Travel demand prediction method and device and electronic equipment

Abstract

The invention provides a travel demand prediction method and device and electronic equipment, and the method comprises the steps: determining a target region of a to-be-predicted travel demand and a future time period, and enabling the target region to comprise a plurality of grids; according to the future time period, determining a plurality of historical time periods corresponding to the future time period; acquiring grid information of each grid in the target area, wherein the grid information comprises a historical travel order issuing amount and a historical order issuing response rate of the grid in each historical time period; according to the plurality of time-space relationship processing modules, processing the plurality of pieces of grid information to obtain time-space features associated with the plurality of pieces of grid information output by each time-space relationship processing module; and processing the plurality of time-space features according to a multi-task model to obtain a predicted travel order issuing amount and a predicted order issuing response rate of each grid in the target area in the future time period. And the travel demand prediction accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of transportation information technology, and in particular, to a travel demand prediction method, apparatus, and electronic device. Background Art

[0002] In the field of shared travel, the way based on online car-hailing travel is an important part. In order to avoid the imbalance between the supply and demand of online car-hailing in the city, it is particularly important to predict the travel demand of users in the city.

[0003] In related technologies, an electronic device can obtain the travel order volume in a certain period of time in a part of the city. The electronic device can process the data of the travel order volume based on a given time slice, and then can predict the travel order volume in a future period in a part of the city based on the sliced data. However, high-frequency time slices and uneven spatio-temporal distribution of demand will cause data sparsity, resulting in low prediction accuracy of travel demand. Summary of the Invention

[0004] This application provides a travel demand prediction method, apparatus, and electronic device, which are used to solve the technical problem of low prediction accuracy of travel demand in the prior art.

[0005] In a first aspect, this application provides a travel demand prediction method, which includes:

[0006] Determine a target area and a future period for the travel demand to be predicted, where the target area includes a plurality of grids;

[0007] According to the future period, determine a plurality of historical periods corresponding to the future period;

[0008] Obtain the grid information of each grid in the target area, where the grid information includes the historical travel order volume and historical order response rate of the grid in each historical period;

[0009] Process a plurality of grid information according to a plurality of spatio-temporal relationship processing modules to obtain spatio-temporal features associated with the plurality of grid information output by each spatio-temporal relationship processing module;

[0010] Process a plurality of spatio-temporal features according to a multi-task model to obtain the predicted travel order volume and predicted order response rate of each grid in the target area in the future period.

[0011] In a second aspect, this application provides a travel demand prediction apparatus, which includes a first determination module, a second determination module, an acquisition module, and a processing module, where:

[0012] The first determination module is configured to determine a target area and a future time period for the travel demand to be predicted, where the target area includes a plurality of grids;

[0013] The second determination module is configured to determine a plurality of historical time periods corresponding to the future time period according to the future time period;

[0014] The obtaining module is configured to obtain grid information of each grid in the target area, where the grid information includes the historical travel order issuance volume and the historical order response rate of the grid in each historical time period;

[0015] The processing module is configured to process a plurality of grid information according to a plurality of spatio-temporal relationship processing modules to obtain spatio-temporal features associated with the plurality of grid information output by each spatio-temporal relationship processing module;

[0016] The processing module is further configured to process a plurality of spatio-temporal features according to a multi-task model to obtain the predicted travel order issuance volume and the predicted order response rate of each grid in the target area in the future time period.

[0017] In a third aspect, the present application provides an electronic device, including: a processor and a memory;

[0018] The memory stores computer execution instructions;

[0019] The processor executes the computer execution instructions stored in the memory, so that the processor executes the travel demand prediction method as described in the first aspect and various possible aspects related to the first aspect.

[0020] In a fourth aspect, the present application provides a computer-readable storage medium, where computer execution instructions are stored in the computer-readable storage medium, and when the computer execution instructions are executed by a processor, the travel demand prediction method as described in the first aspect and various possible aspects related to the first aspect is executed.

[0021] The present application provides a travel demand prediction method, apparatus, and electronic device. The electronic device can determine a target area and a future time period for which travel demand is to be predicted. Among them, the target area may include multiple grids. According to the future time period, multiple historical time periods corresponding to the future time period are determined, and grid information of each grid in the target area is obtained. The grid information includes the historical travel order issuance volume and the historical order response rate of the grid in each historical time period. According to multiple spatio-temporal relationship processing modules, the multiple grid information is processed to obtain spatio-temporal features associated with the multiple grid information output by each spatio-temporal relationship processing module. According to a multi-task model, the multiple spatio-temporal features are processed to obtain the predicted travel order issuance volume and the predicted order response rate of each grid in the target area during the future time period. In the above method, since the electronic device can extract spatio-temporal features from multiple grid information, and the spatio-temporal features can indicate the relationship between multiple grid information on the spatial scale and the relationship between multiple grid information on the time scale, the electronic device can integrate multiple spatio-temporal features related to the target task prediction based on the multi-task model to predict the travel demand during the future time period, thereby improving the accuracy of travel demand prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 FIG. is a schematic diagram of an application scenario provided by an embodiment of the present application;

[0023] Figure 2 FIG. is a schematic flowchart of a travel demand prediction method provided by an embodiment of the present application;

[0024] Figure 3 FIG. is a schematic diagram of a target area provided by an embodiment of the present application;

[0025] Figure 4 FIG. is a schematic diagram of multiple historical time periods provided by an embodiment of the present disclosure;

[0026] Figure 5 FIG. is a schematic structural diagram of a spatio-temporal relationship processing module provided by an embodiment of the present application;

[0027] Figure 6 FIG. is a schematic diagram of a process for determining a graph convolutional neural network established based on distance proximity relationship provided by an embodiment of the present application;

[0028] Figure 7 FIG. is a schematic diagram of a process for determining a graph convolutional neural network established based on functional similarity relationship provided by an embodiment of the present application;

[0029] Figure 8 FIG. is a schematic diagram of a process for determining spatio-temporal features provided by an embodiment of the present application;

[0030] Figure 9A structural schematic diagram of a multi-task model provided by an embodiment of the present application;

[0031] Figure 10 A schematic diagram of a method for determining a predicted travel order volume and a predicted order response rate provided by an embodiment of the present application;

[0032] Figure 11 A structural schematic diagram of a gating network provided by an embodiment of the present application;

[0033] Figure 12 A process schematic diagram of a travel demand prediction method provided by an embodiment of the present application;

[0034] Figure 13 A structural schematic diagram of a travel demand prediction device provided by an embodiment of the present application;

[0035] Figure 14 A hardware structural schematic diagram of an electronic device provided by the present application. Detailed implementation manners

[0036] Here, exemplary embodiments will be described in detail, and examples thereof are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0037] It should be noted that in this document, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the presence of additional identical elements in the process, method, article or device including that element.

[0038] The user information (including but not limited to user device information such as location information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present disclosure are all information and data that have been authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions, and corresponding operation entrances are provided for the user to select authorization or rejection.

[0039] For ease of understanding, some concepts related to the embodiments of the present application will be described below.

[0040] Trip order issuance volume: The trip order issuance volume within a region can indicate the number of online car-hailing orders applied for by users within the region. For example, if the number of online car-hailing orders applied for by multiple users within a region in a day is 100, the electronic device can determine that the trip order issuance volume within this region in a day is 100; if the number of online car-hailing orders applied for by multiple users within a region in the past week is 1000, the electronic device can determine that the trip order issuance volume within this region in the past week is 1000.

[0041] Order response rate: The order response rate within a region can be the proportion of the online car-hailing orders responded to within the region in the trip order issuance volume within this region. For example, if the online car-hailing order applied for by a user is matched with an online car-hailing vehicle and the driver of the online car-hailing vehicle has responded to this order, the electronic device can determine that this online car-hailing order is an order that has been responded to. For example, if the trip order issuance volume within a region is 1000 and the number of online car-hailing orders that have been responded to is 500, the electronic device can determine that the order response rate within this region is 50%.

