Method for updating historical navigation routes in a map navigation cache database

Abstract

The present invention discloses a method for updating historical navigation routes in a map navigation cache database, including: Step 1, obtaining historical navigation routes in the map navigation cache database to establish a corresponding abstract graph model; Step 2, combining the abstract graph model with the introduced cycling routes and historical navigation routes, and determining the historical navigation routes to be updated through deep learning fitting; Step 3, obtaining the latest navigation routes corresponding to the historical navigation routes to be updated: according to the historical navigation routes to be updated, querying a third-party map service provider to obtain the corresponding current latest navigation routes; Step 4, updating the historical navigation routes: according to the current latest navigation routes obtained in Step 3, replacing and updating the corresponding historical navigation routes in the map navigation cache database, and updating the corresponding navigation distances. This method can accurately find out the navigation routes to be updated, reduce the number of updated routes, and lower the update cost.

Description

Technical Field

[0001] The present invention relates to the field of electronic map navigation, and particularly to a method for updating historical navigation routes in a map navigation cache database. Background Art

[0002] Currently, for enterprises that require a large number of map-related services, after obtaining navigation data (such as navigation routes between two points) from a third-party professional map service provider, it is saved in the map navigation cache database. Since accessing the third party will increase the latency, in order to improve the response time, the map navigation cache database stores a large amount of historically obtained navigation data to ensure that there is a corresponding historical navigation data as the return result when an application accesses the map navigation cache database. However, the problem with this is that: there is a large amount of historical navigation data obtained from the third party stored in the map navigation cache database, and due to changes in road conditions in the real world (such as road construction, temporary construction, etc.), most of the shortest distance navigation routes in the historical cache do not match the current latest navigation routes. The obsolescence of navigation data will affect the quality of downstream services. For example, in the takeaway delivery scenario, providing a navigation route that is actually impassable will affect the timely delivery of dishes by the takeaway delivery staff.

[0003] To solve the problem of obsolete cache data in the static cache framework, currently, a commonly used method is to update by combining the access frequency and the last update date:

[0004] In the offline stage, count the paths with high query frequencies from the query logs; access the third-party map service provider to update the top navigation routes whose access frequencies have not been updated for three months;

[0005] Update the navigation routes with high query frequencies that affect the largest number of orders;

[0006] Since the probability that the data not updated for three months is obsolete data is relatively high, the probability of effective update can be increased.

[0007] However, the existing method does not consider the situation of the navigation routes themselves and only considers from the dimension of the non-update time. When querying and updating, there are fewer changed routes before and after the update, resulting in a large amount of obsolete navigation data that remains unupdated in the map navigation cache database. Summary of the Invention

[0008] Aiming at the problems existing in the prior art, the purpose of the present invention is to provide a method for updating historical navigation routes in a map navigation cache database, which can solve the problem that in the existing method for updating obsolete historical navigation routes, since the situation of the navigation routes themselves is not considered and only the non-update time dimension is considered, there are fewer changed navigation routes before and after querying and updating, resulting in a large amount of obsolete navigation routes that remain unupdated in the local database.

[0009] The object of the present invention is achieved by the following technical solutions:

[0010] An embodiment of the present invention provides a method for updating historical navigation routes in a map navigation cache database, including:

[0011] Step 1, establish an abstract graph model: Obtain historical navigation routes in the map navigation cache database, and establish an abstract graph model according to the historical navigation routes;

[0012] Step 2, determine the historical navigation routes to be updated through the abstract graph model: Obtain valid cycling routes, and perform fitting and clustering processing on the valid cycling routes and historical navigation routes corresponding to each abstract edge of the abstract graph model through deep learning to obtain a fitting result of the optimal route distance for the current abstract edge. Determine the abstract edges to be updated according to the fitting result, and determine the navigation routes to be updated according to the determined abstract edges;

[0013] Step 3, obtain the latest navigation routes corresponding to the historical navigation routes to be updated: According to the historical navigation routes to be updated, query a third-party map service provider to obtain the corresponding current latest navigation routes;

[0014] Step 4, update the historical navigation routes: According to the current latest navigation routes obtained in Step 3, replace and update the corresponding historical navigation routes in the map navigation cache database, and update the corresponding navigation distances.

