Cross-modal transferable population travel demand prediction method and device

Abstract

The application relates to the technical field of traffic planning. The application discloses a cross-modal transferable population travel demand prediction method and device, which can improve the accuracy of the total amount of population travel demand. The cross-modal transferable population travel demand prediction method comprises the following steps: obtaining bus travel demand data, subway travel demand data and taxi travel demand data of a region to be predicted; preprocessing the bus travel demand data, the subway travel demand data and the taxi travel demand data to obtain a traffic travel demand feature matrix; and inputting the traffic travel demand feature matrix into a travel demand prediction model, wherein the travel demand prediction model analyzes and processes the traffic travel demand feature matrix to output a prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic planning. More particularly, the present application relates to a cross-modal transferable population travel demand prediction method and device. BACKGROUND

[0002] Population travel demand prediction refers to predicting the travel demand of people in a specific area within a future period of time. The current population travel demand prediction method mostly uses a deep neural network to model a single historical travel time series (such as bus card swiping data, taxi travel data, or shared bicycle usage data) of a specific area, obtains a prediction model, and then uses the prediction model to predict the population travel total data (i.e., the prediction result) of the specific area. However, this method still has shortcomings. Specifically, the formation of the prediction model used by this method relies on single-modal travel data (i.e., one of the bus card swiping data, taxi travel data, or shared bicycle usage data), so that the prediction model can only reflect the travel behavior of a specific group in the specific area and cannot reflect the travel behavior of other groups other than the specific group. As a result, the prediction model has a sampling bias phenomenon, which leads to an inaccurate prediction result of the method and makes it difficult to reflect the population travel demand total. SUMMARY

[0003] The purpose of the embodiments of the present application is to provide a cross-modal transferable population travel demand prediction method and device that can improve the accuracy of population travel demand total. The embodiments of the present application mainly achieve the purpose through the following technical solutions:

[0004] In a first aspect, the embodiments of the present application provide a cross-modal transferable population travel demand prediction method, comprising:

[0005] obtaining bus travel demand data, subway travel demand data, and taxi travel demand data of a to-be-predicted area;

[0006] preprocessing the bus travel demand data, the subway travel demand data, and the taxi travel demand data to obtain a traffic travel demand feature matrix;

[0007] inputting the traffic travel demand feature matrix into a travel demand prediction model, wherein the travel demand prediction model analyzes and processes the traffic travel demand feature matrix to output a prediction result.

[0008] According to an embodiment of the present application, the step of preprocessing the bus travel demand data, the subway travel demand data, and the taxi travel demand data to obtain a traffic travel demand feature matrix comprises:

[0009] The bus trip demand data is subjected to a rejection operation to obtain first to-be-processed data;

[0010] All boarding station information and boarding time information corresponding to each of the boarding station information are extracted from the first to-be-processed data to obtain second to-be-processed data;

[0011] The subway trip demand data is subjected to a rejection operation to obtain third to-be-processed data;

[0012] All station entry position information and station entry time information corresponding to each of the station entry position information are extracted from the third to-be-processed data to obtain fourth to-be-processed data;

[0013] The taxi trip demand data is subjected to a rejection operation to obtain fifth to-be-processed data;

[0014] All departure location information and departure time information corresponding to each of the departure location information are extracted from the fifth to-be-processed data to obtain sixth to-be-processed data;

[0015] A total demand feature matrix is generated based on the second to-be-processed data, the fourth to-be-processed data, and the sixth to-be-processed data;

[0016] The traffic trip demand feature matrix is extracted from the total demand feature matrix in a time window division manner.

[0017] According to an embodiment of the present application, the step of generating the total demand feature matrix based on the second to-be-processed data, the fourth to-be-processed data, and the sixth to-be-processed data comprises:

[0018] A first to-be-processed feature matrix is generated based on the second to-be-processed data, the fourth to-be-processed data, and the sixth to-be-processed data;

[0019] All elements in the first to-be-processed feature matrix are connected in a time dimension to obtain the total demand feature matrix.

[0020] According to an embodiment of the present application, the step of extracting the traffic trip demand feature matrix from the total demand feature matrix in a time window division manner comprises:

[0021] A length of a first sliding window is set;

[0022] A step size of the first sliding window is set;

[0023] The traffic trip demand feature matrix is extracted from the total demand feature matrix based on the length and the step size.

[0024] According to one embodiment of this application, the training steps of the travel demand prediction model include:

[0025] Obtain the training datasets for public transportation, subway, and taxi, as well as the label dataset;

[0026] Preprocessing is performed on the target bus training data, target subway training data, and target taxi training data to obtain a traffic training feature matrix. The target bus training data is any element in the bus training dataset, the target subway training data is an element in the subway training dataset that belongs to the same time series as the target bus training data, and the target taxi training data is an element in the taxi training dataset that belongs to the same time series as the target bus training data.

[0027] The traffic training feature matrix is ​​input into a recurrent neural network model, which analyzes and processes the traffic training feature matrix and outputs the training results.

[0028] Calculate the first loss function based on the training results and the labeled dataset;

[0029] The first model parameters of the recurrent neural network model are adjusted based on the first loss function to form the travel demand prediction model.