[0042] In the related art, the online car-hailing travel mode is an important part of the shared travel field. In order to avoid the imbalance between the supply and demand of online car-hailing in the city, it is particularly important to predict the travel demand of users in the city. For example, if the electronic device predicts that the trip order issuance volume in Region 1 in the next hour is 100 and the trip order issuance volume in Region 2 in the next hour is 10, the shared travel platform can increase the number of online car-hailing vehicles in Region 1 and reduce the number of online car-hailing vehicles in Region 2 after 1 hour, which can avoid the imbalance between the supply and demand of online car-hailing in Region 1 and Region 2 and improve the user experience. Currently, the electronic device can predict the travel demand of users in a region within a future time period according to a pre-trained travel demand prediction model. For example, the pre-trained travel demand prediction model can be a model constructed based on a convolutional neural network. The electronic device can input the trip order issuance volume in the past 1 hour within the region into the travel prediction model, and the travel prediction model can predict the trip order issuance volume in this region 1 hour later based on the trip order issuance volume in the past 1 hour. However, there are many influencing factors for travel demand, and a single-task historical sequence cannot accurately predict the travel demand in a future time period, resulting in a low prediction accuracy of travel demand.

[0043] To solve the technical problems in the related art, an embodiment of the present application provides a travel demand prediction method. An electronic device can determine a target area and a future time period for which the travel demand is to be predicted. Among them, the target area can include multiple grids. According to the future time period, multiple historical time periods corresponding to the future time period are determined. The grid information of each grid in the target area is obtained. The grid information can include the historical travel order issuance volume and the historical order response rate of the grid in multiple historical time periods. According to multiple spatio-temporal relationship processing modules, the multiple grid information is processed to obtain spatio-temporal features associated with the multiple grid information output by each spatio-temporal relationship processing module. The electronic device can determine the target model parameters corresponding to the target area, and replace the model parameters of the multi-task model with the target model parameters to obtain a target multi-task model. And according to the target multi-task model, the multiple spatio-temporal features are processed to obtain the predicted travel order issuance volume and the predicted order response rate. In the above method, since the electronic device can predict the travel demand in the future time period based on multiple spatio-temporal features, and the spatio-temporal features can accurately describe the relationship between multiple grid information on the spatial scale and the relationship between multiple grid information on the time scale, the electronic device can accurately predict the travel demand in the future time period by combining the multi-task model, improving the prediction accuracy of the travel demand.

[0044] Next, in combination with Figure 1 , the application scenario of the embodiment of the present application will be described.

[0045] Figure 1 FIG. is a schematic diagram of an application scenario provided by an embodiment of the present application. Please refer to Figure 1 , which includes a target area and an electronic device. Among them, the target area can include Grid 1, Grid 2, Grid 3, and Grid 4. The electronic device can obtain the grid information of multiple grids and extract the spatio-temporal features between the multiple grid information. Furthermore, according to the multiple spatio-temporal features, the travel demand A of Grid 1 in the future time period, the travel demand B of Grid 2 in the future time period, the travel demand C of Grid 3 in the future time period, and the travel demand D of Grid 4 in the future time period can be accurately predicted. In this way, since the electronic device can extract the spatio-temporal features between multiple grid information, the electronic device can accurately predict the travel demand in the future time period by combining the multi-task model, and thus can improve the prediction accuracy of the travel demand.

[0046] It should be noted that Figure 1 is only an example to illustrate the application scenario of the embodiment of the present application, and is not a limitation on the application scenario of the embodiment of the present application.

[0047] The following uses specific embodiments to elaborate in detail on the technical solution of this application and how the technical solution of this application solves the above technical problems. The following several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0048] Figure 2 It is a schematic flowchart of a travel demand prediction method provided by an embodiment of this application. Please refer to Figure 2 The method flow includes:

[0049] S201. Determine the target area and future time period for the travel demand to be predicted.

[0050] The execution subject of the embodiment of this application can be an electronic device or a travel demand prediction device set in the electronic device. Among them, the travel demand prediction device can be implemented based on software, and the travel demand prediction device can also be implemented based on the combination of software and hardware. This application does not limit this. Optionally, the electronic device can be any device with end-side computing capabilities. For example, the electronic device can be a server, a computer, a mobile device, etc. This application does not limit this.

[0051] Among them, the target area can be the area where the travel demand is to be predicted. For example, if predicting the travel demand in city A in the future time period, the target area can be city A; if predicting the travel demand in city B in the future time period, the target area can be city B.

[0052] Optionally, the future time period can be any time period in the future. For example, the future time period can be the time period from 1 hour to 2 hours in the future, or the time period from 1 day to 2 days in the future. For example, the future time period can be within 10 minutes in the future, within 1 hour in the future, within 1 day in the future, within 1 week in the future, etc. This application does not limit this.

[0053] Among them, the target area can include multiple grids. For example, the target area can be divided into multiple grids, and the size of each grid can be the same. Optionally, the shape of the grid can be a triangle. For example, the target area can include multiple triangular grids of the same size. Optionally, the shape of the grid can be a square. For example, the target area can include multiple square grids of the same size. Optionally, the shape of the grid can be a hexagon. For example, the target area can include multiple hexagonal grids of the same size.

[0054] It should be noted that the shape of the grid in the target area can be any regular or irregular figure, and this application does not limit this.

[0055] Optionally, the electronic device may divide the target area into multiple grids based on a preset grid division method. For example, the electronic device may divide the target area into multiple hexagonal grids of the same size based on the hexagonal hierarchical indexing grid system (i.e., the H3 grid system). For example, when the electronic device divides the target area based on the H3 grid system, the sizes of the grids are different at different grid levels.

[0056] Next, in conjunction with Figure 3 , the target area will be described.

[0057] Figure 3 FIG. is a schematic diagram of a target area provided by an embodiment of the present application. Please refer to Figure 3 , which includes a target area. Among them, the target area is an area after being divided based on the H3 grid. The target area may include Grid 1, Grid 2, Grid 3, Grid 4, Grid 5, Grid 6, and Grid 7. Each grid is a hexagonal shape of the same size. It should be noted that other parts in the target area may also be filled with hexagonal grids, which are not shown in the embodiments of the present application for the sake of easy understanding.

[0058] Optionally, if the H3 grid level in the embodiment shown in Figure 3 is changed, the number of grids in the target area will also change. For example, when the electronic device processes the target area based on different H3 grid levels (such as level 6 or level 7, etc.), the sizes of the hexagonal grids in the target area are different.

[0059] It should be noted that in the actual application process, the electronic device may pre-process the grid division of multiple areas and store the areas after the grid division of multiple areas in the database. For example, the database may include multiple areas of a map, and each area may be an area divided based on the H3 grid system. In this way, the electronic device can obtain the areas that have been grid-divided in the database, improving the efficiency of predicting travel demand. For example, the database may include Area A1 and Area A2 corresponding to Area A, and Area B1 and Area B2 corresponding to Area B. Among them, Area A1 and Area B1 may be areas divided based on the 6th-level H3 grid, and Area A2 and Area B2 may be areas divided based on the 7th-level H3 grid.

[0060] Optionally, the electronic device may determine the target area and the future time period based on the travel prediction request. For example, the travel prediction request may include a region identifier and a time period. The electronic device may receive the travel prediction request sent by other devices, and determine the region indicated by the region identifier in the travel prediction request as the target area, and determine the time period in the travel prediction request as the future time period.

[0061] Optionally, the electronic device may determine the target area and the future time period based on any feasible implementation manner (for example, periodically determine the future time period), and the embodiments of the present application do not limit this.

[0062] S202. Determine multiple historical time periods corresponding to the future time period according to the future time period.

[0063] Optionally, the multiple historical time periods corresponding to the future time period may include at least one of the following: multiple time slices before the same time one week ago, multiple time slices before the same time one day ago, and multiple time slices before the current moment.

[0064] It should be noted that a time slice may be any period of time. For example, one time slice may be a duration of 5 minutes, or one time slice may also be a duration of 1 hour, etc., and the embodiments of the present application do not limit this.

[0065] It should be noted that the above one week and one day are only examples of the embodiments of the present disclosure. The multiple historical time periods may also include multiple time slices before the same time two weeks ago, multiple time slices before the same time one month ago, etc., and the embodiments of the present disclosure do not limit this.

[0066] For example, if the current moment is 9:00 am, the future time period is 10:00 am - 11:00 am, and one time slice is 5 minutes, then the multiple historical time periods corresponding to this future time period may be: 12 time slices before 10:00 am one week ago (multiple time slices before the same time one week ago), 12 time slices before 10:00 am one day ago (multiple time slices before the same time one day ago), and 12 time slices before 9:00 am (multiple time slices before the current moment).