[0015] It can be seen from the technical solutions provided by the present invention above that the method for updating historical navigation routes in the map navigation cache database provided by the embodiments of the present invention has the following beneficial effects:

[0016] By modeling historical navigation routes as an abstract graph and introducing cycling routes, fitting the edge weights of abstract edges by combining the abstract graph with historical navigation routes, determining the abstract edges to be updated according to the fitting result, and then determining the corresponding historical navigation routes to be updated, and updating the historical navigation routes to be updated by querying a third-party map service provider, it can not only effectively update the map historical navigation route library, but also, since only the selected historical navigation routes are updated, it can not only ensure the effectiveness of the update quantity, but also reduce the amount of data to be updated, so that the navigation routes obtained by accessing the database are consistent with the current best navigation routes. This solution belongs to a lightweight and effective update solution and is applicable to most historical navigation data update scenarios. Description of the Drawings

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 Flowchart of the historical navigation route update method for the map navigation cache database provided by the embodiments of the present invention;

[0019] Figure 2 Schematic diagram of the edge weight fitting deep learning model in the method provided by the embodiments of the present invention. Specific embodiments

[0020] The following combines the specific content of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention. The content not described in detail in the embodiments of the present invention belongs to the prior art well-known to those skilled in the art.

[0021] See Figure 1 , the embodiments of the present invention provide a method for updating the historical navigation route of a map navigation cache database, including:

[0022] Step 1, establish an abstract graph model: obtain the historical navigation routes in the map navigation cache database, and establish an abstract graph model according to the historical navigation routes; the historical navigation routes are navigation data stored in a GPS point sequence;

[0023] Step 2, determine the historical navigation routes to be updated through the abstract graph model: obtain the effective cycling routes, and perform fitting and clustering processing on the effective cycling routes corresponding to each abstract edge of the abstract graph model and the historical navigation routes through deep learning to obtain the fitting result of the optimal route distance for the current abstract edge, determine the abstract edges to be updated according to the fitting result, and determine the navigation routes to be updated according to the determined abstract edges; the cycling routes are determined by cycling trajectory data, and the cycling trajectory data is navigation data stored in GPS points;

[0024] Step 3, obtain the latest navigation routes corresponding to the historical navigation routes to be updated: according to the historical navigation routes to be updated, query the third-party map service provider to obtain the corresponding current latest navigation routes;

[0025] Step 4, Update the historical navigation route: According to the currently latest navigation route obtained in Step 3, replace and update the corresponding historical navigation route in the map navigation cache database, and update the corresponding navigation distance.

[0026] Step 3 of the above method further includes:

[0027] Select a group of navigation routes to be preferentially updated from the historical navigation routes to be updated, and query the third-party map service provider through the navigation routes to obtain the corresponding currently latest navigation routes. This way of selecting a group of navigation routes to be preferentially updated can reduce the number of historical navigation routes to be updated, and reduce the update cost and calculation overhead.

[0028] In the above Step 3, select a group of navigation routes to be preferentially updated from the historical navigation routes to be updated in the following way, including:

[0029] Step 31. Calculate the update benefit of each abstract edge according to the usage frequency and change amount of the abstract edges of the abstract graph model;

[0030] Step 32. Select paths from the abstract edges of the abstract graph model so that the sum of the update benefits of the abstract edges included in the selected paths is the largest;

[0031] Step 33. Calculate the sum of the update benefits of the edges included in each path, sort the paths in descending order of benefit, and select the paths in turn according to the path sorting. When the edges included in the selected path have been included in the path selected in Step 32, skip this path and continue to select the next path; This way of selecting paths covers as many edges as possible that are not included in other selected paths on the basis of considering the update benefit;

[0032] Step 34. Repeat Steps 32-33 until the number of selected paths reaches the number of paths in the update budget.

[0033] Step 4 of the above method further includes:

[0034] According to the obtained currently latest navigation route, update the corresponding historical navigation route in the map navigation cache database at preset different periods.

