accurately reflecting the dynamic changes in future regional capacity.

[0006] Through this encoder-decoder architecture, the model can better handle the uncertainty issues in the intercity OD forecast of regional available capacity, improve the prediction accuracy, and provide strong support for the platform to conduct advance scheduling and capacity optimization in complex intercity carpooling scenarios.

[0007] In view of this, this application is filed. Summary of the invention

[0008] The present invention provides a method, device, equipment and medium for predicting regional available transport capacity by inter-city carpooling OD, which can at least partially improve the above-mentioned problems.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] A method for predicting regional available transport capacity based on inter-city carpooling OD, comprising:

[0011] Obtain the road network information of the starting and ending cities, the cities' point of interest data, historical order data, and historical transportation capacity data; obtain the historical demand characteristics and historical inflow characteristics of all areas in the historical time segment based on the city's point of interest data; obtain the available seat number characteristics and vehicle characteristics of all areas in the historical time segment based on the historical transportation capacity data; construct time information characteristics based on the historical order data and time information; construct a total data set; divide the total data set into a training data set and a test data set according to a preset ratio; divide the training data set into training samples according to a preset step size; and input the training samples into a DADGCRN model for training;

[0012] The road network information data of the starting and ending cities are eliminated and topologically processed to realize the regional division of the city, the point of interest data of the city is reclassified and regional POI identification is processed, each POI is weighted by the TF-IDF algorithm, and the POI similarity between regions is calculated using the cosine similarity formula to form a predefined adjacency matrix, and an adjacency relationship graph of the starting and ending cities is formed with a pair of OD regions as nodes and POI similarity as edge weights;

[0013] Input the training data set into a preset time relationship embedding module, and learn the time relationship between the historical order data and the historical transportation data by using a self-attention mechanism to generate a time relationship matrix;

[0014] The predefined adjacency matrix and the time relationship matrix are input into a preset adjacency matrix update module, and are spliced ​​in combination with static topological information and dynamic timing relationships. The spliced ​​matrix is ​​input into a sparse linear layer, and while maintaining computational efficiency, important relationships in the matrix are captured. The processed matrix is ​​dynamically updated using a gated recurrent unit to generate a dynamic adjacency matrix.

[0015] The dynamic adjacency matrix and the training data set are input into a preset dynamic graph convolution loop unit, the matrix penalty in the gated loop unit is replaced with a dynamic graph convolution module, and the dynamic graph convolution loop unit is used to capture the spatiotemporal correlation in the training data set. The hidden state of the last dynamic graph convolution loop unit is used as the final output to obtain the predicted value, and the error between the predicted value and the true value is calculated, and the model parameters with the minimum error are saved as the optimal model parameters. The model training is completed to obtain the intercity carpooling OD to available capacity prediction model;

[0016] The intercity carpooling OD is used to predict the available transport capacity prediction model for the test data set to obtain a prediction output.

[0017] The present invention also provides a device for predicting regional available transport capacity of inter-city carpooling OD, which comprises:

[0018] An input acquisition unit is used to acquire the road network information of the starting and ending cities, the point of interest data of the cities, the historical order data and the historical transportation capacity data, acquire the historical demand characteristics and the historical inflow characteristics of all areas in the historical time segment according to the point of interest data of the cities, acquire the available seat number characteristics and the vehicle characteristics of all areas in the historical time segment according to the historical transportation capacity data, construct the time information characteristics according to the historical order data and the time information, and construct a total data set, divide the total data set into a training data set and a test data set according to a preset ratio, divide the training data set into training samples according to a preset step size, and input the training samples into the DADGCRN model for training;

[0019] The adjacency relationship graph construction unit is used to remove and topologically process the road network information data of the starting and ending cities to realize the regional division of the city, reclassify the point of interest data of the city and process the regional POI identification, weight each POI through the TF-IDF algorithm, use the cosine similarity formula to calculate the POI similarity between regions to form a predefined adjacency matrix, and use a pair of OD regions as nodes and POI similarity as edge weights to form an adjacency relationship graph of the starting and ending cities;

[0020] A time relationship matrix generating unit, used for inputting the training data set into a preset time relationship embedding module, and generating a time relationship matrix by learning the time relationship between the historical order data and the historical transportation data by using a self-attention mechanism;

[0021] A dynamic update unit, used for inputting the predefined adjacency matrix and the time relationship matrix into a preset adjacency matrix update module, performing splicing processing in combination with static topological information and dynamic timing relationship, inputting the spliced ​​matrix into a sparse linear layer, capturing important relationships in the matrix while maintaining computational efficiency, and dynamically updating the processed matrix using a gated recurrent unit to generate a dynamic adjacency matrix;

[0022] A model training unit is used to input the dynamic adjacency matrix and the training data set into a preset dynamic graph convolution loop unit, replace the matrix penalty in the gated loop unit with a dynamic graph convolution module, use the dynamic graph convolution loop unit to capture the spatiotemporal correlation in the training data set, use the hidden state of the last dynamic graph convolution loop unit as the final output, obtain the predicted value, and calculate the error between the predicted value and the true value, save the model parameters with the minimum error as the optimal model parameters, complete the model training, and obtain the intercity carpooling OD to available capacity prediction model;

[0023] The prediction unit is used to use the inter-city carpooling OD to predict the available transportation capacity prediction model for the test data set to obtain a prediction output.

[0024] The present invention also provides a device for predicting regional available transportation capacity by intercity carpooling OD, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement the method for predicting regional available transportation capacity by intercity carpooling OD as described in any one of the above items.

[0025] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the computer-readable storage medium is located, so as to implement the method for predicting regional available transport capacity by intercity carpooling OD as described in any one of the above items.