[0067] In this way, based on the future time period, the electronic device can obtain multiple historical time periods corresponding to the future time period, where each historical time period may include one time slice or multiple time slices, and the embodiments of the present application do not limit this.

[0068] Next, in combination with Figure 4 , the process of determining multiple historical time periods corresponding to the future time period will be described.

[0069] Figure 4 This is a schematic diagram of multiple historical time periods provided by the embodiments of the present disclosure. Please refer to Figure 4, including: a timeline. Among them, the timeline includes a future period, the current moment, historical period 1, historical period 2, and historical period 3. Among them, historical period 1 is a period of time before the current moment, historical period 2 is a period of time one day before the future period, and historical period 3 is a period of time one week before the future period. Optionally, historical period 1, historical period 2, and historical period 3 may include multiple time slices. In the actual application process, each time slice may also be a historical period corresponding to the future period. In this way, by subdividing the historical period, the travel demand in the future period can be accurately predicted.

[0070] S203. Obtain the grid information of each grid in the target area.

[0071] Among them, the grid information may include the historical travel order volume and the historical order response rate of the grid in each historical period.

[0072] Optionally, the historical travel order volume may be the travel order volume of the grid in the target area during the historical period. For example, in the Figure 4 illustrated embodiment, if the travel order volume in the grid during historical period 1 is 10, then the electronic device may determine that the historical travel order volume of the grid corresponding to historical period 1 is 10. If the travel order volume in the grid during historical period 2 is 30, then the electronic device may determine that the historical travel order volume of the grid corresponding to historical period 2 is 30. If the travel order volume in the grid during historical period 3 is 50, then the electronic device may determine that the historical travel order volume of the grid corresponding to historical period 3 is 50.

[0073] Optionally, the historical order response rate may be the order response rate of the grid in the target area during the historical period. For example, in the Figure 4 illustrated embodiment, if the travel order volume in the grid during historical period 1 is 10, and among them, 4 travel orders are responded, then the electronic device may determine that the historical response rate of the grid corresponding to historical period 1 is 40%. If the travel order volume in the grid during historical period 2 is 20, and among them, 14 travel orders are responded, then the electronic device may determine that the historical response rate of the grid corresponding to historical period 2 is 70%. If the travel order volume in the grid during historical period 3 is 30, and each travel order is responded, then the electronic device may determine that the historical response rate of the grid corresponding to historical period 3 is 100%.

[0074] Optionally, each historical period can be one time slice. Therefore, the electronic device can determine the historical trip order volume and historical order response rate corresponding to each time slice. For example, if the historical period determined by the electronic device can be 24 time slices before the current moment, and each time slice is 5 minutes, the electronic device can obtain 24 sets of historical trip order volume and historical order response rate. Among them, each set of historical trip order volume and historical order response rate can be the trip order volume and order response rate of the grid within 5 minutes (such as the first 5 minutes before the current moment, the first 5 minutes - the first 10 minutes before the current moment, or the first 10 minutes - the first 15 minutes before the current moment, etc.).

[0075] Optionally, the electronic device can obtain the historical trip order volume and historical order response rate within each historical period in the database, or can also obtain the historical trip order volume and historical order response rate within each historical period based on any other feasible implementation method. The embodiments of the present application do not limit this.

[0076] Optionally, the grid information can include weather information and point of interest information. Among them, the weather information of the grid can be the weather information at the current moment within the grid. For example, the electronic device can obtain the actual geographical location corresponding to the grid and the weather information at the current moment, and determine the weather information at the current moment as the weather information of the grid. For example, if the actual geographical location corresponding to the grid is rainy at the current moment, the weather information of the grid can include rainy.

[0077] Optionally, the weather information of the grid can be the weather information within the grid in a future period. For example, the electronic device can obtain the actual geographical location corresponding to the grid and the weather information in a future period, and determine the weather information in the future period as the weather information of the grid. For example, if the actual geographical location corresponding to the grid is sunny within a future period, the weather information of the grid can include sunny.

[0078] Optionally, the weather information of the grid can be the weather information of the grid within a historical period. For example, in Figure 4 the shown embodiment, the electronic device can obtain the weather information of the actual geographical location corresponding to the grid in historical period 1, and determine the weather information of historical period 1 as the weather information of the grid. For example, in Figure 4 the shown embodiment, if the actual geographical location corresponding to the grid is rainy within historical period 2, the weather information of the grid can include rainy.

[0079] Optionally, the electronic device can obtain the weather information of multiple grids in the target area based on weather data (such as weather data provided by a weather application, etc.). The electronic device can also determine the weather information of each grid in the target area based on any other feasible implementation method. The embodiments of the present application do not limit this.

[0080] It should be noted that the weather information of the grid may also include information such as humidity, temperature, wind power, etc., which is not limited in the embodiments of the present application.

[0081] Optionally, the point-of-interest information may include the types of points of interest included in the grid and the quantity of each type of point of interest. For example, the types of points of interest (POI) may include gas station type, shopping mall type, school type, etc., which is not limited in the embodiments of the present application. For example, if there are 10 gas stations and 5 shopping malls in the grid, the electronic device may determine that the quantity of points of interest of the gas station type in the grid is 10, and the quantity of points of interest of the shopping mall type is 5. For example, if there are 10 points of interest in the grid, among which 7 points of interest are schools and 3 points of interest are residential buildings, the electronic device may determine that the proportion of points of interest of the school type in the grid is 70%, and the proportion of points of interest of the residential building type in the grid is 30%.

[0082] Optionally, the electronic device may obtain the point-of-interest information in the grid based on map data. For example, the electronic device may obtain the map data in the target area (the map data may include multiple points of interest), and determine the point-of-interest information of each grid in the target area based on the map data. Optionally, the electronic device may also obtain the point-of-interest information of each grid in the target area based on any other feasible implementation manner, which is not limited in the embodiments of the present application.

[0083] Optionally, the grid information may also include the pedestrian flow information in the grid, the traffic information in the grid, large-scale activities in the grid, etc., which is not limited in the embodiments of the present application.

[0084] S204. Process the multiple grid information according to multiple spatio-temporal relationship processing modules to obtain the spatio-temporal features associated with the multiple grid information output by each spatio-temporal relationship processing module.

[0085] Among them, the spatio-temporal relationship processing module is used to extract the spatio-temporal features between the multiple grid information. Among them, the spatio-temporal relationship processing module may include multiple graph convolutional neural networks and long short-term memory networks (LSTM) with an attention mechanism.

[0086] Among them, the spatio-temporal features can be used to indicate the spatial relationship between multiple grid information and the temporal relationship between multiple grid information. The electronic device can determine the spatio-temporal features associated with the multiple grid information output by each spatio-temporal relationship processing module based on the following feasible implementation methods: For any spatio-temporal relationship processing module, the electronic device can process the multiple grid information according to each graph convolutional neural network to obtain the spatial features output by each graph convolutional neural network, and process the multiple spatial features according to the LSTM network with an attention mechanism to obtain the spatio-temporal features. For example, if the number of grids in the target area is R, the number of historical time periods is T, and the dimension of travel demand is D, the target feature image can be a matrix of [R, 1, D], where the LSTM network with an attention mechanism can integrate T into 1 on the time scale.

[0087] Optionally, each spatio-temporal relationship processing module can obtain the spatio-temporal features of multiple grid information based on the grid information of multiple grids. For example, the multiple spatio-temporal features can be travel demands under different spatio-temporal conditions. During the training of the spatio-temporal relationship processing module, each spatio-temporal relationship processing module has different focuses on learning in multiple grid information (for example, spatio-temporal relationship processing module 1 focuses on learning travel demands on rainy days, and spatio-temporal relationship processing module 2 focuses on learning travel demands on sunny days, etc.). Based on multiple spatio-temporal relationship processing modules, accurate information related to travel demands can be learned from multiple grid information.

[0088] Optionally, the spatial features are used to indicate the spatial relationship between multiple grid information. For example, the graph convolutional neural network can integrate multiple grid information on the spatial scale, and each graph convolutional neural network can output a spatial feature, where the spatial features extracted by multiple graph convolutional neural networks have different focuses. For example, the spatio-temporal relationship processing module can include graph convolutional neural network 1 and graph convolutional neural network 2. Graph convolutional neural network 1 can process multiple grid information to obtain spatial feature A, and graph convolutional neural network 2 can process multiple grid information to obtain spatial feature B, where both spatial feature A and spatial feature B can indicate the spatial relationship between multiple grid information.