[0030] According to one embodiment of this application, the calculation formula for the first loss function is as follows:

[0031] ;

[0032] in, Let the first loss function be... For the first model parameters, for The least dissimilar time series , For the first The length of a time series segment The squared error loss function, The tag data is concentrated in the first... The first segment of the time series The label value at each time step. The training result (i.e., the output of the recurrent neural network model). The traffic training feature matrix is ​​the first... The first segment of the time series The feature vectors at each time step.

[0033] According to one embodiment of this application, after the step of adjusting the first model parameters of the recurrent neural network model based on the first loss function, the training step of the travel demand prediction model further includes:

[0034] Set the importance vector;

[0035] A second loss function is constructed based on the aforementioned importance vector;

[0036] The second model parameters of the travel demand prediction model are adjusted based on the second loss function.

[0037] A second aspect of this application provides a cross-modal transferable population travel demand forecasting device, comprising:

[0038] The acquisition module is used to acquire public transport travel demand data, subway travel demand data, and taxi travel demand data for the area to be predicted.

[0039] The transportation demand feature matrix acquisition module is used to preprocess the public transportation demand data, the subway demand data, and the taxi demand data to obtain the transportation demand feature matrix.

[0040] The prediction result output module is used to input the traffic travel demand feature matrix into the travel demand prediction model, and the travel demand prediction model analyzes and processes the traffic travel demand feature matrix to output the prediction result.

[0041] A third aspect of this application provides a terminal device, including a processor and a memory, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to execute the steps of the cross-modal transferable population travel demand prediction method provided in the first aspect of this application.

[0042] A fourth aspect of this application provides a computer-readable storage medium for storing a computer program that causes a computer to perform the steps of the cross-modal transferable population travel demand forecasting method provided in the first aspect of this application.

[0043] The beneficial effects of the embodiments of this application include:

[0044] This application embodiment uses multimodal transportation demand data to predict the total population travel demand. Specifically, this application embodiment uses public transport travel demand data, subway travel demand data, and taxi travel demand data of the area to be predicted as basic data, and preprocesses this basic data to obtain a transportation demand feature matrix, which is used as input to the travel demand prediction model. After the transportation demand feature matrix is ​​analyzed and processed by the travel demand prediction model, the prediction result is output, which is the total population travel demand data. Therefore, the multimodal transportation demand data in this application embodiment can overcome the sampling bias phenomenon caused by single-modal transportation demand data in the prior art, thereby improving the accuracy of the prediction results, that is, improving the accuracy of predicting the total population travel demand. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 The flowcharts for some embodiments of the cross-modal transferable population travel demand forecasting method of this application are shown below;

[0047] Figure 2 This is a schematic diagram of the cross-modal transferable population travel demand prediction device of this application in some embodiments;

[0048] Figure 3 This is a schematic block diagram of the terminal device of this application in some embodiments. Detailed Implementation

[0049] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application are described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0050] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0051] The terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0052] The terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units that are expressly listed, but may include other steps or units that are not expressly listed or that are inherent to such process, method, product, or apparatus.

[0053] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used in this application includes any and all combinations of one or more of the associated listed items.

[0054] The specific embodiments of this application will be further described below with reference to the accompanying drawings.

[0055] <Cross-modal transferable population travel demand forecasting methods>

[0056] like Figure 1 The diagram shown is a flowchart of a cross-modal transferable population travel demand forecasting method provided in the first aspect of an embodiment of this application. Figure 1 The cross-modal transferable population travel demand forecasting method includes:

[0057] S1. Obtain public transport travel demand data, subway travel demand data, and taxi travel demand data for the area to be predicted.

[0058] The public transportation travel demand data includes boarding station information for all first users in the area to be predicted, boarding time information corresponding to each boarding station, alighting station information, and alighting time information corresponding to each alighting station.

[0059] The first user is a user who takes public transportation in the area to be predicted.

[0060] The public transport travel demand data is derived from public transport card swipe data and GPS data. In other implementations, it can also be obtained from other data sources such as traffic surveys.

[0061] The subway travel demand data includes the subway entry location information of all second users in the area to be predicted, the entry time information corresponding to each entry location information, the exit location information, and the exit time information corresponding to each exit location information.

[0062] The second user is a user who takes the subway in the area to be predicted.

[0063] The subway travel demand data is derived from subway card swipe records. In other implementations, it can also be obtained from other data sources such as traffic surveys.

[0064] The taxi travel demand data includes departure location information for all third-party users taking taxis in the area to be predicted, departure time information corresponding to each departure location information, destination information, arrival time information corresponding to each destination information, and service status information corresponding to the departure time information or the arrival time information.

[0065] The third user is a user who takes a taxi in the area to be predicted.

[0066] The taxi travel demand data is derived from taxi GPS datasets. In other implementations, it can also be obtained from other data sources such as traffic surveys.

[0067] In other embodiments, the data used in this application is not limited to the public transportation travel demand data, the subway travel demand data, and the taxi travel demand data. These data can be changed according to different usage scenarios.

[0068] S2. Preprocess the public transport travel demand data, the subway travel demand data, and the taxi travel demand data to obtain a transportation travel demand feature matrix.

[0069] Further, step S2 includes:

[0070] S21. Perform a filtering operation on the public transportation travel demand data to obtain the first data to be processed.

[0071] Furthermore, the removal operation of the public transport travel demand data refers to removing boarding station information or boarding time information that does not have corresponding boarding time information from the public transport travel demand data.