[0035] In the above Step 4, perform the first-period update and the second-period update on the map navigation cache database respectively at the preset first period and second period, where

[0036] The duration of the first period is less than that of the second period;

[0037] The first-period update is: when it is necessary to update at the first period, if the abstract edges included in the historical navigation route remain unchanged, use the edge weights of the obtained latest abstract edges to update the navigation distance of the historical navigation route;

[0038] The second cycle is updated as follows: when the update of the second cycle is required, based on the edge weights of the abstract edges obtained from all the first cycle updates after the previous second cycle update, the latest navigation distances of all historical navigation routes are recalculated on the abstract graph model by the shortest path method, and the corresponding historical navigation routes are updated with the obtained new navigation distances.

[0039] In step 4 above, the shortest path method uses the Dijsktra shortest path algorithm to calculate the shortest path;

[0040] The first cycle is one day;

[0041] The second cycle is one week.

[0042] It can be known that the first and second cycles can also be set to different durations according to needs, as long as the first cycle is shorter than the second cycle.

[0043] This way of updating in different cycles can reduce the number of updates and the update overhead while keeping the historical navigation routes updated in a timely and effective manner.

[0044] In step 1 of the above method, the historical navigation routes are established as an abstract graph model in the following way, including:

[0045] An abstract graph model is established using the intersection points in the historical navigation routes, and the historical navigation routes are mapped into a sequence of abstract nodes of the abstract graph model, and the navigation distances of the historical navigation routes are used as the edge weights of the abstract edges of the abstract graph model.

[0046] In step 1 above, the intersection points in the historical navigation routes are obtained, and the core nodes in the intersection points are extracted through clustering processing as abstract nodes to establish an abstract graph model; preferably, the clustering processing is performed using the DB-SCAN clustering algorithm.

[0047] The historical navigation routes passing through at least two abstract nodes of the abstract graph model are used as an abstract edge, and the edge weight of the abstract edge is the navigation distance of the navigation route between the abstract nodes;

[0048] The historical navigation routes are mapped into a sequence of abstract nodes of the abstract graph model according to the abstract nodes passed through.

[0049] In step 2 of the above method, the effective cycling routes are obtained in the following way, including:

[0050] First, the original cycling routes of the rider are obtained, and based on a preset threshold, the original cycling routes are detected for abnormal points and filtered to remove invalid cycling routes, and the effective cycling routes are obtained;

[0051] Deep learning is used to fit and cluster the effective cycling routes and historical navigation routes corresponding to each abstract edge of the abstract graph model, and obtain the fitting result of the optimal route distance for the current abstract edge, including:

[0052] Deep learning is used to fit the effective cycling routes and historical navigation routes corresponding to each abstract edge of the abstract graph, and obtain the fitting result of the optimal route distance for the current abstract edge;

[0053] Cluster the fitting results output by the abstract edges with multiple cycling routes obtained from the above fitting process, and take the class with the largest number as the fitting result. Preferably, the DBSCAN clustering algorithm is used for clustering;

[0054] See Figure 2 In step 2 above, deep learning uses an edge weight fitting deep learning model composed of a first preprocessing module, a first LSTM network, a second preprocessing module, a second LSTM network, a Bi-LSTM network, a first MLP network, a Fast-DTW module, a similarity measurement module, a basic feature module, and a second MLP network. Among them,

[0055] The input of the first preprocessing module is the historical navigation route, and the output of the first preprocessing module is connected to the first LSTM network;

[0056] The input of the second preprocessing module is the cycling route, and the output of the second preprocessing module is connected to the second LSTM network;

[0057] The outputs of the first LSTM network and the second LSTM network are connected to the Bi-LSTM network, and the output of the Bi-LSTM network is connected to the first MLP network;

[0058] The input of the Fast-DTW module is the historical navigation route and the cycling route, and the output of the Fast-DTW module is connected to the similarity measurement module and the basic feature module in sequence;

[0059] The outputs of the first MLP network and the basic feature module are both connected to the second MLP network, and the output of the second MLP network is the predicted navigation distance.

[0060] In the above edge weight fitting simulation, the Fast-DTW module runs the Fast-DTW algorithm, and the similarity measurement module runs the similarity measurement algorithm; this edge weight fitting simulation is mainly composed of LSTM and Bi-LSTM, aggregates the clustering results and the intermediate results of the model, and uses a fully connected model to output the final predicted navigation distance.