[0026] In summary, the method for predicting regional available capacity for intercity carpooling OD pairs is based on the historical available capacity sequence and the historical demand sequence to predict the available capacity of the future starting and ending cities divided by the road network information. By predicting the available capacity of OD pairs in advance, the relevant transportation agencies can flexibly configure vehicles to achieve the goal of optimal matching of passengers with available capacity, improve vehicle utilization and reduce passenger waiting time. The present invention still has high accuracy in the face of the high correlation between the available capacity distribution and the passenger demand distribution, but the strong randomness of passenger demand leads to the uncertainty of the available capacity distribution. It solves the dynamic spatiotemporal correlation between the available capacity data and the passenger demand data within the city and between cities, and provides a good solution for the prediction of available capacity for intercity carpooling OD pairs, which meets the prediction of regional available capacity for intercity carpooling OD pairs in real life. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of the method for predicting regional available transport capacity by inter-city carpooling OD provided by the first aspect of the present invention;

[0028] Figure 2 It is a flow chart of the method for predicting regional available transport capacity by inter-city carpooling OD provided by the second aspect of the present invention;

[0029] Figure 3 It is a training optimization diagram of the method for predicting regional available transportation capacity of inter-city carpooling OD provided by an embodiment of the present invention;

[0030] Figure 4 It is a module schematic diagram of the device for predicting regional available transport capacity of inter-city carpooling OD provided by the first aspect of the present invention;

[0031] Figure 5 It is a module schematic diagram of the device for predicting regional available transport capacity of inter-city carpooling OD provided by the second aspect of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0033] refer to Figure 1 , Figure 2 , Figure 3 As shown, the first embodiment of the present invention discloses a method for predicting the available transportation capacity of an intercity carpooling OD to a region, which can be executed by a device for predicting the available transportation capacity of an intercity carpooling OD to a region (hereinafter referred to as a prediction device), and in particular, executed by one or more processors in the prediction device to implement the following method:

[0034] S1, obtaining the road network information of the starting and ending cities, the cities' point of interest data, historical order data, and historical transportation capacity data, obtaining the historical demand characteristics and historical inflow characteristics of all areas in the historical time segment based on the city's point of interest data, obtaining the available seat number characteristics and vehicle characteristics of all areas in the historical time segment based on the historical transportation capacity data, constructing the time information characteristics based on the historical order data and time information, and constructing a total data set, dividing the total data set into a training data set and a test data set according to a preset ratio, dividing the training data set into training samples according to a preset step size, and inputting the training samples into the DADGCRN model for training;

[0035] Specifically, step S1 includes: obtaining the road network information Road of the starting and ending cities, ,in, It is the main traffic road in the city. It is a secondary urban road. It is a third-level road. For highways, For expressway, For residential roads;

[0036] Get the city's point of interest data POI, ,in, For the name, For the address, are coordinates, For categories;

[0037] Get historical order data Data, ,in, For departure point, For the destination, For departure time, is the latitude and longitude;

[0038] Get historical capacity data CS, ,in, Number the driver. For the remaining seats in the vehicle;

[0039] According to the historical order data Data, the historical demand characteristics of all OD pairs within the time step are obtained, and the formula is: , get the historical demand characteristics ,in, For all OD pairs in The historical demand characteristics matrix of the time segment, For the i-th OD pair A collection of historical demand characteristics for a time segment, is the set of OD pairs of starting and ending cities, For the moment, is the time step, is the number of OD pairs on the previous day Travel demand in a certain time period, is the number of OD pairs in the previous week Travel demand in a certain time period, For the i-th OD pair history week The highest travel demand in the time segment, For the i-th OD pair history week The minimum travel demand within a time segment, For the i-th OD pair history week Average travel demand in a time segment;

[0040] According to the historical order data Data, the inflow characteristics of all regions in the historical time segment are obtained, and the formula is: , get the historical inflow characteristics ,in, is the number of OD pairs in the time step The number of net inflow passengers departing from the urban area in the time segment, For the i-th OD pair The number of net inflow passengers departing from the urban area during the time segment;

[0041] The historical available capacity characteristics of all OD pairs within the time step are obtained according to the historical capacity data CS, and the formula is: , get the historical available seats feature , Vehicle characteristics ,in, For The number of available seats in the i-th OD pair in the time segment, For The number of vehicles in the i-th OD pair in the time segment;

[0042] According to the point of interest data POI, the OD pair POI data feature set is obtained, and the expression is: ,in, , represents the number of POIs of each type in the i-th OD pair, is the number of catering and shopping categories, The number of life services, is the number of residential buildings, The number of office science and education categories, The number of leisure and entertainment categories, is the number of transportation facilities, The number of public facilities;

[0043] Construct time features based on time information. The formula is: , get the time information feature ,in, For the week category, For weekend features, For holiday characteristics;

[0044] Weekly Category Use periodic function to map and obtain the periodic function formula: ,in, is a cosine periodic function, Monday to Sunday;

[0045] Based on the historical available seat number characteristics , Vehicle characteristics , Historical demand characteristics , Historical inflow characteristics and time information features Constructing the total dataset , and divide the total data set into training data set and test data set according to the ratio of 7:3, and divide the training data set into test data set according to the fixed input step size Divide into training samples, and divide the training samples Input into the DADGCRN model for training, the training sample expression is:

[0046] ;

[0047] The training output expression is:

[0048] ;

[0049] in, is the logarithm of OD, is the output step size, Indicated in The Nth training output in the time segment.

[0050] In this embodiment, the road network information Road and point of interest (POI) data POI of the starting and ending cities are obtained to construct spatial features, and the number of available seats in all areas within the historical time segment is obtained from the available capacity data CS. and available vehicles , obtain the passenger demand of all regions in the historical time segment from the order data Data and inflow characteristics , and the temporal features constructed by temporal information , construct a dataset from all the above acquired features , according to the proportion of the data set Divide into training data set and test data set, and input the training data set according to the fixed input step size Divide into training samples, and divide the training samples Input DADGCRN model for training.

[0051] Specifically, the road network information of the starting and ending cities is obtained from the OpenStreetMap official website. ,in, It is the main traffic road in the city. It is a secondary urban road. It is a third-level road. For highways, For expressway, Residential roads.

[0052] Get the city's POI data from the Amap official website. ,in, For the name, For the address, are coordinates, For category.

[0053] Historical order data: ,in, For departure point, For the destination, For departure time, is the longitude and latitude.

[0054] Historical capacity data: ,in, Number the driver. For the remaining seats in the vehicle.