[0072] S22. Extract all boarding station information and boarding time information corresponding to each boarding station from the first data to be processed to obtain the second data to be processed.

[0073] It should be understood that the second data to be processed only includes all boarding station information and boarding time information corresponding to each boarding station information in the first data to be processed, but does not include alighting station information and alighting time information corresponding to each alighting station information.

[0074] S23. Perform a filtering operation on the subway travel demand data to obtain the third data to be processed.

[0075] Furthermore, the process of removing subway travel demand data refers to removing entry location information or entry time information that does not have corresponding entry time information from the subway travel demand data.

[0076] S24. Extract all entry location information and entry time information corresponding to each entry location information from the third data to be processed to obtain the fourth data to be processed.

[0077] It should be understood that the fourth data to be processed only includes all the entry location information and the entry time information corresponding to each of the entry location information in the third data to be processed, and does not include the exit location information and the exit time information corresponding to each of the exit location information.

[0078] S25. Perform a filtering operation on the taxi travel demand data to obtain the fifth data to be processed.

[0079] Furthermore, the process of removing taxi travel demand data refers to removing departure location information, service status information, and departure location information from the taxi travel demand data that do not have corresponding departure time information.

[0080] Furthermore, this application embodiment employs a state change recognition mechanism to set the service status information. For example, when the taxi is in an idle state (i.e., the taxi does not have the departure time information), the service status information is set to 0; when the taxi is in a service state (i.e., the taxi has the departure time information), the service status information is set to 1. It should be understood that when a taxi trip has not started, it is in an idle state; when a taxi trip has started, it is in a service state.

[0081] S26. Extract all departure location information and departure time information corresponding to each departure location information from the fifth data to be processed to obtain the sixth data to be processed.

[0082] The sixth data to be processed contains only all departure location information and departure time information corresponding to each departure location information in the fifth data to be processed, that is, the departure location and time of each taxi trip.

[0083] S27. Generate a total demand feature matrix based on the second data to be processed, the fourth data to be processed, and the sixth data to be processed.

[0084] Further, step S27 includes:

[0085] S271. Generate a first feature matrix to be processed based on the second data to be processed, the fourth data to be processed, and the sixth data to be processed.

[0086] More specifically, step S271 includes:

[0087] S2711. The area to be predicted is divided into multiple first traffic zones using a spatial partitioning method. In this embodiment, the area to be predicted is divided into 1386 first traffic zones based on the Voronoi diagram of the main roads in the area to be predicted. In other embodiments, the spatial partitioning method is not limited to the above-mentioned method of using a Voronoi diagram for partitioning; those skilled in the art can choose according to actual needs. Furthermore, the number of first traffic zones is not limited to the aforementioned 1386; the specific number of first traffic zones can also be determined by those skilled in the art according to actual needs.

[0088] S2712. Using the GeoPandas library in Python, the second data to be processed, the fourth data to be processed, and the sixth data to be processed are matched to the multiple first traffic zones. Then, the first traffic demand of different modes of transportation in each first traffic zone is statistically analyzed at a time resolution of 1 hour. The first feature matrix to be processed is composed of all the first traffic demand.

[0089] The first feature matrix to be processed can be expressed as the following formula:

[0090] ;

[0091] in, The first feature matrix to be processed. Each element in the formula is... The form of representation, in which Indicates the first sky, Indicates the first Hour, Indicates the first The type of transportation in the first Heavenly The hourly travel demand data is as follows: the second data to be processed belongs to the first mode of transportation, the fourth data to be processed belongs to the second mode of transportation, and the sixth data to be processed belongs to the third mode of transportation. .

[0092] The first feature matrix to be processed is the travel demand feature matrix of various modes of transportation in the region to be predicted. S272, All elements in the first feature matrix to be processed are concatenated along the time dimension to obtain the total demand feature matrix.

[0093] The time dimension is a days dimension. In other embodiments, the time dimension can be set by those skilled in the art according to actual needs.

[0094] S28. The transportation demand feature matrix is ​​obtained by using a time windowing method to extract the total demand feature matrix.

[0095] Further, step S28 includes:

[0096] S281. Set the length of the first sliding window.

[0097] The length of the first sliding window is one hour, one day, or one week. The appropriate length of the first sliding window (i.e., the time window) is selected based on the specific forecasting task (short-term forecasting or long-term forecasting).

[0098] S282. Set the step size of the first sliding window.

[0099] The step size of the first sliding window can be determined by those skilled in the art based on actual needs.

[0100] S283. Extract the transportation demand feature matrix from the total demand feature matrix based on the length and step size of the first sliding window.

[0101] The transportation demand feature matrix is ​​a sequence extracted from the total demand feature matrix. That is, starting from the beginning of the time dimension of the total demand feature matrix, the first sliding window is slid according to the step size to extract the data of the total demand feature matrix corresponding to the current first sliding window.

[0102] S3. Input the traffic travel demand feature matrix into the travel demand prediction model. The travel demand prediction model analyzes and processes the traffic travel demand feature matrix and outputs the prediction result.

[0103] The prediction results are the total population travel data.

[0104] In the above technical solution, the embodiments of this application predict the total population travel demand data by using multimodal traffic travel demand data. Therefore, the embodiments of this application can overcome the sampling bias caused by single-modal traffic travel demand data in the prior art, thereby improving the accuracy of the prediction results, that is, improving the accuracy of predicting the total population travel demand.

