A method, device and system for obtaining a navigation route

Abstract

The present invention discloses a navigation route acquisition method, device, and system. The method comprises: recalling multiple route-calculation target navigation routes based on a route calculation request and at least one set of single route calculation target weight combinations, wherein the set of single route calculation target weight combinations includes weights corresponding to at least two single route calculation targets; recalling single route calculation target navigation routes based on the route calculation request and a determined single route calculation target; and sorting the multiple route calculation target navigation routes and single route calculation target navigation routes to determine a recommended navigation route that is pushed to the user. The present invention solves the problem in the prior art of using a single route calculation target to recall a single route calculation target navigation route, resulting in the resulting navigation route not meeting the user's navigation needs.

Description

Technical Field

[0001] The present invention relates to the field of electronic map navigation, and in particular to a navigation route acquisition method, device and system. Background Art

[0002] In existing technologies, navigation routes generally use a recall strategy based on a single route calculation objective. Multiple routes recalled for each single route calculation objective are sorted and then pushed to the user. However, routes recalled based on a recall strategy based on a single route calculation objective cannot simultaneously optimize multiple route calculation objectives. This is because navigation routes that simultaneously meet multiple route calculation objectives cannot be prioritized for recall based on a single route calculation objective recall strategy. For example, when the route calculation objective is shortest time or shortest distance, the route with the shortest time or shortest distance is prioritized for push notification to the user. Routes that are not optimal for either the shortest time or shortest distance objectives, but have a better overall performance for both objectives, are unlikely to be prioritized for recall and pushed to the user. However, when users use travel applications for navigation route planning, they have multifaceted route requirements. For example, they may simultaneously consider road congestion, driving distance, driving time, tolls, and more. However, existing technologies based on a single route calculation objective cannot support such product requirements. Summary of the Invention

[0003] In view of the above problems, the present invention is proposed to provide a navigation route acquisition method, device and system that overcome the above problems or at least partially solve the above problems.

[0004] In a first aspect, an embodiment of the present invention provides a method for obtaining a navigation route, comprising the following steps:

[0005] Recalling a multi-path calculation target navigation route based on a path calculation request and at least one set of single path calculation target weight combinations, wherein the set of single path calculation target weight combinations includes weights corresponding to at least two single path calculation targets respectively;

[0006] Recalling a navigation route of the single route calculation target according to the route calculation request and the determined single route calculation target;

[0007] The multi-route calculation target navigation routes and the single-route calculation target navigation routes are sorted to determine a recommended navigation route to be pushed to the user.

[0008] In one or some optional embodiments, the method further comprises:

[0009] Determine a single route calculation target corresponding to the route preference indicator according to the route preference indicator carried in the route calculation request;

[0010] If more than two single route calculation targets are determined, the step of recalling navigation routes with multiple route calculation targets is performed.

[0011] In one or some optional embodiments, the step of recalling navigation routes with multiple path calculation targets based on the path calculation request and at least one set of single path calculation target weight combinations includes:

[0012] According to the start and end point information carried in the route calculation request, the route segments included in the multi-target navigation route to be recalled are determined using each set of single route calculation target weight combinations;

[0013] Weighting the weights of the route segments included in each multi-target navigation route to be recalled to obtain the route weight of the corresponding multi-target navigation route to be recalled;

[0014] A preset number of routes are selected from the multi-target navigation routes to be recalled in ascending order of route weights as multi-route target navigation routes.

[0015] In one or some optional embodiments, determining the route segments of the multi-target navigation route to be recalled based on the start and end point information carried in the route calculation request and using each set of single route calculation target weight combinations includes:

[0016] Divide the road network into multiple grids in advance;

[0017] Based on the start and end point information carried in the route calculation request, determine all the grids passed from the start point to the end point;

[0018] Using the weights of each single-route target in each set of single-route target weight combinations, weight the attribute values ​​of each single-route target in each route segment corresponding to each grid from the start point to the end point, and obtain the weight of each route segment corresponding to each grid;

[0019] For each grid, a preset number of route segments are selected in order of weight from small to large as the route segments of the multi-target navigation route to be recalled.

[0020] In one or some optional embodiments, the route preference indicator includes:

[0021] Shortest travel time, shortest driving distance, least number of traffic lights, least turning movements and avoidance of congestion.

[0022] In one or some optional embodiments, sorting the multi-route calculation target navigation routes and the single-route calculation target navigation routes to determine a recommended navigation route to be pushed to the user includes:

[0023] Inputting the at least one multi-route target navigation route and the at least one single-route target navigation route into a trained ranking model, and outputting ranking scores for the at least one multi-route target navigation route and the at least one single-route target navigation route;

[0024] Sort by ranking score and get the recommended navigation route pushed to the user.