[0089] Optionally, the electronic device processes the multiple spatial features according to the LSTM network with an attention mechanism to obtain the spatio-temporal features, which can specifically be: splicing the multiple spatial features to obtain a spliced feature, and processing the spliced feature according to the LSTM network with an attention mechanism to obtain the spatio-temporal features.

[0090] Among them, the splicing feature can be the feature after splicing multiple spatial features output by multiple graph convolutional neural networks. For example, if the number of grids in the target area is R, the number of historical time periods is T, and the dimension of travel demand is D, then the multiple grid information can be a matrix of [R, T, D]. The graph convolutional neural network 1 processes the multiple grid information to obtain the spatial feature A (a matrix of [R, T, D]), and the graph convolutional neural network 2 processes the multiple grid information to obtain the spatial feature B (a matrix of [R, T, D]). Therefore, the electronic device can splice the spatial feature A and the spatial feature B to obtain the splicing feature.

[0091] Optionally, after the electronic device determines the splicing feature, it can process the splicing feature based on the LSTM network with an attention mechanism to obtain the spatio-temporal feature. For example, if the splicing feature is a matrix of [R, T, D1], the LSTM network with an attention mechanism can be integrated on the time scale to obtain a matrix of [R, 1, D2].

[0092] Optionally, the spatio-temporal relationship processing module can include a graph convolutional neural network established based on the distance proximity relationship and a graph convolutional neural network established based on the functional similarity relationship. For example, the edge weights between the grids in the graph convolutional neural network established based on the distance proximity relationship can be determined based on the distance between the grids. If the distance between the grids is closer, the edge weight between the grids is higher; if the distance between the grids is farther, the edge weight between the grids is lower. The edge weights between the grids in the graph convolutional neural network established based on the functional similarity relationship can be determined based on the functional types between the grids. If the functional type similarity between the grids is higher, the edge weight between the grids is higher; if the functional type similarity between the grids is lower, the edge weight between the grids is lower.

[0093] Next, in combination with Figure 5 , the structure of the spatio-temporal relationship processing module will be described.

[0094] Figure 5 This is a schematic structural diagram of a spatio-temporal relationship processing module provided by an embodiment of the present application. Please refer to Figure 5, including: a spatio-temporal relationship processing module. Among them, the spatio-temporal relationship processing module may include a graph convolutional neural network established based on distance proximity relationship, a graph convolutional neural network established based on functional similarity relationship, and a long short-term memory network with an attention mechanism. Among them, the long short-term memory network with an attention mechanism can be connected to the graph convolutional neural network established based on distance proximity relationship and the graph convolutional neural network established based on functional similarity relationship. In this way, the spatio-temporal relationship processing module can extract the spatial features between multiple grid information through the graph convolutional neural network established based on distance proximity relationship and the graph convolutional neural network established based on functional similarity relationship, and integrate multiple spatial features on the time scale based on the LSTM network with an attention mechanism to obtain spatio-temporal features.

[0095] Optionally, in Figure 5 In the embodiment shown, since multiple grid information includes data of multiple time scales, the electronic device can integrate multiple grid information on the time scale based on the LSTM network with an attention mechanism to obtain spatio-temporal features. For example, if the number of grids in the target area is R, the number of historical time periods is T, and the dimension of travel demand is D, the grid information of multiple grids can be a matrix of (R, T, D1). Integrating the spatial features output by multiple graph convolutional neural networks based on the LSTM network with an attention mechanism can obtain a matrix of (R, 1, D2).

[0096] Optionally, the electronic device can determine the graph convolutional neural network established based on distance proximity relationship according to the following feasible implementation method: obtain the distances between grids in the target area, and determine the edge weights between grids according to the distances between grids, so as to obtain the graph convolutional neural network established based on distance proximity relationship.

[0097] Optionally, the electronic device can obtain the distances between grids in the map data, or obtain the distances between grids based on any other feasible implementation method (for example, after processing the target area based on the H3 grid system, the distances between grids can be obtained). The embodiments of the present application do not limit this.

[0098] Optionally, the electronic device can determine the edge weights between grids according to the following formula:

[0099]

[0100] Among them, A i,jIt can be the edge weight between grid i and grid j in the graph convolutional neural network established based on the distance proximity relationship; k can be a preset coefficient (which can represent the influence degree of distance on the weight and can be arbitrarily selected, and the embodiments of this application do not limit this), dist(i,j) is the distance between grid i and grid j, and H is a preset threshold.

[0101] For example, for grids i and j in the graph convolutional neural network established based on the distance proximity relationship, if the distance between grid i and grid j is greater than H, the electronic device can determine that the edge weight between grid i and grid j in the graph convolutional neural network established based on the distance proximity relationship is 0. If the distance between grid i and grid j is less than H, the electronic device can determine the edge weight between grid i and grid j in the graph convolutional neural network established based on the distance proximity relationship according to the above formula.

[0102] Optionally, the electronic device can determine the graph convolutional neural network established based on the functional similarity relationship based on the following feasible implementation: obtain the interest point information in each grid in the target area, determine the similarity between each grid according to the interest point information between each grid, and obtain the graph convolutional neural network established based on the functional similarity relationship according to the similarity between each grid.

[0103] Optionally, the electronic device can determine the similarity between each grid based on the interest point information in the grid. For example, each type of interest point has a specific vector, and the electronic device can determine the vector corresponding to the grid based on the vectors of all interest points in the grid. For example, the interest point of the school type is 1, the interest point of the shopping mall type is 2, and the interest point of the gas station type is 3. If the grid includes 3 schools, 2 shopping malls and 1 gas station, the vector corresponding to this grid can be 1, 1, 1, 2, 2, 3. In this way, based on the vectors of all the interest points in the grid, the vector corresponding to this grid can be obtained.

[0104] For example, after the electronic device obtains the point-of-interest information between each grid, it can determine the vectors corresponding to each grid based on the point-of-interest information. The electronic device can determine the cosine similarity between the vectors corresponding to each grid, and based on the cosine similarity, determine the edge weights between each grid, thereby obtaining a graph convolutional neural network established based on the functional similarity relationship. For example, if the vector corresponding to grid 1 is vector A and the vector corresponding to grid 2 is vector B, the electronic device can calculate the cosine similarity between vector A and vector B. If the cosine similarity is less than a preset value, the electronic device can determine that the edge weight between grid 1 and grid 2 in the graph convolutional neural network established based on the functional similarity relationship is 0. If the cosine similarity is greater than the preset value, the electronic device can determine the edge weight between grid 1 and grid 2 in the graph convolutional neural network established based on the functional similarity relationship based on this cosine similarity (the cosine similarity can be determined as the edge weight, or the cosine similarity can be transformed to obtain the edge weight, and the embodiments of the present application do not limit this).

[0105] Next, in conjunction with Figure 6 , the process of determining the graph convolutional neural network established based on the distance proximity relationship will be described.

[0106] Figure 6 FIG. is a schematic diagram of a process for determining a graph convolutional neural network established based on the distance proximity relationship provided by an embodiment of the present application. Please refer to Figure 6 , including: grid 1, grid 2, grid 3, grid 4, grid 5, grid 6, and grid 7. Among them, the electronic device ( Figure 6 not shown) can obtain the actual geographical distance between any two grids, and based on the actual geographical distance between any two grids, determine the edge weight between these two grids in the graph convolutional neural network established based on the distance proximity relationship, that is, the electronic device can determine the graph convolutional neural network established based on the distance proximity relationship according to the distance. Among them, this graph convolutional neural network can also include 7 grids and the edge weights between the grids. For example, since the distance between grid 4 and each grid is relatively small, in the graph convolutional neural network established based on the distance proximity relationship, there are edge weights between grid 4 and each grid. Since the distance between grid 1 and grid 6 is relatively far, there is no edge weight between grid 1 and grid 6. However, since both grid 1 and grid 6 are connected to grid 4, there is also a relevant connection between grid 1 and grid 6.

[0107] Next, in conjunction with Figure 7 , the process of determining the graph convolutional neural network established based on the functional similarity relationship will be described.