[0105] In some implementations, the training steps of the travel demand prediction model include:

[0106] S41. Obtain the bus training dataset, subway training dataset, taxi training dataset, and label dataset.

[0107] Each element in the public transport training dataset has the same data structure and data type as the public transport travel demand data. Similarly, each element in the subway training dataset has the same data structure and data type as the subway travel demand data. Finally, each element in the taxi training dataset has the same data structure and data type as the taxi travel demand data.

[0108] The bus training dataset, the subway training dataset, the taxi training dataset, and the label dataset were all collected from the same research area.

[0109] S42. Preprocess the target bus training data, target subway training data, and target taxi training data to obtain a traffic training feature matrix. The target bus training data is any element in the bus training dataset, the target subway training data is an element in the subway training dataset that belongs to the same time series as the target bus training data, and the target taxi training data is an element in the taxi training dataset that belongs to the same time series as the target bus training data.

[0110] Furthermore, each element in the labeled dataset is a population travel demand data corresponding to the target bus training data, the target subway training data, and the target taxi training data.

[0111] In this embodiment, the population travel demand data corresponding to the target bus training data, the target subway training data, and the target taxi training data are set as the target population travel demand data. The target population travel demand data originates from mobile phone signaling data, which is user location information provided by mobile communication operators. The user location information includes base station handover records and call information of the mobile terminal. In other embodiments, the target population travel demand data can be obtained through traffic surveys or other data sources. The target population travel demand data is one element in the tag dataset.

[0112] Furthermore, the step of preprocessing the target bus training data, target subway training data, and target taxi training data to obtain the traffic training feature matrix includes:

[0113] S421. Perform a removal operation on the target bus training data to obtain the seventh data to be processed.

[0114] Specifically, the removal operation of the target bus training data refers to removing boarding station information or boarding time information that does not have corresponding boarding time information from the target bus training data.

[0115] S422. Extract all boarding station information and boarding time information corresponding to each boarding station from the seventh data to be processed to obtain the eighth data to be processed.

[0116] The eighth data to be processed includes only all boarding station information and boarding time information corresponding to each boarding station information in the seventh data to be processed, but does not include alighting station information and alighting time information corresponding to each alighting station information.

[0117] S423. Perform a removal operation on the target subway training data to obtain the ninth data to be processed.

[0118] Specifically, the removal operation of the target subway training data refers to removing entry location information or entry time information that does not have corresponding entry time information from the target subway training data.

[0119] S424. Extract all entry location information and entry time information corresponding to each entry location information from the ninth data to be processed to obtain the tenth data to be processed.

[0120] The tenth data to be processed includes only all the entry location information and the entry time information corresponding to each entry location information in the ninth data to be processed, but does not include the exit location information and the exit time information corresponding to each exit location information.

[0121] S425. Perform a removal operation on the target taxi training data to obtain the eleventh data to be processed.

[0122] Specifically, the removal operation of the target taxi training data refers to removing departure location information, service status information, and departure location information that do not have corresponding departure time information from the target taxi training data.

[0123] S426. Extract all departure location information and departure time information corresponding to each departure location information from the eleventh data to be processed to obtain the twelfth data to be processed.

[0124] The twelfth set of data to be processed contains only all departure location information and departure time information corresponding to each departure location information from the eleventh set of data to be processed.

[0125] S427. Generate a traffic training demand feature matrix based on the eighth data to be processed, the tenth data to be processed, and the twelfth data to be processed.

[0126] Further, step S427 includes:

[0127] S4271. Generate a second feature matrix to be processed based on the eighth data to be processed, the tenth data to be processed, and the twelfth data to be processed.

[0128] More specifically, step 4271 includes:

[0129] S42711. The study area is divided into multiple second traffic zones using a spatial partitioning method. In this embodiment, the study area is divided into 1386 second traffic zones based on the Voronoi diagram of the main roads in the study area. In other embodiments, the number of second traffic zones is not limited to the aforementioned 1386, and the specific number of second traffic zones can be determined by those skilled in the art according to actual needs. The number of second traffic zones can be the same as the number of first traffic zones.

[0130] S42712. Using the GeoPandas library in Python, the eighth, tenth, and twelfth data to be processed are matched to the multiple second traffic zones. Then, the second traffic demand of different modes of transportation (i.e., buses, subways, and taxis) in each second traffic zone is statistically analyzed at a time resolution of 1 hour. The second traffic demand constitutes the second feature matrix to be processed.

[0131] The formula for the second feature matrix to be processed can be referred to the first feature matrix to be processed. The formula is as follows: S4272. Connect all elements in the second feature matrix to be processed along the time dimension to obtain the traffic training demand feature matrix.

[0132] S428. The traffic training feature matrix is ​​obtained by using a time windowing method to extract the traffic training demand feature matrix.

[0133] Further, step S428 includes:

[0134] S4281, Set the length of the second sliding window.

[0135] The length of the second sliding window can be one hour, one day, or one week. The appropriate length of the second sliding window is selected based on the specific forecasting task.

[0136] S4282. Set the step size of the second sliding window.

[0137] The step size of the second sliding window can be determined by those skilled in the art based on actual needs.