[0061] The method of the present invention can effectively update the historical navigation route library, making the navigation route obtained by accessing the database consistent with the current best navigation route. It is a lightweight and effective method for updating the map navigation cache database, which can be deployed in a navigation database with a scale of hundreds of millions, providing real-time updates. Compared with the existing methods, this method significantly improves the accuracy of the historical navigation routes in the map navigation cache database.

[0062] The following further describes the embodiments of the present invention in detail.

[0063] The embodiment of the present invention provides a method for updating historical navigation routes in a map navigation cache database. It is a method that models historical navigation routes as an abstract graph, introduces a cycling route to predict the navigation distance, determines the historical navigation routes that need to be updated according to the predicted navigation distance, and updates the historical navigation routes in the map historical navigation route library. The method includes the following steps:

[0064] Step 1, establish an abstract graph model: Establish an abstract graph model according to the historical navigation routes in the map navigation cache database, and map the historical navigation routes to an abstract node sequence on the abstract graph model, which is convenient for subsequent processing of updating historical navigation routes. The historical navigation routes are navigation data stored by GPS points;

[0065] Step 2, perform edge weight fitting on the abstract graph model to determine the historical navigation routes that need to be updated: Obtain the cycling route of the rider, and based on the cycling route and historical navigation routes, estimate the navigation distance of the current best navigation route corresponding to the abstract edge of the abstract graph model through deep learning. Determine which historical navigation routes corresponding to the abstract edge need to be updated according to the obtained navigation distance; that is, the initial edge weight of the established abstract graph model is the information of the historical cache, that is, the original navigation distance of the historical navigation route. After the introduced cycling route is fitted with the abstract edge of the abstract graph model, the latest navigation distance of the latest cycling route can be obtained, and then it can be determined which historical navigation routes have changed in the original navigation distance. These historical navigation routes with changed navigation distances are the navigation routes that need to be updated;

[0066] Step 3, obtain the corresponding navigation routes for update: According to the determined historical navigation routes that need to be updated, query the third-party map service provider to obtain their corresponding current latest navigation routes;

[0067] Step 4, update the historical navigation routes, and use the obtained current latest navigation routes to update the corresponding historical navigation routes and corresponding navigation distances in the map navigation cache database.

[0068] Furthermore, in step 3, a group of navigation routes to be updated first are selected from the determined historical navigation routes that need to be updated, and the third-party map service provider is queried with the group of navigation routes to be updated first to obtain the corresponding current latest navigation routes. This method can reduce the number of updated routes and reduce the update cost, and is more suitable for scenarios where all routes cannot be updated due to budget constraints.

[0069] In the above step 3, since the historical navigation route is updated by querying the third-party map service provider to obtain the accurate current latest navigation route, rather than directly updating the historical navigation route based on the cycling route, it avoids the problem that the collected cycling routes may violate traffic rules such as going in the opposite direction, making the cycling routes inaccurate and compliant, thereby affecting the accuracy of updating the historical navigation route.

[0070] Furthermore, in step 4 above, a dual-cycle update is adopted: that is, after obtaining the latest current navigation route, the corresponding historical navigation data is updated using a large and small cycle method. This can not only ensure the timeliness of the update, but also save computing overhead. In actual processing, after obtaining the latest edge weight information of the abstract graph model, the shortest path information between abstract node pairs is updated using a large and small cycle method.

[0071] The process of establishing the abstract graph model in the above step 1 is specifically as follows: obtain the intersection points in the historical navigation routes in the map navigation cache database, and use the DB-SCAN clustering algorithm to extract the core nodes as the abstract nodes of the abstract graph model; if there are two abstract nodes that the historical navigation route passes through successively, it is recorded that there is an abstract edge, and the edge weight of the abstract edge is the navigation distance of the historical navigation route within the two abstract nodes, and the historical navigation route is mapped to the abstract node sequence of the abstract graph model according to the abstract nodes passed through.