[0108] Figure 7 FIG. is a schematic diagram of a process for determining a graph convolutional neural network established based on the functional similarity relationship provided by an embodiment of the present application. Please refer toFigure 7 , including Grid 1, Grid 2, Grid 3, Grid 4, Grid 5, Grid 6, and Grid 7. Among them, an electronic device ( Figure 7 not shown) can obtain the point-of-interest information in each grid, and determine the edge weights between any two grids in the graph convolutional neural network established based on the functional similarity relationship according to the point-of-interest information in each grid, that is, the electronic device can determine the graph convolutional neural network established based on the functional similarity relationship according to the points of interest. For example, although Grid 1 and Grid 2 are relatively close in distance, since the functions of Grid 1 and Grid 2 are quite different, there is no edge weight between Grid 1 and Grid 2 in the graph convolutional neural network established based on the functional similarity relationship. Although Grid 1 and Grid 6 are relatively far in distance, since the functions of Grid 1 and Grid 6 are similar, there is an edge weight between Grid 1 and Grid 6 in the graph convolutional neural network established based on the functional similarity relationship.

[0109] In this way, through the graph convolutional neural network established based on the distance proximity relationship and the graph convolutional neural network established based on the functional similarity relationship, the electronic device can integrate the grid information of multiple grids from different perspectives, and thus can improve the accuracy of predicting travel demand.

[0110] Next, in combination with Figure 8 , the process of determining spatio-temporal features will be described.

[0111] Figure 8 FIG. is a schematic diagram of a process for determining spatio-temporal features provided by an embodiment of the present application. Please refer to Figure 8 , including: grid information of 7 grids, a graph convolutional neural network established based on the distance proximity relationship, a graph convolutional neural network established based on the functional similarity relationship, and a long short-term memory network with an attention mechanism. Among them, the grid information of 7 grids can be a matrix of [R, T, D1], where R can be 7. The graph convolutional neural network established based on the distance proximity relationship may include Grid 1, Grid 2, Grid 3, Grid 4, Grid 5, Grid 6, and Grid 7, and the distance-related edge weights between each grid. The graph convolutional neural network established based on the functional similarity relationship may include Grid 1, Grid 2, Grid 3, Grid 4, Grid 5, Grid 6, and Grid 7, and the function-type-related edge weights between each grid.

[0112] Please refer to Figure 8, after the graph convolutional neural network established based on the distance proximity relationship processes the grid information, spatial feature 1 can be obtained. This spatial feature 1 can be a matrix of [R, T, D2]. After the graph convolutional neural network established based on the functional similarity relationship processes the grid information, spatial feature 2 can be obtained. This spatial feature 2 can be a matrix of [R, T, D2]. By splicing spatial feature 1 and spatial feature 2, a spliced feature can be obtained. This spliced feature can be a matrix of [R, T, D3]. The electronic device can input the spliced feature into the long short-term memory network with an attention mechanism, and then a spatio-temporal feature can be obtained. This spatio-temporal feature can be a matrix of [R, 1, D4]. In this way, based on multiple spatio-temporal relationship processing modules, the electronic device can obtain multiple spatio-temporal features between multiple grid information, and then improve the accuracy of travel demand prediction.

[0113] S205. Process multiple spatio-temporal features according to the multi-task model to obtain the predicted travel order volume and predicted order response rate of each grid in the target area in the future time period.

[0114] Among them, the predicted travel order volume can be the travel order volume of each grid in the target area predicted by the electronic device in the future time period, and the predicted order response rate can be the order response rate of each grid in the target area predicted by the electronic device in the future time period. For example, if there are 100 grids in the target area, the electronic device can predict the travel order volume and order response rate within each grid range in the future time period based on the grid information of these 100 grids. For example, if there are 100 grids in the target area and the future time period is from the next 1 hour to the next 2 hours, the electronic device can predict the travel order volume and order response rate of each grid from the next 1 hour to the next 2 hours based on the grid information of the 100 grids.

[0115] Optionally, the electronic device can process multiple spatio-temporal features based on the architecture of the multi-task model, and then obtain the predicted travel order volume and predicted order response rate. For example, the multi-task model can be a Multi-gate-Mixture-of-Experts (MMOE) model. The expert module in this MMOE model can be the spatio-temporal relationship processing module in this application. In this way, the electronic device can determine the predicted travel order volume and predicted order response rate of each grid in the target area in the future time period according to the spatio-temporal features extracted by the MMOE model and the spatio-temporal relationship processing module. Since multiple spatio-temporal features can accurately describe the relationship between multiple grid information in the time scale and the space scale, the accuracy of the predicted travel order volume and predicted order response rate is improved, and then the travel demand can be predicted more accurately.

[0116] Next, in combination withFigure 9 , the structure of the multi-task model will be described.

[0117] Figure 9 It is a schematic structural diagram of a multi-task model provided by an embodiment of the present application. Please refer to Figure 9 , including: a multi-task model. Among them, the multi-task model may include a spatio-temporal relationship processing module 1, a spatio-temporal relationship processing module 2, a spatio-temporal relationship processing module 3, a gating network a, a gating network b, a decoding module A, and a decoding module B. Among them, the spatio-temporal relationship processing module may be an expert module (expert) in the MMOE model, the gating network may be a gate network (gate) in the MMOE, and the decoding module may be a tower network in the MMOE model. After the results output by graph convolution 1, graph convolution 2, and graph convolution 3 are concatenated, they are combined with the result output by the gating network a, and the combined result is processed based on the decoding module A to obtain the result of the first task (predicting the number of travel order dispatches). After the results output by graph convolution 1, graph convolution 2, and graph convolution 3 are concatenated, they are combined with the result output by the gating network b, and the combined result is processed based on the decoding module B to obtain the result of the second task (predicting the order dispatch response rate).

[0118] It should be noted that Figure 5 it is only an exemplary illustration of the structure of the multi-task model and does not limit the structure of the multi-task model. For example, the number of spatio-temporal relationship processing modules in the multi-task model can be 5, 7, etc., and the embodiments of the present application do not limit this. Moreover, since the training tasks in the embodiments of the present application are 2 tasks (the number of travel order dispatches and the order dispatch response rate), therefore, the multi-task model in the embodiments of the present application may include 2 gating networks and 2 decoding modules. If the training tasks increase, the number of gating networks and decoding modules in the multi-task model can also increase. For example, if the training tasks are to give the number of travel order dispatches, the order dispatch response rate, and the response failure rate in a future time period, then each prediction requirement in the multi-task model can be associated with a decoding module and a gating network, and the embodiments of the present application do not limit this.

[0119] An embodiment of the present application provides a travel demand prediction method. An electronic device can determine a target area and a future time period for which travel demand is to be predicted. Among them, the target area may include multiple grids. According to the future time period, multiple historical time periods corresponding to the future time period are determined, and grid information of each grid in the target area is obtained. The grid information may include the historical travel order issuance volume and the historical order response rate of the grid in multiple historical time periods. According to multiple spatio-temporal relationship processing modules, the multiple grid information is processed to obtain spatio-temporal features associated with the multiple grid information output by each spatio-temporal relationship processing module. The electronic device can process the multiple spatio-temporal features according to a multi-task model to obtain the predicted travel order issuance volume and the predicted order response rate of each grid in the target area in the future time period. In this way, since the electronic device can determine multiple spatio-temporal features based on multiple spatio-temporal relationship processing modules, and the multiple spatio-temporal features can accurately describe the relationship between the multiple grid information on the spatial scale and the relationship between the multiple grid information on the time scale, the electronic device can accurately predict the travel demand in the future time period according to the multi-task model and the multiple spatio-temporal features, and improve the prediction accuracy of the travel demand.

[0120] Based on the embodiment shown in Figure 2 , the following will describe, in combination with Figure 10 , the method of processing multiple spatio-temporal features according to a multi-task model to obtain the predicted travel order issuance volume and the predicted order response rate of each grid in the target area in the future time period in the above travel demand prediction method.

[0121] Figure 10 It is a schematic diagram of a method for determining the predicted travel order issuance volume and the predicted order response rate provided by an embodiment of the present application. Please refer to Figure 10 , and the method process may include:

[0122] S1001. Determine the target model parameters corresponding to the target area, and replace the model parameters of the multi-task model with the target model parameters to obtain a target multi-task model.

[0123] Optionally, since the areas for which travel demand is to be predicted are different, the model parameters of the multi-task model are also different. Therefore, the electronic device can determine the target model parameters corresponding to the target area, and replace the model parameters of the current multi-task model with the target model parameters to obtain a target multi-task model. In this way, the target multi-task model can be a model for predicting the predicted travel order issuance volume and the predicted order response rate of the target area.