[0138] S4283. Extract the traffic training feature matrix from the traffic training demand feature matrix based on the length and step size of the second sliding window.

[0139] Furthermore, in this embodiment of the application, the travel demand data of the target population also needs to be preprocessed to obtain a population travel demand matrix.

[0140] Specifically, the step of preprocessing the travel demand data of the target population to obtain the population travel demand matrix includes:

[0141] S51. Perform a removal operation on the travel demand data of the target population to obtain the thirteenth data to be processed.

[0142] It should be understood that the removal operation of the target population travel demand data refers to removing mobile signaling data that has switched to different base stations multiple times within a short period of time, as well as mobile signaling data that is not within the study area. In other embodiments, other abnormal signaling records may also be removed.

[0143] S52. Using the GeoPandas library in Python, the thirteenth data to be processed is matched to the multiple second traffic zones. Then, the population travel demand in each second traffic zone is statistically analyzed at a time resolution of 1 hour. The third feature matrix to be processed is formed by all the population travel demand.

[0144] The third feature matrix to be processed can be represented in the following form:

[0145] ;

[0146] in, The third feature matrix to be processed. Each element in the formula is... The form of representation, in which Indicates the first sky, Indicates the first Hour.

[0147] S53. Connect all elements in the third feature matrix to be processed along the time dimension to obtain the population training requirement feature matrix.

[0148] S54. Set the length and step size of the third sliding window. The length and step size of the third sliding window are the same as those of the second sliding window.

[0149] S55. Extract the population travel demand matrix from the population training demand feature matrix based on the length and step size of the third sliding window.

[0150] It should also be understood that when the second sliding window is at the first time step, the elements in the traffic training demand feature matrix corresponding to the first time step are used as the traffic training feature matrix; at the same time, the elements in the population training demand feature matrix corresponding to the third sliding window at the second time step are used as the population travel demand matrix, or the elements in the population training demand feature matrix corresponding to multiple time steps after the first time step of the third sliding window are used as the population travel demand matrix.

[0151] In this embodiment, the traffic training feature matrix is ​​used as the input to the recurrent neural network model, and the population travel demand matrix is ​​used as the output target of the recurrent neural network model. Therefore, the labeled dataset can also be understood as consisting of multiple population travel demand matrices.

[0152] Furthermore, the relationship between the input and output of the recurrent neural network model described in this application embodiment can be expressed by a basic formula as follows: ,in, This is the population travel demand matrix. The parameters of the recurrent neural network model (i.e., the model prediction parameters) are as follows. The traffic training feature matrix is... This refers to the third model parameter of the recurrent neural network model, which is also the first model parameter. In the basic formula, it can be understood that this application predicts the population travel demand matrix using the traffic training feature matrix, that is, it uses multimodal traffic travel demand data to predict the full population travel demand data. This basic formula establishes the relationship between the population travel demand matrix and the traffic training feature matrix, enabling the recurrent neural network model to have high accuracy.

[0153] In other embodiments, the step of preprocessing the target population travel demand data to obtain a population travel demand matrix further includes:

[0154] S61. Sort the base station switching records of each user in the target population travel demand data in chronological order to construct the activity trajectory of each user.

[0155] S62. Identify the dwell time of each user at the target base station, calculate the duration of the continuous signaling record of the user at the target base station, and determine the activity area of ​​the user based on the dwell time and the duration, wherein the target base station is any base station in the base station handover record corresponding to the user.

[0156] S63. Set the dwell threshold.

[0157] S64. If the dwell time exceeds the dwell threshold, the target base station is regarded as point O, and point O is the departure point.

[0158] S65. If the user switches from the target base station to another base station and stays at the other base station for a period of time exceeding the stay threshold, the other base station is regarded as point D.

[0159] Point D represents the end of the travel event.

[0160] S66. Record the timestamp of point O as the departure time and record the timestamp of point D as the arrival time.

[0161] By following steps S51-S55 and S61-S66, we can calculate the departure point and time for each user in each segment of the trip.

[0162] S43. Input the traffic training feature matrix into the recurrent neural network model, and the recurrent neural network model analyzes and processes the traffic training feature matrix to output the training results.

[0163] It should be understood that, in order to overcome the problems of timeliness and transferability of existing models, this application embodiment sets up an optimization model (i.e., an optimization problem). The technical means adopted by the optimization model is to segment time-series data to maximize the differences between the segmented time-series data. Specifically, the optimization model is expressed as:

[0164] ;

[0165] ;

[0166] in, for The least dissimilar time series; These are hyperparameters used to avoid over-segmentation; For the first Time series; For the first Time series; This is a distance metric. For the first The length of a time series segment; For the first The length of a time series segment; and These are predefined parameters used to avoid the optimization model from obtaining cumbersome solutions; This represents the total length of the time series.

[0167] The optimization model can be solved using a greedy algorithm or a dynamic programming algorithm to obtain... The least dissimilar time series segment. The purpose of the optimization model can be simply understood as dividing the initially input data segment into... The segment, that is, the one mentioned The least similar time series.

[0168] Furthermore, in this embodiment, CORAL distance (Deep Correlation Alignment) is used as the distance metric, and the specific calculation formula is as follows:

[0169] ;

[0170] in, For the first The length of a time series segment As the dimension of the feature, For the first The length of a time series segment Let Frobenius norm be the matrix. For the first The covariance matrix of a time series For the first The covariance matrix of a time series.