[0055] The historical demand characteristics of all OD pairs within the time step, that is, the historical demand characteristics of all areas of the departure city A to all areas of the destination city B within the time step, are expressed as: , Indicates that all OD pairs are Historical demand characteristics of time segments, where , Indicates that the i-th OD pair is A collection of historical demand characteristics for a time segment; Indicates that the i-th OD pair was Travel demand in a certain time period, Indicates the number of OD pairs in the previous week Travel demand in a certain time period, Indicates the i-th OD pair in history for one week The highest travel demand in the time segment, Indicates the i-th OD pair in history for one week The minimum travel demand within a time segment, Indicates the i-th OD pair in history for one week Average travel demand within a time segment.

[0056] The historical inflow characteristics of the OD pair within the time step, that is, the number of passengers who flowed into the departure city area from the destination city B area in a historical day, can be expressed as: ,in Indicates that the i-th OD pair is The net inflow of passengers departing from the urban area during the time segment.

[0057] The historical available capacity characteristics of all OD pairs within the time step, that is, the historical available seats and vehicle characteristics of all areas of the departure city A to all areas of the destination city B within the time step, are expressed as: ,in, Indicated in The number of available seats in the i-th OD pair in the time segment, Indicated in The number of vehicles in the i-th OD pair in the time segment.

[0058] The OD expression for POI data feature set is: ,in, , represents the number of POIs of each type in the i-th OD pair, is the number of catering and shopping categories, The number of life services, is the number of residential buildings, The number of office science and education categories, The number of leisure and entertainment categories, is the number of transportation facilities, The number of public facilities.

[0059] The expression of OD for time information feature set is: ,in, is the week category. In order to fully reflect the time correlation between weeks, the week category is mapped using a periodic function. For example, the inter-city travel demand on Sunday is greater than that on Monday. However, the traditional setting of Monday to Sunday from 0 to 6 or using one-hot encoding cannot reflect the above time correlation. It is a weekend feature, with a weekend value of 1 and a weekday value of 0; It is a holiday feature, with holidays as 1 and working days as 0;

[0060] The periodic function is: ,in, is a cosine periodic function, Use 1 to 7 for Monday to Sunday.

[0061] The above historical available seats feature , Vehicle characteristics , Historical demand characteristics , Historical inflow characteristics And time information features Constructed as a total data set , the total data set is divided into training data set and test data set according to the ratio of 7:3, and the training data set is divided into training data set and test data set according to the fixed input step size Divide into training samples, and divide the training samples Input to the model for training, the training sample is expressed as:

[0062] ;

[0063] The training output is represented as:

[0064] ;

[0065] in, is the logarithm of OD, is the output step size.

[0066] S2, removing and topologically processing the road network information data of the starting and ending cities to realize the regional division of the city, reclassifying the point of interest data of the city and processing the regional POI identification, weighting each POI by TF-IDF algorithm, using the cosine similarity formula to calculate the POI similarity between regions to form a predefined adjacency matrix, and forming an adjacency relationship graph of the starting and ending cities with a pair of OD regions as nodes and POI similarity as edge weights;

[0067] Specifically, step S2 includes: using ArcGIS software to perform data removal and topological processing on the road network information Road of the starting and ending cities, so as to realize regional division of the starting and ending cities, and dividing the starting city A and the destination city B into regions O and D respectively; , forming o×d OD pairs;

[0068] The POI data of the city are processed with regional identification, and the TF-IDF algorithm is used to calculate the weight of each type of POI in each OD pair. The POI similarity between the OD pairs is calculated by the cosine similarity formula, and the adjacency matrix of the OD pairs of the starting and ending cities is constructed. The TF-IDF algorithm formula is:

[0069] ;

[0070] ;

[0071] ;

[0072] ;

[0073] ;

[0074] in, is the POI weight of the departure area o, is the POI weight of the destination area d, is the number of POIs of one type in the departure area o, is the total number of POIs in the departure area o, is the number of POIs of one type in the departure area d, is the total number of POIs in the departure area d, is the number of OD logarithms containing a type of POI;

[0075] The cosine similarity formula is: ,in, is the cosine similarity value, Represents the weight vector of all departure area POIs, Represents the weight vector of all POIs in the destination area;

[0076] Construct a predefined adjacency matrix , ,in, is the POI similarity between the Nth OD pair and the Nth OD pair;

[0077] According to the regional division of the starting and ending cities, OD pairs are formed and the POI similarity between the OD pairs is used to construct an adjacency relationship graph. The expression of the adjacency relationship graph is: .

[0078] In this embodiment, the road network information of the starting and ending cities is removed and topologically processed to divide the cities into regions, the POI data is reclassified, and then the regional POI is identified. Finally, each POI is weighted by the TF-IDF algorithm. The POI similarity between regions is calculated by cosine similarity to form a predefined adjacency matrix. The adjacency relationship graph of the starting and ending cities is formed by a pair of OD regions as nodes and POI similarity as edge weights. .

[0079] Specifically, based on the road network information Road, ArcGIS software is used to perform data elimination and topological processing to divide the starting and ending cities into regions. The departure city A and the destination city B are divided into regions O and D. , finally forming o×d OD pairs;

[0080] The acquired POI data is regionally labeled, and the TF-IDF algorithm is used to calculate the weight of each type of POI in each OD pair, and the POI similarity between OD pairs is calculated by cosine similarity to construct the adjacency matrix of the OD pairs of the starting and ending cities. The TF-IDF algorithm is:

[0081] ;

[0082] ;

[0083] ;

[0084] ;

[0085] ;

[0086] in, is the POI weight of the departure area o, is the POI weight of the destination area d, is the number of POIs of one type in the departure area o, is the total number of POIs in the departure area o, is the number of OD logarithms containing a type of POI, and N is the number of OD logarithms;

[0087] The cosine similarity formula is: ,in, is the cosine similarity value, Represents the weight vector of all departure area POIs, Represents the weight vector of all POIs in the destination area.

[0088] Constructing POI similarity matrix : ,in, is the POI similarity between the first OD pair and the Nth OD pair.

[0089] The obtained regional division of the starting and ending cities forms OD pairs and the POI similarity between the OD pairs is used to construct an adjacency relationship graph. The adjacency relationship graph of the OD pairs of the starting and ending cities is expressed as: ,in, Represents the node set consisting of the departure city area and the destination city area. The departure city A and the destination city B are divided into areas O and D. , , and finally form o×d OD pairs, that is, , is the POI similarity matrix.