[0025] In one or some optional embodiments, the ranking model is obtained by:

[0026] According to the preset navigation target combination, the loss function is set using the linear rectification ReLU function;

[0027] Determine a Pareto route set based on multiple historical route calculation requests using a selected multi-objective algorithm, and recall at least one multi-route calculation objective navigation route from the Pareto route set based on at least one set of single route calculation objective weight combinations initially set or obtained during training;

[0028] Recalling at least one single-route-calculation-target navigation route according to the multiple historical route calculation requests and the determined single-route-calculation target;

[0029] Inputting the obtained route set consisting of at least one multi-route calculation target navigation route and at least one single-route calculation target navigation route into a sorting model, and determining the optimal route in the route set and the optimal route in the at least one single-target navigation route;

[0030] According to the attribute value of the optimal route in the route set and the attribute value of the optimal route in the single-route target navigation route, the loss value of the loss function is determined, and the weights of each single-route target in each set of single-route target weight combinations and the parameters of the sorting model are adjusted according to the loss value of the loss function. The above process is repeated until at least one set of single-route target weight combinations and the parameters of the sorting model are obtained when the loss function is minimized.

[0031] In one or some optional embodiments, the navigation objective combination includes an optimization objective and at least one objective serving as a constraint condition, and the optimization objective and the objective serving as a constraint condition are selected from the following objectives:

[0032] Estimated travel time, travel distance, tolls, congestion distance, side roads distance, clear road distance, number of traffic lights and number of turns.

[0033] In a second aspect, an embodiment of the present invention provides a navigation route acquisition device, comprising: a multi-target recall module, configured to retrieve a navigation route with multiple route calculation targets based on a route calculation request and at least one set of single route calculation target weight combinations, wherein the set of single route calculation target weight combinations includes weights corresponding to at least two single route calculation targets;

[0034] A single target recall module, configured to recall a single target navigation route according to the route calculation request and the determined single target;

[0035] The sorting module is used to sort the navigation routes with multiple route calculation targets and the navigation routes with a single route calculation target, and determine a recommended navigation route to be pushed to the user.

[0036] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having computer instructions stored thereon, wherein the instructions, when executed by a processor, implement the navigation route acquisition method as described above.

[0037] In a fourth aspect, an embodiment of the present invention provides a computer device, comprising: a processor, and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the navigation route acquisition method as described above.

[0038] The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least:

[0039] Embodiments of the present invention provide a navigation route acquisition method, device, and system. The navigation route acquisition method uses at least one set of single-route-calculation target weight combinations to recall multi-route-calculation target navigation routes, sorts the obtained multi-route-calculation target navigation routes and the single-route-calculation target navigation routes obtained based on the determined single-route-calculation target, and determines a recommended navigation route to be pushed to the user. Not only does the number of navigation routes increase, but also navigation routes that are relatively good in terms of multiple route calculation targets can be pushed to the user. The pushed recommended navigation routes are more in line with the user's actual route planning needs. Moreover, according to the needs of actual route planning, multiple route calculation targets can be considered at the same time, and the number of single-route-calculation target weight combinations can be planned in advance according to actual needs. The number of recalled multi-route-calculation target navigation routes can be adjusted according to the number of recommended navigation routes pushed to the user.

[0040] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0041] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0043] Figure 1 This is a flow chart of a method for obtaining a navigation route in an embodiment of the present invention;

[0044] Figure 2This is a flow chart of a method for recalling a multi-route target navigation route according to an embodiment of the present invention;

[0045] Figure 3 A flowchart of a process for determining a route segment of a multi-target navigation route to be recalled in a multi-route target navigation route recall method according to an embodiment of the present invention;

[0046] Figure 4 This is a flow chart of the sorting model training process in an embodiment of the present invention;

[0047] Figure 5 A schematic diagram of a sorting model provided in an embodiment of the present invention;

[0048] Figure 6 Schematic diagram of a navigation route acquisition device according to an embodiment of the present invention;

[0049] Figure 7 Schematic diagram of a navigation route acquisition system in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0051] In view of the problems existing in the above-mentioned prior art, the embodiment of the present invention provides a method for obtaining a navigation route. Figure 1 As shown, the following steps are included:

[0052] S101: Recalling a multi-path calculation target navigation route according to a path calculation request and at least one set of single path calculation target weight combinations, wherein the set of single path calculation target weight combinations includes weights corresponding to at least two single path calculation targets.

[0053] In the above step S101, at least one set of single-path calculation target weight combinations is obtained through machine learning model training based on a pre-set multiple-path calculation target combination containing at least two single-path calculation targets. Before the machine learning model training begins, the value K of the number of sets of single-path calculation target weight combinations can be pre-set according to actual needs, where K is a positive integer greater than or equal to 1, so as to recall an appropriate number of multiple-path calculation target navigation routes. When a path calculation request is received, the process of recalling multiple-path calculation target navigation routes can be to perform path calculation based on the start and end point information in the path calculation request and each set of single-path calculation target weight combinations in the K sets of single-path calculation target weight combinations, obtain the path calculation result corresponding to each set of single-path calculation target weight combinations, and determine at least one multiple-path calculation target navigation route corresponding to each set of single-path calculation target weight combinations based on the path calculation result.