[0171] In other embodiments, the distance measurement method can be set by those skilled in the art according to actual needs.

[0172] Furthermore, this embodiment of the application uses a gated recurrent unit of a recurrent neural network model for modeling. The calculation formula of the gated recurrent unit is as follows:

[0173] ;

[0174] ;

[0175] ;

[0176] ;

[0177] ;

[0178] in, This is the update gate for the gated loop unit; Use the Sigmoid activation function; This is the weight matrix corresponding to the update gate; This is the hidden state from the previous time step; For the current input, Belonging to the The least dissimilar time series data (used in this application embodiment) The least dissimilar time series data is used as the input to the gated loop unit. This is the reset door for the gated loop unit; This is the weight matrix corresponding to the reset gate; This is the candidate hidden state; It is the hyperbolic tangent activation function; The weight matrix is ​​the one corresponding to the hyperbolic tangent activation function. The symbol for element-wise multiplication; Currently hidden; This refers to the output at that moment, specifically the travel data output at that moment. To and The corresponding weight matrix; To and The corresponding bias term.

[0179] S44. Calculate the first loss function based on the training results and the labeled dataset.

[0180] Furthermore, the formula for calculating the first loss function is as follows:

[0181] ;

[0182] in, Let the first loss function be used. The first model parameters, also known as the undetermined parameters of the model prediction part (this part of the parameters includes the hidden layer parameters of the recurrent neural network). and output layer parameters ); for The least dissimilar time series; ; For the first The length of a time series segment; The squared error loss function; The tag data is concentrated in the first... The first segment of the time series The label value at each time step; The training result is the output of the recurrent neural network model. The traffic training feature matrix is ​​the first... The first segment of the time series The feature vectors at each time step.

[0183] Furthermore, the formula for calculating the squared error loss function is as follows:

[0184] ;

[0185] in, For the first The length of a time series segment The tag data is concentrated in the first... The first segment of the time series The label value at each time step. The training results are as described.

[0186] In other embodiments, the formula for calculating the squared error loss function is not limited to the above formula and can be set by those skilled in the art according to actual needs.

[0187] S45. Adjust the first model parameters of the recurrent neural network model based on the first loss function to form the travel demand prediction model.

[0188] In some implementations, it is assumed that if the model can reduce the above-mentioned... The model exhibits the strongest generalization ability when the distribution differences of the least dissimilar time series are identified. Therefore, after the step of adjusting the first model parameters of the recurrent neural network model based on the first loss function, the training steps of the travel demand prediction model further include:

[0189] S46. Set the importance vector.

[0190] The importance vector is used to describe the relative importance of each hidden state.

[0191] The importance vector is introduced in this application embodiment in order to learn the recurrent neural network (the recurrent neural network model) in... The relative importance of each hidden layer at each time step is determined by normalizing all hidden states. Weighted and minimizing time series pairs The distribution distance of the hidden states in each time series can be calculated to dynamically reduce the distribution distance between time series, so as to learn the shared knowledge of time series with different distributions, thereby effectively solving the timeliness and transferability problems of existing models.

[0192] S47. Construct a second loss function based on the importance vector.

[0193] Furthermore, the formula for calculating the second loss function is as follows:

[0194] ;

[0195] in, For the second loss function, For the second model parameters, Let be the importance vector. for dimensional vectors (i.e.) (Importance weight of hidden layer at each time point) For the first loss function, For the undetermined parameters in the model prediction, As a weighing factor, The average distribution distance between each time segment, for The least dissimilar time series For the first A time series, For the first A time series, For the third loss function, For the first The length of a time series segment For the first The length of a time series segment These are the parameters for the fourth model.

[0196] Furthermore, the formula for calculating the third loss function is as follows:

[0197] ;

[0198] in, for At that moment, , Time sequence and exist The importance of the distribution of time points. This is a distance metric. Time sequence exist The hidden state at all times Time sequence exist The hidden state at all times For the fourth model parameters, The importance vector is defined as follows. The fourth model parameter is a part of the undetermined parameters for the model transfer regularization term, which refers to the parameters in the time-series distribution matching process. This part of the parameters includes the hidden layer parameters of the recurrent neural network adjusted by the first loss function. Parameters in the distance measurement process .

[0199] S48. Adjust the second model parameters of the travel demand prediction model based on the second loss function.

[0200] In some implementations, the hidden layers learn weak representations early in training, which leads to weak importance vectors. Because it cannot learn effectively, this application embodiment includes a pre-training stage, which uses data from all time periods to learn the model parameters. To promote The learning.

[0201] Specifically, promoting the importance vector The learning steps include:

[0202] S71. Set up a pre-training round to obtain parameters. .

[0203] S72, Based on the parameters The importance vector is learned using the following Boosting process. It should be understood that initially, for each layer of a recurrent neural network, all weights are initialized to the same value, i.e. , For the first Time series and the first A time series in The relative importance of each moment.

[0204] S73. If the distribution distance in the (n+1)th round shows an increasing trend compared to the distribution distance in the nth round, then the corresponding importance vector is increased. This reduces the diversity of distributions and promotes knowledge sharing among different distribution time segments.