[0072] The edge weight fitting process in the above step 2 is specifically as follows: first, the rider's original riding route is subjected to threshold-based outlier detection and filtering to remove invalid riding routes, and a valid riding route is obtained. Then, the valid rider trajectory data and historical navigation routes corresponding to each abstract edge of the abstract graph model obtained in step 1 are input into the deep learning edge weight fitting deep learning model for deep learning estimation processing. The deep learning edge weight fitting deep learning model outputs the fitting result of the optimal route distance of the current abstract edge. For abstract edges with multiple riding routes, the DB-SCAN clustering algorithm is used to cluster the fitting results output by the deep learning edge weight fitting deep learning model, and the class with the largest number is taken as the final fitting result.

[0073] The path selection in step 3 above is specifically as follows: Select a set of navigation routes for priority update, and use them to query the third-party map service provider again to obtain the corresponding current latest navigation routes. The pseudo-code of the specific selection algorithm for navigation routes is as follows, which specifically includes the following steps (the path selection is all implemented based on the operation of the abstract graph model):

[0074] Step 31, calculate the update benefit of each edge based on the usage frequency and change amount of the edge;

[0075] Step 32, select paths so that the sum of the update benefits of the edges included in the selected paths is the largest;

[0076] Step 33, calculate the sum of the update benefits of the edges included in each path, sort the paths from largest to smallest according to the benefit, and select paths in the order of path sorting. When the edges included in the selected path have been included in the path selected in step 32, skip this path and continue to select the next path; this way of selecting paths can cover as many edges not included in other selected paths as possible on the basis of considering the update benefit, and can avoid the situation that the selected paths contain too many common edges and an edge is updated and queried multiple times;

[0077] Step 34, repeat steps 32 to 33 until the number of selected paths reaches the number of update budgets.

[0078]

[0079] The double-period update in step 4 above is specifically as follows: It is divided into two types of updates. The first update is the first-period (short-period update, such as daily update) update. When the update in the first period is required, if the abstract edges included in the historical navigation route remain unchanged, use the edge weights of the obtained latest abstract edges to update the navigation distances of the corresponding historical navigation routes in the map navigation cache database;

[0080] The second update is the second-period (long-period update, such as weekly update) update: When the update in the second period is required, according to all the abstract edge weights obtained from the first-period updates after the last second-period update, use the Dijsktra shortest path algorithm on the abstract graph to recalculate the latest navigation distances of all historical navigation routes, and use the obtained latest navigation distances to update all historical navigation routes.

[0081] The method of the present invention can be deployed in a large-scale database server and can be implemented and deployed in software.

[0082] Those of ordinary skill in the art will understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0083] As mentioned above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for updating historical navigation routes in a map navigation cache database, characterized in that, it includes: Step 1, establish an abstract graph model: Obtain historical navigation routes in the map navigation cache database, and establish an abstract graph model according to the historical navigation routes; The historical navigation routes are established as an abstract graph model in the following manner, including: Establish an abstract graph model with the intersection points in the historical navigation routes, map the historical navigation routes into an abstract node sequence of the abstract graph model, and use the navigation distance of the historical navigation routes as the edge weight of the abstract edges of the abstract graph model; In this step 1, obtain the intersection points in the historical navigation routes, extract the core nodes in the intersection points through clustering processing as abstract nodes to establish an abstract graph model; Use the historical navigation routes passing through at least two abstract nodes of the abstract graph model as an abstract edge, and the edge weight of this abstract edge is the navigation distance between the abstract nodes; Map the historical navigation routes into an abstract node sequence of the abstract graph model according to the passing abstract nodes; Step 2, determine the historical navigation routes to be updated through the abstract graph model: Obtain valid cycling routes, and through deep learning, fit and cluster the valid cycling routes and historical navigation routes corresponding to each abstract edge of the abstract graph model to obtain a fitting result of the optimal route distance for the current abstract edge, and determine the abstract edges to be updated according to the fitting result, and determine the navigation routes to be updated according to the determined abstract edges; The following is the way to perform deep learning to fit and cluster the valid cycling routes and historical navigation routes corresponding to each abstract edge of the abstract graph model to obtain a fitting result of the optimal route distance for the current abstract edge, including: Through deep learning, fit the valid cycling routes and historical navigation routes corresponding to each abstract edge of the abstract graph to obtain a fitting result of the optimal route distance for the current abstract edge; Perform clustering processing on the fitting results output by the abstract edges with multiple cycling routes obtained from the fitting processing, and take the class with the largest number as the fitting result; Step 3, obtain the latest navigation routes corresponding to the historical navigation routes to be updated: According to the historical navigation routes to be updated, query a third-party map service provider to obtain the corresponding current latest navigation routes; Step 4, update the historical navigation routes: Update the corresponding historical navigation routes in the map navigation cache database according to the current latest navigation routes obtained in Step 3.