[0090] S3, inputting the training data set into a preset time relationship embedding module, and using a self-attention mechanism to learn the time relationship between the historical order data and the historical transportation data to generate a time relationship matrix;

[0091] Specifically, step S3 includes: inputting the training data set into a preset time relationship embedding module, capturing the time transformation relationship between available capacity and historical order data by using a self-attention mechanism, and obtaining a time relationship matrix , the expression of the self-attention mechanism is:

[0092] ;

[0093] ;

[0094] ;

[0095] ;

[0096] in, , , denote the query matrix, key matrix and value matrix respectively, , , Respectively , , The weight matrix of represents the scaling factor, represents the activation function Softmax, is the transpose of the matrix.

[0097] In this embodiment, the above-mentioned training data set is input into the temporal relationship embedding module, and the temporal relationship between historical order data and capacity data is learned through the self-attention mechanism. A temporal relationship matrix is ​​generated to reflect the correlation between each time segment in the input sequence, thereby reducing the impact of order uncertainty on capacity data.

[0098] Specifically, the training data set is input into the temporal relationship embedding module, and the temporal transformation relationship between available capacity and historical order data is captured through the self-attention mechanism to obtain the temporal relationship matrix , self-attention mechanism expression:

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] in, , , denote the query matrix, key matrix and value matrix respectively, , , Respectively , , The weight matrix of Represents the scaling factor so that the result is not too large or too small. Represents the activation function Softmax.

[0104] S4, inputting the predefined adjacency matrix and the time relationship matrix into a preset adjacency matrix update module, performing splicing processing in combination with static topological information and dynamic timing relationship, inputting the spliced ​​matrix into a sparse linear layer, capturing important relationships in the matrix while maintaining computational efficiency, and dynamically updating the processed matrix using a gated recurrent unit to generate a dynamic adjacency matrix;

[0105] Specifically, step S4 includes: converting the predefined adjacency matrix and time relationship matrix Splicing process to get the matrix , ,in, Represents a splicing operation;

[0106] The matrix Input sparse linear layer to get matrix , ,in, Represents the coefficient matrix of the sparse linear layer;

[0107] Use the preset gated recurrent unit according to the matrix and the predefined adjacency matrix , generate a dynamic adjacency matrix , the calculation formula is:

[0108] ;

[0109] ;

[0110] ;

[0111] ;

[0112] in, , They are the update gate and reset gate in the gated recurrent unit, represents the Hadamard product, , , are the parameters of the corresponding sparse connection layer, is the gated recurrent unit state, is the t-th derived dynamic adjacency matrix, is the activation function, for Activation function.

[0113] In this embodiment, the above-mentioned predefined adjacency matrix and the time relationship matrix are input into the adjacency matrix update module, and the predefined adjacency matrix and the time relationship matrix are spliced, combining the static topological information and the dynamic timing relationship. In order to effectively process the spliced ​​matrix, a sparse linear layer is input to capture the important relationships in the matrix while maintaining computational efficiency. Finally, a gated recurrent unit is input to dynamically update the processed matrix to generate a new dynamic adjacency matrix.

[0114] Specifically, the predefined adjacency matrix The time relationship matrix above Concatenate to get the matrix , ,in, Represents a splicing operation;

[0115] The concatenated matrix is ​​input into the sparse linear layer for processing, and we get , ,in, Represents the coefficient matrix of the sparse linear layer;

[0116] Will get The historical spatial information is the predefined adjacency matrix, and the dynamic adjacency matrix is ​​generated by the modified gated recurrent unit , the calculation process is as follows:

[0117] ;

[0118] ;

[0119] ;

[0120] ;

[0121] In the formula, represents the Hadamard product, , , are the parameters of the corresponding sparsely connected layer, , They are the update gate and reset gate in the gated recurrent unit, is the gated recurrent unit state, It is the derived dynamic adjacency matrix, which contains context information and the spatiotemporal relationship information between capacity data and order data.

[0122] S5, input the dynamic adjacency matrix and the training data set into a preset dynamic graph convolution loop unit, replace the matrix penalty in the gated loop unit with a dynamic graph convolution module, use the dynamic graph convolution loop unit to capture the spatiotemporal correlation in the training data set, use the hidden state of the last dynamic graph convolution loop unit as the final output, obtain the predicted value, and calculate the error between the predicted value and the true value, save the model parameters with the minimum error as the optimal model parameters, complete the model training, and obtain the intercity carpooling OD to available capacity prediction model;

[0123] Specifically, step S5 includes: converting the dynamic adjacency matrix Input a single convolutional element of the dynamic graph, use the weighted average sum of the input graph signal and the dynamic adjacency matrix in each graph convolution layer, and use bidirectional graph convolution calculation. The calculation formula is:

[0124] ;

[0125] ;

[0126] ;

[0127] ;

[0128] ;

[0129] in, is a hyperparameter, and Represent the input and output node states respectively, is the graph convolution operation, is a learnable parameter, To spread depth;

[0130] The training data set is input into the gated recurrent unit to obtain the hidden value output, and the calculation formula is:

[0131] ;

[0132] ;

[0133] ;

[0134] ;

[0135] in, represents the output hidden state of the dynamic graph loop module at time t, represents the sigmoid activation function, , They are the reset gate and update gate in the dynamic graph loop module at time t. , , Respectively represent the defined dynamic graph convolution;

[0136] The spatiotemporal correlation in the training data set is captured by a dynamic graph convolutional recurrent unit, and the hidden state of the last dynamic graph convolutional recurrent unit is used as the final output to obtain a predicted value;

[0137] The error between the predicted value and the true value is calculated, and the model parameters with the minimum error are saved as the optimal model parameters. The model training is completed to obtain the intercity carpooling OD prediction model for available capacity.

[0138] In this embodiment, the new dynamic adjacency matrix obtained above and the training data set are input into the dynamic graph convolution loop unit, which is obtained by replacing the matrix penalty in the gated loop unit with the dynamic graph convolution module. The dynamic graph convolution loop unit captures the spatiotemporal correlation in the training data set, and the hidden state of the last dynamic graph convolution loop unit is used as the final output to obtain the predicted value. The error between the predicted value and the true value is calculated, and the model parameters with the minimum error are saved as the optimal model to complete the model training.