[0205] In some implementations, the Boosting process can be described as follows:

[0206] ;

[0207] in, ;

[0208] in, ;

[0209] in, ;

[0210] in, For the first In the wheel That is, the first Cycle sequence and exist The importance of the distribution of time points; For the first In the wheel That is, the first Cycle sequence and exist The importance of the distribution of time points; Use the Sigmoid activation function; For the first Wheel in Time of the first Time series and the first The distribution distance of hidden layers in a time series; As a cross-domain distribution distance measurement method, the embodiments of this application adopt a domain adversarial difference measurement method; For domain discriminator; Time sequence exist The hidden state at any given moment; Time sequence exist The hidden state at any given moment; For the first Time sequence The losses incurred in combating situations where the state is constantly hidden; For the first Time sequence The constant concealment of the state of the opponent's losses.

[0211] The travel demand prediction model provided in this application is applicable to regions with rapidly changing travel demand and significant spatial heterogeneity, thereby improving the model's applicability and reducing its training cost.

[0212] <Cross-modal transferable population travel demand forecasting device>

[0213] like Figure 2 The diagram shown illustrates the principle block diagram of a cross-modal transferable population travel demand prediction device provided in the second aspect of this application. Figure 2 The cross-modal transferable population travel demand prediction device 100 includes:

[0214] The acquisition module 101 is used to acquire public transport travel demand data, subway travel demand data, and taxi travel demand data for the area to be predicted.

[0215] The transportation demand feature matrix acquisition module 102 is used to preprocess the public transport demand data, the subway travel demand data and the taxi travel demand data to obtain the transportation demand feature matrix.

[0216] The prediction result output module 103 is used to input the traffic travel demand feature matrix into the travel demand prediction model, and the travel demand prediction model analyzes and processes the traffic travel demand feature matrix and outputs the prediction result.

[0217] Terminal Equipment

[0218] A third aspect of this application provides a terminal device, the schematic diagram of which is as follows: Figure 3As shown. The terminal device includes a processor, memory, network interface, display screen, and temperature sensor connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the terminal device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a cross-modal transferable population travel demand forecasting method. The display screen can be a liquid crystal display screen or an e-ink display screen, and the temperature sensor is pre-installed inside the terminal device to detect the operating temperature of the internal components.

[0219] Those skilled in the art will understand that Figure 3 The schematic diagram shown is only a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0220] In some embodiments, this application provides a terminal device including a processor and a memory for storing computer programs. The processor is used to call and run the computer programs stored in the memory to perform the steps of the cross-modal transferable population travel demand forecasting method provided in the first aspect of this application.

[0221] Computer-readable storage media

[0222] A fourth aspect of this application provides a computer-readable storage medium for storing a computer program that causes a computer to perform the steps of the cross-modal transferable population travel demand forecasting method provided in the first aspect of this application.

[0223] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0224] The technical features of the above embodiments can be combined without changing the basic principles of this application. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0225] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the patent protection scope of this application should be determined by the appended claims.

Claims

1. A cross-modal transferable method for predicting population travel demand, characterized in that, include: Obtain public transport travel demand data, subway travel demand data, and taxi travel demand data for the area to be predicted; The public transport travel demand data, the subway travel demand data, and the taxi travel demand data are preprocessed to obtain a transportation travel demand feature matrix. The traffic demand feature matrix is ​​input into the travel demand prediction model, which analyzes and processes the traffic demand feature matrix and outputs the prediction result. The training steps of the travel demand prediction model include: acquiring a public transport training dataset, a subway training dataset, a taxi training dataset, and a label dataset; preprocessing the target public transport training data, target subway training data, and target taxi training data to obtain a traffic training feature matrix, wherein the target public transport training data is any element in the public transport training dataset, the target subway training data is an element in the subway training dataset belonging to the same time series as the target public transport training data, and the target taxi training data is an element in the taxi training dataset belonging to the same time series as the target public transport training data; inputting the traffic training feature matrix into a recurrent neural network model, the recurrent neural network model analyzing and processing the traffic training feature matrix and outputting training results; calculating a first loss function based on the training results and the label dataset; and adjusting the first model parameters of the recurrent neural network model based on the first loss function to form the travel demand prediction model. After the step of adjusting the first model parameters of the recurrent neural network model based on the first loss function, the training step of the travel demand prediction model further includes: setting an importance vector; constructing a second loss function based on the importance vector; and adjusting the second model parameters of the travel demand prediction model based on the second loss function. The formula for calculating the first loss function is: ； ； The formula for calculating the second loss function is: ； The formula for calculating the third loss function is: ； in, Let the first loss function be... The first model parameters are undetermined parameters in the model prediction part, and the undetermined parameters in the model prediction part include the hidden layer parameters of the recurrent neural network. and output layer parameters , for The least dissimilar time series , For the first The length of a time series segment The squared error loss function, The tag data is concentrated in the first... The first segment of the time series The label value at each time step. The training results are as described. The traffic training feature matrix is ​​the first... The first segment of the time series Feature vectors at each time step; For the second loss function, For the second model parameters, Let be the importance vector. for A dimensional vector is also The importance weight of the hidden layer at each time point As a weighing factor, The average distribution distance between each time segment, For the first A time series, For the first A time series, For the third loss function, For the first The length of a time series segment For the fourth model parameters; for At that moment, , Time sequence and exist The importance of the distribution of time points. This is a distance metric. Time sequence exist The hidden state at all times Time sequence exist The hidden state at time step, the fourth model parameter is a part of the undetermined parameters of the model transfer regularization term, the part of the undetermined parameters of the model transfer regularization term is a parameter in the temporal distribution matching process, the part of the undetermined parameters of the model transfer regularization term includes the parameters of the hidden layer of the recurrent neural network adjusted by the first loss function. Parameters in the distance measurement process .