[0124] For example, if the expert module (spatiotemporal relationship processing module) and other modules in the multitask model are determined based on multiple grids in Region 1, then this set of model parameters of the multitask model corresponds to Region 1. For example, if the expert module (spatiotemporal relationship processing module) and other modules in the multitask model are determined based on multiple grids in Region 1, and Region 1 can include Region 2 and Region 3, then the electronic device can determine that Region 1, Region 2, and Region 3 all correspond to this set of model parameters of the multitask model.

[0125] Optionally, the electronic device can determine the target model parameters corresponding to the target region based on the identifier of the target region. For example, if the target region is City A, the electronic device can obtain the target model parameters corresponding to City A. If the target region is City B, the electronic device can obtain the target model parameters corresponding to City B.

[0126] Optionally, the electronic device can determine the target model parameters corresponding to the target region according to the corresponding relationship between multiple regions and model parameters. It should be noted that the electronic device can also determine the target model parameters corresponding to the target region based on any other feasible implementation manner, and the embodiments of the present disclosure do not limit this.

[0127] S1002. Process multiple spatiotemporal features according to the target multitask model to obtain a predicted travel order volume and a predicted order response rate.

[0128] Among them, the electronic device can determine the predicted travel order volume and the predicted order response rate based on the following feasible implementation manner: In the target multitask model, obtain multiple first weights corresponding to multiple spatiotemporal features and multiple second weights corresponding to multiple spatiotemporal features, and determine the predicted travel order volume and the predicted order response rate according to the multiple spatiotemporal features, the multiple first weights, and the multiple second weights.

[0129] Among them, the first weight is a weight generated based on the gating network related to the predicted travel order volume in the target multitask model, and the second weight is a weight generated based on the gating network related to the predicted order response rate in the target multitask model. For example, the electronic device can obtain multiple spatiotemporal features based on multiple spatiotemporal relationship processing modules. The electronic device can extract information related to the travel order volume from the multiple spatiotemporal features based on the multiple first weights, and extract information related to the order response rate from the multiple spatiotemporal features based on the multiple second weights.

[0130] Among them, the number of spatio-temporal relationship processing modules is the same as the number of first weights, and the number of spatio-temporal relationship processing modules is the same as the number of second weights. For example, if the multi-task model includes 10 spatio-temporal relationship processing modules (i.e., outputs 10 spatio-temporal features), then the number of first weights in the multi-task model is 10, and the number of second weights is 10. If the multi-task model includes 100 spatio-temporal relationship processing modules (i.e., outputs 100 spatio-temporal features), then the number of first weights in the multi-task model is 100, and the number of second weights is 100.

[0131] Optionally, in the target multi-task model, the electronic device obtains multiple first weights corresponding to multiple spatio-temporal features and multiple second weights corresponding to multiple spatio-temporal features. Specifically, it can be: according to the gating network related to predicting the order dispatch volume in the target multi-task model, process multiple grid information to obtain multiple first weights, and according to the gating network related to predicting the order response rate in the target multi-task model, process multiple grid information to obtain multiple second weights.

[0132] It should be noted that after the training of the multi-task model is completed, the parameters in the gating network are determined. After different gating networks process multiple grid information, the obtained weights are also different. For example, in the actual application process, if the spatio-temporal feature 1 has a high correlation with the task, then for this task, among the multiple weights output by the gating network, the weight corresponding to the spatio-temporal feature 1 is larger. If the spatio-temporal feature 2 has a low correlation with the task, then for this task, among the multiple weights output by the gating network, the weight corresponding to the spatio-temporal feature 2 is smaller.

[0133] Optionally, the structure of the gating network is the same as that of the spatio-temporal relationship processing module. For example, the gating network can also include multiple graph convolutional neural networks and an LSTM network with an attention mechanism. However, since the gating network is to determine multiple weights, the output result of the gating network can be processed based on the softmax layer to obtain multiple weights.

[0134] Next, in combination with Figure 11 , the structure of the gating network will be described.

[0135] Figure 11 FIG. is a schematic structural diagram of a gating network provided by an embodiment of the present application. Please refer to Figure 11 , including: a gating network. Among them, the gating network can include a graph convolutional neural network established based on distance proximity relationship, a graph convolutional neural network established based on functional similarity relationship, an LSTM network with an attention mechanism, and a softmax layer. The electronic device ( Figure 11(not shown) may input the grid information of multiple grids into the graph convolutional neural network established based on distance proximity relationship and the graph convolutional neural network established based on functional similarity relationship. The graph convolutional neural network established based on distance proximity relationship and the graph convolutional neural network established based on functional similarity relationship may output two spatial features. After the electronic device splices the two spatial features, a spliced feature is obtained. The electronic device may process the spliced feature based on an LSTM network with an attention mechanism, and process the result output by the LSTM with an attention mechanism based on a softmax layer, and then multiple weights may be obtained.

[0136] It should be noted that in the actual application process, if the multi-task model can determine the predicted order dispatch volume and the predicted order dispatch response rate in the future time period, the number of learning tasks of the multi-task model is 2. Therefore, the number of gating networks in the multi-task model is 2. One gating network is used to output multiple first weights related to the prediction of the order dispatch volume, and the other gating network is used to output multiple second weights related to the prediction of the order dispatch response rate.

[0137] Among them, the electronic device determines the predicted order dispatch volume and the predicted order dispatch response rate according to multiple spatio-temporal features, multiple first weights, and multiple second weights. Specifically, it may be: determining the weighted sum of multiple spatio-temporal features and multiple first weights to obtain the feature associated with the predicted order dispatch volume, determining the weighted sum of multiple spatio-temporal features and multiple second weights to obtain the feature associated with the predicted order dispatch response rate, decoding the feature associated with the predicted order dispatch volume to obtain the predicted order dispatch volume, and decoding the feature associated with the predicted order dispatch response rate to obtain the predicted order dispatch response rate.

[0138] Optionally, since the number of spatio-temporal features is the same as the number of first weights, the electronic device may determine the weighted sum of multiple spatio-temporal features and multiple first weights to obtain the feature associated with the predicted order dispatch volume. For example, if the electronic device processes the grid information of multiple grids based on 3 spatio-temporal relationship processing modules, spatio-temporal feature X1, spatio-temporal feature X2, and spatio-temporal feature X3 may be obtained. The electronic device processes the grid information of multiple grids based on the gating network related to the order dispatch volume, and weights q1, q2, and q3 may be obtained. Then the electronic device may perform weighted sum processing on spatio-temporal feature X1, spatio-temporal feature X2, spatio-temporal feature X3, and weights q1, q2, and q3, and then obtain the feature associated with the predicted order dispatch volume (X1*q1 + X2*q2 + X3*q3).

[0139] Optionally, since the number of spatio-temporal features is the same as the number of second weights, the electronic device can determine the weighted sum of multiple spatio-temporal features and multiple second weights to obtain the features associated with the predicted order dispatch response rate. For example, if the electronic device processes multiple grid information based on 3 spatio-temporal relationship processing modules, spatio-temporal feature X1, spatio-temporal feature X2, and spatio-temporal feature X3 can be obtained. If the electronic device processes multiple grid information based on a gating network related to the order dispatch volume of trips, weights p1, p2, and p3 can be obtained. Then, the electronic device can perform weighted sum processing on spatio-temporal feature X1, spatio-temporal feature X2, spatio-temporal feature X3, and weights p1, p2, and p3, and further obtain the features associated with the predicted order dispatch response rate (X1*p1 + X2*p2 + X3*p3).

[0140] Optionally, the electronic device can perform decoding processing on the features associated with the predicted order dispatch volume of trips, and further obtain the predicted order dispatch volume of trips for each grid in the future time period. For example, after the electronic device determines the features associated with the predicted order dispatch volume of trips, it can perform decoding processing on the features based on the decoding module in the multi-task model, and further obtain the predicted order dispatch volume of trips. Similarly, the electronic device can obtain the predicted order dispatch response rate for each grid in the future time period.