[0139] Specifically, the above dynamic adjacency matrix To the dynamic graph convolution, the weighted average sum of the input graph signal and the dynamic adjacency matrix is ​​used in each graph convolution layer. In order to more effectively utilize the information in the graph, a bidirectional graph convolution is adopted. The calculation process is as follows:

[0140] ;

[0141] ;

[0142] ;

[0143] ;

[0144] ;

[0145] in, is a hyperparameter used to control the weights of different components. and Represent the input and output node states respectively, is the graph convolution operation, is a learnable parameter, To spread the depth.

[0146] The training data set is input into the modified gated recurrent unit, and the hidden value output obtained is the output value. The calculation process is as follows:

[0147] ;

[0148] ;

[0149] ;

[0150] ;

[0151] in, represents the output hidden state of the dynamic graph loop module at time t, represents the Hadamard product, represents the sigmoid activation function, , They are the reset gate and update gate in the dynamic graph loop module at time t. , , They represent the dynamic graph convolution defined in the above steps. The dynamic graph convolution loop unit captures the spatiotemporal correlation in the training data set. The hidden state of the last dynamic graph convolution loop unit is used as the final output to obtain the predicted value. The error between the predicted value and the true value is calculated, and the model parameters with the minimum error are saved as the optimal model to complete the model training.

[0152] S6, using the intercity carpooling OD available transport capacity prediction model to predict the test data set to obtain a prediction output.

[0153] Specifically, step S6 includes: according to a preset fixed time step Divide the test data set into test samples , the test sample Input the trained intercity carpooling OD into the available capacity prediction model to obtain the prediction output;

[0154] The expression of the test sample is:

[0155] ;

[0156] The expression for the predicted output is:

[0157] ;

[0158] in, is the prediction result matrix, Representing the future Available capacity for the Nth OD pair area in the time segment.

[0159] In this embodiment, the optimal model is loaded, and the above test data set is input into the loaded optimal DADGCRN model to obtain a prediction output.

[0160] Specifically, the above test data set is divided into Form a test sample and place the test sample Input into the intercity carpooling OD pair available capacity prediction model to obtain the prediction output; the test sample is expressed as:

[0161] ;

[0162] The prediction output expression of the intercity carpooling OD to the available capacity prediction model is:

[0163] ;

[0164] in, is the prediction result matrix, Representing the future Available capacity for the Nth OD pair area in the time segment.

[0165] In this embodiment, once the DADGCRN-based intercity carpooling OD to regional available capacity prediction model is trained, the OD to regional available capacity of intercity carpooling can be predicted according to the trained DADGCRN-based intercity carpooling OD to regional available capacity demand prediction model, and new data with a fixed time step is input into the OD to regional available capacity prediction model to realize the prediction of the OD to regional available capacity for the next time step in the future.

[0166] The following case is used to further illustrate this embodiment. The historical data of a company's intercity carpooling orders are selected as the research object, specifically the passenger carpooling order data set and available transportation capacity data set for round trips between City A and City B during March 2023. The passenger carpooling order data includes the scheduled boarding time, the longitude and latitude of the boarding point, the longitude and latitude of the alighting point, and the number of companions, etc. The available transportation capacity data set includes the longitude and latitude of the vehicle and the number of remaining seats. As shown in Table 1, the statistical data of the data set are summarized.

[0167] Table 1 Statistics of the dataset

[0168]

[0169] The data from March 1, 2023 to March 25, 2023 are divided into the training set for model training, and the data from March 25, 2023 to March 31, 2023 are divided into the test set for model testing. The model parameter settings are shown in Table 2.

[0170] Table 2 Model parameter settings

[0171]

[0172] The present invention compares some typical prediction model methods, such as:

[0173] (1) Diffusion Convolutional Recurrent Neural Network is a deep learning model specifically used for spatiotemporal data prediction. It uses diffusion graph convolution to capture spatial dependencies and introduces a recurrent neural network structure to handle temporal dependencies. It is suitable for prediction of dynamically changing spatiotemporal data.

[0174] (2) Dynamic Graph Convolutional Recurrent Network (DGRN) is designed to process dynamic graph structures that change over time. Unlike static graph convolutional networks, DGRN can adapt to the dynamic changes of graph structures and node features. It uses dynamic graph convolution to capture dynamic spatial dependencies and recurrent neural networks to capture the temporal dependencies of time series data.

[0175] (3) Adaptive Graph Convolutional Recurrent Network aims to capture complex spatiotemporal dependencies by adaptively learning graph structures. The core part is the adaptive graph convolution layer, which dynamically learns the adjacency matrix of the graph during training. It does not rely on a predefined graph structure and uses a recurrent neural network to capture temporal dependencies.

[0176] (4) Decomposition Dynamic graph convolutional recurrent network aims to adaptively capture complex spatiotemporal dependencies through signal decomposition. The core part is to perform bidirectional prediction after the recurrent graph convolution module. The reverse prediction is the decomposition process to obtain the unlearned part and use the recurrent neural network to capture the temporal dependency of time series data.

[0177] The performance of the model method of the present invention is compared by calculating the mean absolute error MAE, root mean square error RMSE, mean absolute percentage error MAPE, and symmetric mean absolute percentage error SMAPE of the above method and the present invention. The lower the value of each evaluation index, the better the prediction accuracy of the model. The calculation formula is as follows:

[0178] ;

[0179] ;

[0180] ;

[0181] ;

[0182] in, and are the true value and the corresponding predicted value respectively, is the output step length. The comparison results are shown in Table 3.