2. The method for updating historical navigation routes in a map navigation cache database according to claim 1, characterized in that, Step 3 further includes: Select a group of navigation routes to be preferentially updated from the historical navigation routes to be updated, and query a third-party map service provider through the navigation routes to obtain the corresponding current latest navigation routes.

3. The method for updating historical navigation routes in a map navigation cache database according to claim 1, characterized in that, The following is the way to select a group of navigation routes to be preferentially updated from the historical navigation routes to be updated, including: Step 31. Calculate the update benefit of each abstract edge based on the usage frequency and variation amount of the abstract edges of the abstract graph model; Step 32. Select paths from the abstract edges of the abstract graph model such that the sum of the update benefits of the abstract edges included in the selected paths is the largest; Step 33. Calculate the sum of the update benefits of the edges included in each path, sort the paths in descending order of benefit, and sequentially select paths according to the path sorting. When the edges included in the selected path have already been included in the path selected in Step 32, skip this path and continue to select the next path; Step 34. Repeat Steps 32 to 33 until the number of selected paths reaches the number of the update budget.

4. The method for updating the historical navigation routes of the map navigation cache database according to any one of claims 1 to 3, wherein, the said Step 4 further includes: According to the currently obtained latest navigation route, update the corresponding historical navigation routes in the map navigation cache database at preset different periods.

5. The method for updating the historical navigation routes of the map navigation cache database according to claim 4, wherein, in the said Step 4, the map navigation cache database is respectively updated in the first period and the second period according to preset first and second periods, wherein, the duration of the first period is less than that of the second period; The first-period update is: when it is necessary to update at the first period, if the abstract edges included in the historical navigation route remain unchanged, use the edge weights of the obtained latest abstract edges to update the navigation distance of the historical navigation route; The second-period update is: when it is necessary to update at the second period, according to the edge weights of all the abstract edges obtained by the first-period updates after the previous second-period update, recalculate the latest navigation distances of all the historical navigation routes on the abstract graph model in the shortest path manner, and update the corresponding historical navigation routes with the obtained new navigation distances.

6. The method for updating the historical navigation routes of the map navigation cache database according to claim 5, wherein, the said shortest path manner uses the Dijsktra shortest path algorithm to calculate the shortest path; the said first period is one day; the said second period is one week.

7. The method for updating the historical navigation routes of the map navigation cache database according to claim 1, wherein, in the said Step 2, the effective riding routes are obtained in the following manner, including: First, obtain the original riding routes of the rider, and perform anomaly point detection and filtering on the original riding routes based on a preset threshold to remove invalid riding routes, thereby obtaining effective riding routes.

8. The method for updating the historical navigation routes of the map navigation cache database according to claim 1 or 2, wherein, in the said Step 2, the deep learning uses an edge weight fitting deep learning model composed of a first preprocessing module, a first LSTM network, a second preprocessing module, a second LSTM network, a Bi-LSTM network, a first MLP network, a Fast-DTW module, a similarity measurement module, a basic feature module, and a second MLP network, wherein, The input of the first preprocessing module is the historical navigation route, and the output of the first preprocessing module is connected to the first LSTM network; The input of the second preprocessing module is the cycling route, and the output of the second preprocessing module is connected to the second LSTM network; The outputs of the first LSTM network and the second LSTM network are connected to the Bi-LSTM network, and the output of the Bi-LSTM network is connected to the first MLP network; The input of the Fast-DTW module is the historical navigation route and the cycling route, and the output of the Fast-DTW module is sequentially connected to the similarity measurement module and the basic feature module; The outputs of the first MLP network and the basic feature module are both connected to the second MLP network, and the output of the second MLP network is the predicted navigation distance.

Images