[0141] It should be noted that if the number of tasks changes, the structure of the multi-task model will also change. For example, if the predicted travel demands of the electronic device can include the predicted order dispatch volume of trips and the predicted order dispatch response rate for future time period 1, future time period 2, and future time period 3, then there can be 6 decoders in the multi-task model (2 decoding modules are required for each future time period), and there can also be 6 gating networks in the multi-task model (2 gating networks are required for each future time period). In this way, the terminal device can predict the travel demands in the future time period in multiple steps, improving the flexibility of travel demand prediction.

[0142] It should be noted that since the spatio-temporal relationship processing module in the multi-task model can learn the travel demand information under different spatio-temporal conditions, the number of spatio-temporal relationship processing modules in the multi-task model can be set arbitrarily. For example, the multi-task model can include 3 spatio-temporal relationship processing modules, 5 spatio-temporal relationship processing modules, or 10 spatio-temporal relationship processing modules. The embodiments of the present application do not limit this.

[0143] It should be noted that when obtaining the grid information of each grid in the target area, the electronic device can determine relevant grids among multiple grids and add the same information to the relevant grids, thereby improving the accuracy of predicting travel demand. For example, the size of the H3 grid at level A is smaller than that of the H3 grid at level B. Based on the H3 grid system at level A, the target area is divided into grids 1, 2, 3, and 4. The electronic device can also divide the target area based on the H3 grid system at level B to obtain grids a and b. Among them, if grid a includes grids 1 and 2, and grid b includes grids 3 and 4, then the electronic device can determine that grids 1 and 2 are relevant grids, and grids 3 and 4 are relevant grids. Therefore, the electronic device can add the grid information of grid a to the grid information of grid 1, add the grid information of grid a to the grid information of grid 2, add the grid information of grid b to the grid information of grid 3, and add the grid information of grid b to the grid information of grid 4. In this way, the correlation between grids 1 and 2 can be improved, the correlation between grids 3 and 4 can be improved, and more grid information can be obtained, further improving the accuracy of the electronic device's prediction of travel demand.

[0144] The embodiment of the present application provides a method for determining the predicted travel order volume and the predicted order response rate. Determine the target model parameters corresponding to the target area, and replace the model parameters of the multi-task model with the target model parameters to obtain the target multi-task model. In the target multi-task model, obtain multiple first weights corresponding to multiple spatio-temporal features and multiple second weights corresponding to multiple spatio-temporal features, determine the weighted sum of the multiple spatio-temporal features and the multiple first weights to obtain the features associated with the predicted travel order volume, determine the weighted sum of the multiple spatio-temporal features and the multiple second weights to obtain the features associated with the predicted order response rate, perform decoding processing on the features associated with the predicted travel order volume to obtain the predicted travel order volume, and perform decoding processing on the features associated with the predicted order response rate to obtain the predicted order response rate. In this way, since multiple spatio-temporal relationship processing modules can obtain travel demand information under different spatio-temporal conditions, the electronic device can accurately predict the predicted travel order volume and the predicted order response rate in each grid in the future period. Moreover, the electronic device can further reflect the travel demand in the target area from both the supply and demand sides, thereby improving the accuracy of determining travel demand.

[0145] Based on any one of the above embodiments, hereinafter, in combination with Figure 12 , the process of the above travel demand prediction method will be described.

[0146] Figure 12 It is a schematic diagram of the process of a travel demand prediction method provided by an embodiment of the present application. Please refer to Figure 12, including: grid information of multiple grids in the target area and a multi-task model. The multi-task model may include a spatio-temporal relationship processing module 1, a spatio-temporal relationship processing module 2, a spatio-temporal relationship processing module 3, a gating network 1, a gating network 2, a decoder A, and a decoder B. The structures of the spatio-temporal relationship processing module, the gating network, and the decoder are not elaborated in this embodiment of the present application. An electronic device ( Figure 12 not shown) may obtain the grid information of multiple grids in the target area based on a future time period, and input the grid information of the multiple grids into the multi-task model.

[0147] Please refer to Figure 12 . After the spatio-temporal relationship processing module 1 processes the grid information of the multiple grids, spatio-temporal feature x1 can be obtained. After the spatio-temporal relationship processing module 2 processes the grid information of the multiple grids, spatio-temporal feature x2 can be obtained. After the spatio-temporal relationship processing module 3 processes the grid information of the multiple grids, spatio-temporal feature x3 can be obtained. After the gating network 1 processes the grid information of the multiple grids, first weights q1, q2, and q3 can be output. After the gating network 2 processes the grid information of the multiple grids, second weights p1, p2, and p3 can be output.

[0148] Please refer to Figure 12 . The multi-task model can determine the weighted sum between the spatio-temporal feature x1, the spatio-temporal feature x2, the spatio-temporal feature x3 and the first weights q1, q2, and q3 to obtain q1 * x1 + q2 * x2 + q3 * x3. The multi-task model can determine the weighted sum between the spatio-temporal feature x1, the spatio-temporal feature x2, the spatio-temporal feature x3 and the second weights p1, p2, and p3 to obtain p1 * x1 + p2 * x2 + p3 * x3.

[0149] Please refer to Figure 12 . The multi-task model can perform decoding processing on the weighted sum q1 * x1 + q2 * x2 + q3 * x3 based on decoder A to obtain the predicted order dispatch volume of each grid in the future time period. The multi-task model can perform decoding processing on the weighted sum p1 * x1 + p2 * x2 + p3 * x3 based on decoder B to obtain the predicted order response rate of each grid in the future time period. In this way, the electronic device can determine whether the number of online car-hailing vehicles in each grid of the target area is sufficient based on the predicted order dispatch volume and the predicted order response rate, accurately predict travel demands from both the supply and demand sides. And since multiple spatio-temporal relationship processing modules can obtain travel demand information under different spatio-temporal conditions, the electronic device can accurately predict the predicted order dispatch volume and the predicted order response rate in each grid in the future time period.

[0150] Figure 13 This is a schematic structural diagram of a travel demand prediction device provided by an embodiment of the present application. Please refer toFigure 13 , the travel demand prediction device 130 includes a first determination module 131, a second determination module 132, an acquisition module 133, and a processing module 134, where:

[0151] The first determination module 131 is configured to determine a target area and a future time period for the travel demand to be predicted, where the target area includes a plurality of grids;

[0152] The second determination module 132 is configured to determine a plurality of historical time periods corresponding to the future time period according to the future time period;

[0153] The acquisition module 133 is configured to acquire grid information of each grid in the target area, where the grid information includes the historical travel order issuance volume and the historical order response rate of the grid in each historical time period;

[0154] The processing module 134 is configured to process a plurality of grid information according to a plurality of spatio-temporal relationship processing modules to obtain spatio-temporal features associated with the plurality of grid information output by each spatio-temporal relationship processing module;

[0155] The processing module 134 is further configured to process a plurality of spatio-temporal features according to a multi-task model to obtain the predicted travel order issuance volume and the predicted order response rate of each grid in the target area in the future time period.

[0156] In a possible implementation manner, the processing module 134 is specifically configured to:

[0157] For any one spatio-temporal relationship processing module;

[0158] Process the plurality of grid information according to each graph convolutional neural network to obtain spatial features output by each graph convolutional neural network;

[0159] Process a plurality of spatial features according to the LSTM network with an attention mechanism to obtain the spatio-temporal features.

[0160] In a possible implementation manner, the processing module 134 is specifically configured to:

[0161] Perform splicing processing on the plurality of spatial features to obtain a spliced feature;

[0162] Process the spliced feature according to the LSTM network with an attention mechanism to obtain the spatio-temporal features.

[0163] In a possible implementation manner, the spatio-temporal relationship processing module includes a graph convolutional neural network established based on distance proximity relationship and a graph convolutional neural network established based on functional similarity relationship.

[0164] In a possible implementation, the processing module 134 is specifically configured to:

[0165] Determine the target model parameters corresponding to the target area, and replace the model parameters of the multi-task model with the target model parameters to obtain a target multi-task model;

[0166] Process the multiple spatio-temporal features according to the target multi-task model to obtain the predicted order dispatch volume and the predicted order dispatch response rate.

[0167] In a possible implementation, the processing module 134 is specifically configured to:

[0168] In the target multi-task model, obtain multiple first weights corresponding to the multiple spatio-temporal features and multiple second weights corresponding to the multiple spatio-temporal features, where the first weight is a weight generated based on a gating network related to predicting the order dispatch volume in the target multi-task model, and the second weight is a weight generated based on a gating network related to predicting the order dispatch response rate in the target multi-task model;

[0169] Determine the predicted order dispatch volume and the predicted order dispatch response rate according to the multiple spatio-temporal features, the multiple first weights, and the multiple second weights.