[0183] Table 3 Performance of different methods in prediction effect

[0184]

[0185] By observing and comparing the MAE, RMSE, MAPE and SMAPE of different methods, it can be found that the MAE, MAPE and SMAPE values ​​calculated by the OD-to-region available capacity prediction method for intercity carpooling provided by the present invention are lower than those of other typical calculation and prediction methods, and the RMSE value of the method of the present invention is lower than that of most other typical calculation and prediction methods. However, compared with the DGCRN method, when the RMSE value is similar, the values ​​of other indicators are improved. Therefore, compared with the prior art, the scheme of the present invention has the following beneficial effects:

[0186] The method for predicting available transport capacity for regional demand of OD pairs of intercity carpooling provided by the present invention predicts available transport capacity for future start and end cities divided by road network information based on historical demand sequences and historical available transport capacity sequences. By predicting the available transport capacity of OD pairs in advance, relevant transportation agencies can flexibly configure vehicles in advance to achieve the goal of multi-person carpooling and shared travel, and improve vehicle utilization. In terms of the problem of predicting available transport capacity for OD pairs in a region, the present invention considers the impact of order data on transport capacity distribution. As time goes by, order data causes the relationship between various regions to change continuously. By fusing order data with transport capacity data to learn the time relationship between them, a time relationship matrix is ​​generated, and it is combined with a predefined adjacency matrix to form a dynamic adjacency matrix, which better captures the complex spatiotemporal relationship between OD pairs and between orders and transport capacity, reduces the uncertainty caused by the randomness of order generation to capacity prediction, and improves the accuracy of capacity prediction.

[0187] In summary, by predicting the available capacity of OD pairs in advance through the intercity carpooling OD pair regional available capacity prediction method, the approximate distribution of available capacity can be understood in advance, the goal of optimal matching of passenger demand and available capacity can be achieved, the utilization rate of vehicles can be improved, and the waiting time of passengers can be reduced. This meets the requirements of the OD pair regional available capacity prediction for intercity carpooling in real life.

[0188] See also Figure 4 , Figure 5 The second embodiment of the present invention provides a device for predicting regional available transportation capacity of inter-city carpooling OD, which includes:

[0189] An input acquisition unit 201 is used to acquire the road network information of the starting and ending cities, the points of interest data of the cities, the historical order data and the historical transportation capacity data, acquire the historical demand characteristics and the historical inflow characteristics of all areas in the historical time segment according to the points of interest data of the cities, acquire the available seat number characteristics and the vehicle characteristics of all areas in the historical time segment according to the historical transportation capacity data, construct the time information characteristics according to the historical order data and the time information, and construct a total data set, divide the total data set into a training data set and a test data set according to a preset ratio, divide the training data set into training samples according to a preset step size, and input the training samples into the DADGCRN model for training;

[0190] The adjacency relationship graph construction unit 202 is used to remove and topologically process the road network information data of the starting and ending cities to realize the regional division of the city, reclassify the point of interest data of the city and process the regional POI identification, weight each POI through the TF-IDF algorithm, use the cosine similarity formula to calculate the POI similarity between regions to form a predefined adjacency matrix, and use a pair of OD regions as nodes and POI similarity as edge weights to form an adjacency relationship graph of the starting and ending cities;

[0191] A time relationship matrix generating unit 203 is used to input the training data set into a preset time relationship embedding module, and learn the time relationship between the historical order data and the historical transportation data by using a self-attention mechanism to generate a time relationship matrix;

[0192] A dynamic updating unit 204 is used to input the predefined adjacency matrix and the time relationship matrix into a preset adjacency matrix updating module, perform splicing processing in combination with static topological information and dynamic timing relationship, input the spliced ​​matrix into a sparse linear layer, capture important relationships in the matrix while maintaining computational efficiency, and dynamically update the processed matrix using a gated recurrent unit to generate a dynamic adjacency matrix;

[0193] The model training unit 205 is used to input the dynamic adjacency matrix and the training data set into a preset dynamic graph convolution loop unit, replace the matrix penalty in the gated loop unit with a dynamic graph convolution module, use the dynamic graph convolution loop unit to capture the spatiotemporal correlation in the training data set, use the hidden state of the last dynamic graph convolution loop unit as the final output, obtain the predicted value, and calculate the error between the predicted value and the true value, save the model parameters with the minimum error as the optimal model parameters, complete the model training, and obtain the intercity carpooling OD to available capacity prediction model;

[0194] The prediction unit 206 is used to use the inter-city carpooling OD available transportation capacity prediction model to predict the test data set and obtain a prediction output.

[0195] The third embodiment of the present invention provides a device for predicting the available regional transportation capacity of an intercity carpooling OD, including a memory and a processor, wherein a computer program is stored in the memory, and the computer program can be executed by the processor to implement the method for predicting the available regional transportation capacity of an intercity carpooling OD as described in any one of the above items.

[0196] The fourth embodiment of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the computer-readable storage medium is located, so as to implement the method for predicting regional available transportation capacity of inter-city carpooling OD as described in any one of the above items.

[0197] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for predicting regional available transport capacity based on inter-city carpooling OD, characterized in that: include: Obtain the road network information of the starting and ending cities, the cities' point of interest data, historical order data, and historical transportation capacity data; obtain the historical demand characteristics and historical inflow characteristics of all areas in the historical time segment based on the city's point of interest data; obtain the available seat number characteristics and vehicle characteristics of all areas in the historical time segment based on the historical transportation capacity data; construct time information characteristics based on the historical order data and time information; construct a total data set; divide the total data set into a training data set and a test data set according to a preset ratio; divide the training data set into training samples according to a preset step size; and input the training samples into a DADGCRN model for training; The road network information data of the starting and ending cities are eliminated and topologically processed to realize the regional division of the city, the point of interest data of the city is reclassified and regional POI identification is processed, each POI is weighted by the TF-IDF algorithm, and the POI similarity between regions is calculated using the cosine similarity formula to form a predefined adjacency matrix, and an adjacency relationship graph of the starting and ending cities is formed with a pair of OD regions as nodes and POI similarity as edge weights; Input the training data set into a preset time relationship embedding module, and learn the time relationship between the historical order data and the historical transportation data by using a self-attention mechanism to generate a time relationship matrix; The predefined adjacency matrix and the time relationship matrix are input into a preset adjacency matrix update module, and are spliced ​​in combination with static topological information and dynamic timing relationships. The spliced ​​matrix is ​​input into a sparse linear layer, and while maintaining computational efficiency, important relationships in the matrix are captured. The processed matrix is ​​dynamically updated using a gated recurrent unit to generate a dynamic adjacency matrix. The dynamic adjacency matrix and the training data set are input into a preset dynamic graph convolution loop unit, the matrix penalty in the gated loop unit is replaced with a dynamic graph convolution module, and the dynamic graph convolution loop unit is used to capture the spatiotemporal correlation in the training data set. The hidden state of the last dynamic graph convolution loop unit is used as the final output to obtain the predicted value, and the error between the predicted value and the true value is calculated, and the model parameters with the minimum error are saved as the optimal model parameters. The model training is completed to obtain the intercity carpooling OD to available capacity prediction model; The intercity carpooling OD is used to predict the available capacity prediction model for the test data set to obtain a prediction output.