2. The cross-modal transferable population travel demand forecasting method according to claim 1, characterized in that, The step of preprocessing the public transport travel demand data, the subway travel demand data, and the taxi travel demand data to obtain a transportation travel demand feature matrix includes: The public transportation travel demand data is filtered out to obtain the first data to be processed; Extract all boarding station information and boarding time information corresponding to each boarding station from the first data to be processed to obtain the second data to be processed; The subway travel demand data is filtered out to obtain the third set of data to be processed. Extract all entry location information and entry time information corresponding to each entry location information from the third data to be processed to obtain the fourth data to be processed. The taxi travel demand data is filtered out to obtain the fifth set of data to be processed; Extract all departure location information and departure time information corresponding to each departure location information from the fifth data to be processed to obtain the sixth data to be processed. A total demand feature matrix is ​​generated based on the second data to be processed, the fourth data to be processed, and the sixth data to be processed; The transportation travel demand feature matrix is ​​obtained by using a time windowing method to extract it from the total demand feature matrix.

3. The cross-modal transferable population travel demand forecasting method according to claim 2, characterized in that, The step of generating the total demand feature matrix based on the second data to be processed, the fourth data to be processed, and the sixth data to be processed includes: A first feature matrix to be processed is generated based on the second data to be processed, the fourth data to be processed, and the sixth data to be processed. All elements in the first feature matrix to be processed are concatenated along the time dimension to obtain the total demand feature matrix.

4. The cross-modal transferable population travel demand forecasting method according to claim 2, characterized in that, The step of extracting the transportation demand feature matrix from the total demand feature matrix using a time windowing method includes: Set the length of the first sliding window; Set the step size of the first sliding window; The transportation demand feature matrix is ​​extracted from the total demand feature matrix based on the length and the step size.

5. A cross-modal transferable population travel demand forecasting device, characterized in that, include: The acquisition module is used to acquire public transport travel demand data, subway travel demand data, and taxi travel demand data for the area to be predicted. The transportation demand feature matrix acquisition module is used to preprocess the public transportation demand data, the subway demand data, and the taxi demand data to obtain the transportation demand feature matrix. The prediction result output module is used to input the traffic travel demand feature matrix into the travel demand prediction model, and the travel demand prediction model analyzes and processes the traffic travel demand feature matrix and outputs the prediction result. The cross-modal transferable population travel demand prediction device is further used to acquire public transport training datasets, subway training datasets, taxi training datasets, and label datasets; preprocess the target public transport training data, target subway training data, and target taxi training data to obtain a traffic training feature matrix, wherein the target public transport training data is any element in the public transport training dataset, the target subway training data is an element in the subway training dataset belonging to the same time series as the target public transport training data, and the target taxi training data is an element in the taxi training dataset belonging to the same time series as the target public transport training data; input the traffic training feature matrix into a recurrent neural network model, the recurrent neural network model analyzes and processes the traffic training feature matrix, and outputs training results; calculate a first loss function based on the training results and the label dataset; adjust the first model parameters of the recurrent neural network model based on the first loss function to form the travel demand prediction model; The cross-modal transferable population travel demand prediction device is also used to set an importance vector; construct a second loss function based on the importance vector; and adjust the second model parameters of the travel demand prediction model based on the second loss function. The formula for calculating the first loss function is: ； ； The formula for calculating the second loss function is: ； The formula for calculating the third loss function is: ； in, Let the first loss function be... The first model parameters are undetermined parameters in the model prediction part, and the undetermined parameters in the model prediction part include the hidden layer parameters of the recurrent neural network. and output layer parameters , for The least dissimilar time series , For the first The length of a time series segment The squared error loss function, The tag data is concentrated in the first... The first segment of the time series The label value at each time step. The training results are as described. The traffic training feature matrix is ​​the first... The first segment of the time series Feature vectors at each time step; For the second loss function, For the second model parameters, Let be the importance vector. for A dimensional vector is also The importance weight of the hidden layer at each time point As a weighing factor, The average distribution distance between each time segment, For the first A time series, For the first A time series, For the third loss function, For the first The length of a time series segment For the fourth model parameters; for At that moment, , Time sequence and exist The importance of the distribution of time points. This is a distance metric. Time sequence exist The hidden state at all times Time sequence exist The hidden state at time step, the fourth model parameter is a part of the undetermined parameters of the model transfer regularization term, the part of the undetermined parameters of the model transfer regularization term is a parameter in the temporal distribution matching process, the part of the undetermined parameters of the model transfer regularization term includes the parameters of the hidden layer of the recurrent neural network adjusted by the first loss function. Parameters in the distance measurement process .

6. A terminal device, characterized in that, include: A processor and a memory, the memory for storing a computer program, the processor for calling and running the computer program stored in the memory to perform the steps of the cross-modal transferable population travel demand forecasting method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, Used to store a computer program that causes a computer to perform the steps of the cross-modal transferable population travel demand forecasting method according to any one of claims 1 to 4.

Images