[0170] In a possible implementation, the processing module 134 is specifically configured to:

[0171] Determine the weighted sum of the multiple spatio-temporal features and the multiple first weights to obtain the features associated with the predicted order dispatch volume;

[0172] Determine the weighted sum of the multiple spatio-temporal features and the multiple second weights to obtain the features associated with the predicted order dispatch response rate;

[0173] Perform decoding processing on the features associated with the predicted order dispatch volume to obtain the predicted order dispatch volume;

[0174] Perform decoding processing on the features associated with the predicted order dispatch response rate to obtain the predicted order dispatch response rate.

[0175] In a possible implementation, the processing module 134 is specifically configured to:

[0176] Process the multiple grid information according to the gating network related to the order dispatch volume in the multi-task model to obtain the multiple first weights;

[0177] Process the multiple grid information according to the gating network related to the order dispatch response rate in the multi-task model to obtain the multiple second weights.

[0178] The travel demand prediction device provided by the embodiments of the present application can execute the technical solutions shown in the above method embodiments. The implementation principles and beneficial effects are similar and will not be elaborated here.

[0179] Figure 14 It is a schematic hardware structure diagram of the electronic device provided by the present application. Please refer to Figure 14 , the electronic device 140 may include: a processor 141 and a memory 142. Among them, the processor 141 and the memory 142 can communicate. Exemplarily, the processor 141 and the memory 142 communicate through a communication bus 143. The memory 142 is used to store program instructions, and the processor 141 is used to call the program instructions in the memory to execute the travel demand prediction method shown in any of the above method embodiments.

[0180] Optionally, the electronic device 140 may further include a communication interface, and the communication interface may include a transmitter and / or a receiver.

[0181] Optionally, the above-mentioned processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the present application can be directly embodied as being executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

[0182] The embodiments of the present application provide a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the travel demand prediction methods as described in the first aspect and various possible aspects related to the first aspect are executed.

[0183] The embodiments of the present application may further provide a computer program product, which includes a computer program. When the computer program is executed by a processor, the travel demand prediction methods as described in the first aspect and various possible aspects related to the first aspect are executed.

[0184] All or part of the steps of the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a readable memory. When the program is executed, it executes the steps including the above method embodiments; and the foregoing memory (storage medium) includes: read-only memory (abbreviation: ROM), RAM, flash memory, hard disk, solid state drive, magnetic tape, floppy disk, optical disc, and any combination thereof.

[0185] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processing unit of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable electronic devices to generate a machine, such that the instructions executed by the processing unit of the computer or other programmable electronic devices generate a device for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0186] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable electronic device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0187] These computer program instructions can also be loaded onto a computer or other programmable electronic device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0188] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these changes and modifications.

[0189] In this application, the term "including" and its variants may refer to non-limiting inclusion; the term "or" and its variants may refer to "and / or". In this application, the terms "first", "second", etc. are used to distinguish similar objects and do not necessarily have to be used to describe a specific order or sequence. In this application, "a plurality of" means two or more. "And / or" describes the relationship between related objects and indicates that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the related objects before and after are in an "or" relationship.

Claims

1. A travel demand prediction method, characterized in that, Including: Determine the target area and future time period for predicting travel demand, where the target area includes multiple grids; Determine multiple historical time periods corresponding to the future time period according to the future time period; Obtain the grid information of each grid in the target area, where the grid information includes the historical travel order volume and historical order response rate of the grid in each historical time period; Process the multiple grid information according to multiple spatio-temporal relationship processing modules to obtain spatio-temporal features associated with the multiple grid information output by each spatio-temporal relationship processing module; Process the multiple spatio-temporal features according to a multi-task model to obtain the predicted travel order volume and predicted order response rate of each grid in the target area during the future time period.

2. The method according to claim 1, wherein The spatio-temporal relationship processing module includes multiple graph convolutional neural networks and a long short-term memory network LSTM with an attention mechanism. For any one spatio-temporal relationship processing module; processing the multiple grid information according to multiple spatio-temporal relationship processing modules to obtain spatio-temporal features associated with the multiple grid information output by each spatio-temporal relationship processing module, including: Process the multiple grid information according to each graph convolutional neural network to obtain spatial features output by each graph convolutional neural network; Process the multiple spatial features according to the LSTM network with an attention mechanism to obtain the spatio-temporal features.

3. The method according to claim 2, wherein The process of processing the multiple spatial features according to the LSTM network with an attention mechanism to obtain the spatio-temporal features includes: Perform splicing processing on the multiple spatial features to obtain a spliced feature; Process the spliced feature according to the LSTM network with an attention mechanism to obtain the spatio-temporal features.

4. The method according to claim 2, wherein The spatio-temporal relationship processing module includes a graph convolutional neural network established based on distance proximity relationship and a graph convolutional neural network established based on functional similarity relationship.

5. The method according to any one of claims 1-4, characterized in that, The process of processing the multiple spatio-temporal features according to a multi-task model to obtain the predicted travel order volume and predicted order response rate of each grid in the target area during the future time period includes: Determine the target model parameters corresponding to the target area, and replace the model parameters of the multi-task model with the target model parameters to obtain a target multi-task model; Process the multiple spatio-temporal features according to the target multi-task model to obtain the predicted travel order volume and the predicted order response rate.

6. The method according to claim 5, wherein The process of processing the multiple spatio-temporal features according to the target multi-task model to obtain the predicted travel order volume and the predicted order response rate includes: In the target multi-task model, obtain multiple first weights corresponding to the multiple spatio-temporal features and multiple second weights corresponding to the multiple spatio-temporal features. The first weight is a weight generated based on a gating network related to the predicted travel order volume in the target multi-task model, and the second weight is a weight generated based on a gating network related to the predicted order response rate in the target multi-task model; Determine the predicted order issuance volume for trips and the predicted order response rate according to the multiple spatio-temporal features, the multiple first weights, and the multiple second weights.

7. The method according to claim 6, characterized in that, The determining the predicted order issuance volume for trips and the predicted order response rate according to the multiple spatio-temporal features, the multiple first weights, and the multiple second weights includes: Determine the weighted sum of the multiple spatio-temporal features and the multiple first weights to obtain the features associated with the predicted order issuance volume for trips; Determine the weighted sum of the multiple spatio-temporal features and the multiple second weights to obtain the features associated with the predicted order response rate; Perform decoding processing on the features associated with the predicted order issuance volume for trips to obtain the predicted order issuance volume for trips; Perform decoding processing on the features associated with the predicted order response rate to obtain the predicted order response rate.

8. The method according to claim 6 or 7, characterized in that The obtaining, in the target multi-task model, the multiple first weights corresponding to the multiple spatio-temporal features and the multiple second weights corresponding to the multiple spatio-temporal features includes: Process the multiple grid information according to the gating network associated with the predicted order issuance volume for trips in the target multi-task model to obtain the multiple first weights; Process the multiple grid information according to the gating network associated with the predicted order response rate in the target multi-task model to obtain the multiple second weights.

9. A travel demand prediction device, characterized in that, It includes a first determination module, a second determination module, an acquisition module, and a processing module, where: The first determination module is configured to determine the target area and the future time period of the travel demand to be predicted, and the target area includes multiple grids; The second determination module is configured to determine, according to the future time period, the multiple historical time periods corresponding to the future time period; The acquisition module is configured to acquire the grid information of each grid in the target area, and the grid information includes the historical order issuance volume and the historical order response rate of the grid in each of the historical time periods; The processing module is configured to process the multiple grid information according to multiple spatio-temporal relationship processing modules to obtain the spatio-temporal features associated with the multiple grid information output by each spatio-temporal relationship processing module; The processing module is further configured to process the multiple spatio-temporal features according to a multi-task model to obtain the predicted order issuance volume for trips and the predicted order response rate of each grid in the target area during the future time period.

10. An electronic device, characterized in that, It includes: A processor and a memory; The memory stores computer execution instructions; The processor executes the computer execution instructions stored in the memory, so that the processor executes the travel demand prediction method according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, Computer execution instructions are stored in the computer-readable storage medium, and when the computer execution instructions are executed by a processor, they are used to implement the travel demand prediction method according to any one of claims 1-8.

Images