2. The method for predicting regional available transport capacity based on intercity carpooling OD according to claim 1 is characterized in that: Obtain the road network information of the starting and ending cities, the city's point of interest data, historical order data, and historical transportation capacity data; obtain the historical demand characteristics and historical inflow characteristics of all areas in the historical time segment based on the road network information of the starting and ending cities and the city's point of interest data; obtain the available seat number characteristics and vehicle characteristics of all areas in the historical time segment based on the historical transportation capacity data; construct time information characteristics based on the historical order data and time information; construct a total data set; divide the total data set into a training data set and a test data set according to a preset ratio; divide the training data set into training samples according to a preset step size; and input the training samples into the DADGCRN model for training, specifically: Get the road network information Road of the starting and ending cities, ,in, It is the main traffic road in the city. It is a secondary urban road. It is a third-level road. For highways, For expressway, For residential roads; Get the city's point of interest data POI, ,in, For the name, For the address, is the coordinate, For categories; Get historical order data Data, ,in, For departure point, For the destination, For departure time, is the latitude and longitude; Get historical capacity data CS, ,in, Number the driver. For the remaining seats in the vehicle; According to the historical order data Data, the historical demand characteristics of all OD pairs within the time step are obtained, and the formula is: , get the historical demand characteristics ,in, For all OD pairs in The historical demand characteristics matrix of the time segment, is the i-th OD pair in A collection of historical demand characteristics for a time segment, is the set of OD pairs of starting and ending cities, For the moment, is the time step, is the number of OD pairs on the previous day Travel demand in a certain time period, is the number of OD pairs in the previous week Travel demand in a certain time period, For the i-th OD pair history week The highest travel demand in the time segment, For the i-th OD pair history week The minimum travel demand within a time segment, For the i-th OD pair history week Average travel demand in a time segment; According to the historical order data Data, the inflow characteristics of all regions in the historical time segment are obtained, and the formula is: , get the historical inflow characteristics ,in, is the number of OD pairs in the time step The number of net inflow passengers departing from the urban area in the time segment, is the i-th OD pair in The number of net inflow passengers departing from the urban area during the time segment; The historical available capacity characteristics of all OD pairs within the time step are obtained according to the historical capacity data CS, and the formula is: , get the historical available seats feature , Vehicle characteristics ,in, For The number of available seats in the i-th OD pair in the time segment, For The number of vehicles in the i-th OD pair in the time segment; According to the point of interest data POI, the OD pair POI data feature set is obtained, and the expression is: ,in, , represents the number of POIs of each type in the i-th OD pair, is the number of catering and shopping categories, The number of life services, is the number of residential buildings, The number of office science and education categories, The number of leisure and entertainment categories, is the number of transportation facilities, The number of public facilities; Construct time features based on time information. The formula is: , get the time information feature ,in, For the week category, For weekend features, For holiday characteristics; Weekly Category Use periodic function to map and obtain the periodic function formula: ,in, is a cosine periodic function, Monday to Sunday; Based on the historical available seat number characteristics , Vehicle characteristics , Historical demand characteristics , Historical inflow characteristics and time information features Constructing the total dataset , and divide the total data set into training data set and test data set according to the ratio of 7:3, and divide the training data set into test data set according to the fixed input step size Divide into training samples, and divide the training samples Input into the DADGCRN model for training, the training sample expression is: ； The training output expression is: ； in, is the logarithm of OD, is the output step size, Indicated in The Nth training output in the time segment.

3. The method for predicting regional available transport capacity based on intercity carpooling OD according to claim 2 is characterized in that: The road network information data of the starting and ending cities are eliminated and topologically processed to realize the regional division of the city, the point of interest data of the city is reclassified and regional POI identification is processed, each POI is weighted by the TF-IDF algorithm, and the cosine similarity formula is used to calculate the POI similarity between regions to form a predefined adjacency matrix, and a pair of OD regions are used as nodes and POI similarity is used as the edge weight to form an adjacency relationship graph of the starting and ending cities, specifically: ArcGIS software is used to remove data and perform topological processing on the road network information of the starting and ending cities, so as to realize the regional division of the starting and ending cities, and the departure city A and the destination city B are divided into regions O and D respectively. , forming o×d OD pairs; The POI data of the city are processed with regional identification, and the TF-IDF algorithm is used to calculate the weight of each type of POI in each OD pair. The POI similarity between the OD pairs is calculated by the cosine similarity formula, and the adjacency matrix of the OD pairs of the starting and ending cities is constructed. The TF-IDF algorithm formula is: ； ； ； ； ； in, is the POI weight of the departure area o, is the POI weight of the destination area d, is the number of POIs of one type in the departure area o, is the total number of POIs in the departure area o, is the number of POIs of one type in the departure area d, is the total number of POIs in the departure area d, is the number of OD logarithms containing a type of POI; The cosine similarity formula is: ,in, is the cosine similarity value, Represents the weight vector of all departure area POIs, Represents the weight vector of all POIs in the destination area; Construct a predefined adjacency matrix , ,in, is the POI similarity between the Nth OD pair and the Nth OD pair; According to the regional division of the starting and ending cities, OD pairs are formed and the POI similarity between the OD pairs is used to construct an adjacency relationship graph. The expression of the adjacency relationship graph is: .

4. The method for predicting regional available transport capacity based on intercity carpooling OD according to claim 3 is characterized in that: The training data set is input into a preset time relationship embedding module, and the time relationship between the historical order data and the historical transportation data is learned by using a self-attention mechanism to generate a time relationship matrix, specifically: The training data set is input into the preset time relationship embedding module, and the time transformation relationship between available capacity and historical order data is captured by using the self-attention mechanism to obtain the time relationship matrix , the expression of the self-attention mechanism is: ； ； ； ； in, , , denote the query matrix, key matrix and value matrix respectively, , , Respectively , , The weight matrix of represents the scaling factor, represents the activation function Softmax, is the transpose of the matrix.

5. The method for predicting regional available transport capacity based on intercity carpooling OD according to claim 4 is characterized in that: The predefined adjacency matrix and the time relationship matrix are input into a preset adjacency matrix update module, and are spliced ​​in combination with static topological information and dynamic temporal relationships. The spliced ​​matrix is ​​input into a sparse linear layer, and while maintaining computational efficiency, important relationships in the matrix are captured. The processed matrix is ​​dynamically updated using a gated recurrent unit to generate a dynamic adjacency matrix, specifically: The predefined adjacency matrix and time relationship matrix Splicing process to get the matrix , ,in, Represents a splicing operation; The matrix Input sparse linear layer to get matrix , ,in, Represents the coefficient matrix of the sparse linear layer; Use the preset gated recurrent unit according to the matrix and the predefined adjacency matrix , generate a dynamic adjacency matrix , the calculation formula is: ； ； ； ； in, , They are the update gate and reset gate in the gated recurrent unit, represents the Hadamard product, , , are the parameters of the corresponding sparse connection layer, is the gated recurrent unit state, is the t-th derived dynamic adjacency matrix, is the activation function, for Activation function.

6. The method for predicting regional available transport capacity based on intercity carpooling OD according to claim 5 is characterized in that: The dynamic adjacency matrix and the training data set are input into a preset dynamic graph convolution loop unit, the matrix penalty in the gated loop unit is replaced with a dynamic graph convolution module, and the dynamic graph convolution loop unit is used to capture the spatiotemporal correlation in the training data set. The hidden state of the last dynamic graph convolution loop unit is used as the final output to obtain the predicted value, and the error between the predicted value and the true value is calculated. The model parameters with the minimum error are saved as the optimal model parameters, and the model training is completed to obtain the intercity carpooling OD to available capacity prediction model, which is specifically: The dynamic adjacency matrix Input a single convolutional element of the dynamic graph, use the weighted average sum of the input graph signal and the dynamic adjacency matrix in each graph convolution layer, and use bidirectional graph convolution calculation. The calculation formula is: ； ； ； ； ； in, is a hyperparameter, and Represent the input and output node states respectively, is the graph convolution operation, is a learnable parameter, To spread depth; The training data set is input into the gated recurrent unit to obtain the hidden value output, and the calculation formula is: ； ； ； ； in, represents the output hidden state of the dynamic graph loop module at time t, represents the sigmoid activation function, , They are the reset gate and update gate in the dynamic graph loop module at time t. , , Respectively represent the defined dynamic graph convolution; The spatiotemporal correlation in the training data set is captured by a dynamic graph convolutional recurrent unit, and the hidden state of the last dynamic graph convolutional recurrent unit is used as the final output to obtain a predicted value; The error between the predicted value and the true value is calculated, and the model parameters with the minimum error are saved as the optimal model parameters. The model training is completed to obtain the intercity carpooling OD prediction model for available capacity.

7. The method for predicting regional available transport capacity based on intercity carpooling OD according to claim 6 is characterized in that: The intercity carpooling OD is used to predict the available capacity prediction model for the test data set to obtain the prediction output, which is specifically: According to the preset fixed time step Divide the test data set into test samples , the test sample Input the trained intercity carpooling OD into the available capacity prediction model to obtain the prediction output; The expression of the test sample is: ； The expression for the predicted output is: ； in, is the prediction result matrix, Representing the future Available capacity for the Nth OD pair area in the time segment.

8. A device for predicting regional available transport capacity for inter-city carpooling OD, characterized in that: include: An input acquisition unit is used to acquire the road network information of the starting and ending cities, the point of interest data of the cities, the historical order data and the historical transportation capacity data, acquire the historical demand characteristics and the historical inflow characteristics of all areas in the historical time segment according to the point of interest data of the cities, acquire the available seat number characteristics and the vehicle characteristics of all areas in the historical time segment according to the historical transportation capacity data, construct the time information characteristics according to the historical order data and the time information, and construct a total data set, divide the total data set into a training data set and a test data set according to a preset ratio, divide the training data set into training samples according to a preset step size, and input the training samples into the DADGCRN model for training; The adjacency relationship graph construction unit is used to remove and topologically process the road network information data of the starting and ending cities to realize the regional division of the city, reclassify the point of interest data of the city and process the regional POI identification, weight each POI through the TF-IDF algorithm, use the cosine similarity formula to calculate the POI similarity between regions to form a predefined adjacency matrix, and use a pair of OD regions as nodes and POI similarity as edge weights to form an adjacency relationship graph of the starting and ending cities; A time relationship matrix generating unit, used for inputting the training data set into a preset time relationship embedding module, and generating a time relationship matrix by learning the time relationship between the historical order data and the historical transportation data by using a self-attention mechanism; A dynamic update unit, used for inputting the predefined adjacency matrix and the time relationship matrix into a preset adjacency matrix update module, performing splicing processing in combination with static topological information and dynamic timing relationship, inputting the spliced ​​matrix into a sparse linear layer, capturing important relationships in the matrix while maintaining computational efficiency, and dynamically updating the processed matrix using a gated recurrent unit to generate a dynamic adjacency matrix; A model training unit is used to input the dynamic adjacency matrix and the training data set into a preset dynamic graph convolution loop unit, replace the matrix penalty in the gated loop unit with a dynamic graph convolution module, use the dynamic graph convolution loop unit to capture the spatiotemporal correlation in the training data set, use the hidden state of the last dynamic graph convolution loop unit as the final output, obtain the predicted value, and calculate the error between the predicted value and the true value, save the model parameters with the minimum error as the optimal model parameters, complete the model training, and obtain the intercity carpooling OD to available capacity prediction model; The prediction unit is used to use the inter-city carpooling OD to predict the available transportation capacity prediction model for the test data set to obtain a prediction output.

9. An inter-city carpooling OD prediction device for regional available transport capacity, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement the method for predicting the available regional transportation capacity of the inter-city carpooling OD as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored, and the computer program can be executed by a processor of the device where the computer-readable storage medium is located to implement the method for predicting the available regional transportation capacity of inter-city carpooling OD as described in any one of claims 1 to 7.

Images