[0054] S102: Recalling a navigation route of the single route calculation target according to the route calculation request and the determined single route calculation target.

[0055] In the above step S102, based on the route calculation request and the determined single route calculation target, a single target recall strategy in the prior art can be used to recall at least one single route calculation target navigation route. For example, the route calculation can be performed using a single target and Dijkstra algorithm in the prior art, or using a cell-based routing (CBR) algorithm.

[0056] S103: Sort the navigation routes with multiple route calculation targets and the navigation routes with a single route calculation target, and determine a recommended navigation route to be pushed to the user.

[0057] In the above step S103, the multi-route target navigation routes and the single-route target navigation routes obtained in the above steps S101 and S102 are input into the trained sorting model, sorted according to the pre-set sorting index, and at least one navigation route is obtained according to the sorting result. The navigation client can obtain the at least one navigation route, and the user can select the appropriate navigation route according to actual needs to complete the route planning.

[0058] The navigation route acquisition method provided by the embodiment of the present invention uses at least one set of single-route-calculation target weight combinations to recall multi-route-calculation target navigation routes, sorts the obtained multi-route-calculation target navigation routes and the single-route-calculation target navigation routes obtained based on the determined single-route-calculation target, and determines the recommended navigation routes to be pushed to the user. Not only does this increase the number of navigation routes, but it can also push navigation routes that are relatively superior in terms of multiple single-route-calculation targets to the user. The pushed recommended navigation routes are more in line with the user's actual route planning needs. Moreover, multiple single-route-calculation targets can be considered simultaneously according to the needs of actual route planning, and the number of single-route-calculation target weight combinations can be planned in advance according to actual needs. The number of recalled multi-route-calculation target navigation routes can be adjusted according to the number of recommended navigation routes pushed to the user.

[0059] The specific embodiments of the present invention are described in detail below.

[0060] Example 1:

[0061] In a specific embodiment, the navigation route acquisition method provided in the above embodiment may further include, before executing the step of recalling the multi-route target navigation route:

[0062] Determine a single route calculation target corresponding to the route preference indicator according to the route preference indicator carried in the route calculation request;

[0063] If more than two single route calculation targets are determined, the step of recalling navigation routes with multiple route calculation targets is performed.

[0064] In an embodiment of the present invention, the user's route preference indicator includes one or more of the following route preference indicators: shortest travel time, shortest driving distance, fewest number of traffic lights, fewest turns, congestion avoidance, toll avoidance, no expressways, and expressway priority. Different users have different route preferences. When a route calculation request is received, one or more single route calculation targets corresponding to the route preference indicators can be determined based on the route preference indicators carried in the calculation request. When more than two single route calculation targets are determined, the group of more than two single route calculation targets is combined as a multiple route calculation target combination, and at least one corresponding set of single route calculation target weight combinations is selected based on the multiple route calculation target combination.

[0065] In a specific embodiment, referring to Figure 2 As shown in the figure, the specific process of recalling the multi-route target navigation route is as follows:

[0066] S201: Determine the route segments of the multi-target navigation route to be recalled based on the start and end point information carried in the route calculation request and using each set of single route calculation target weight combinations;

[0067] S202: Weighting the weights of the route segments included in each multi-destination navigation route to be recalled to obtain a route weight of the corresponding multi-destination navigation route to be recalled;

[0068] S203: Selecting a preset number of routes as multi-route target navigation routes from the multi-target navigation routes to be recalled in ascending order of route weights.

[0069] In a specific embodiment, referring to Figure 3 As shown, in the above step S201, based on the start and end point information carried in the route calculation request, the route segments included in the multi-target navigation route to be recalled are determined by using each set of single route calculation target weight combinations, which can be achieved through the following process:

[0070] S301: Divide the road network into multiple grids in advance;

[0071] S302: Determine all grids passed from the starting point to the end point based on the start and end point information carried in the path calculation request;

[0072] S303: Using the weights of each single path calculation target in each set of single path calculation target weight combinations, weight the attribute values ​​of each single path calculation target in each route segment corresponding to each grid passed through from the start point to the end point to obtain the weight of each route segment corresponding to each grid;

[0073] S304: For each grid, select a preset number of route segments in order of weight from small to large as the route segments of the multi-destination navigation route to be recalled.

[0074] In an embodiment of the present invention, to facilitate determining the route from the starting point to the end point of a route calculation request, the road network is pre-divided into multiple grids. When recalling a route, the grids where the starting point and end point of the route calculation request are located are first determined, and then all grids traversed from the starting point to the end point are determined. The weights corresponding to each single route calculation target in each set of single route calculation target weight combinations are used to weight the attribute values ​​of the corresponding targets in each route segment corresponding to each grid traversed from the starting point to the end point, respectively, to obtain the weights of multiple route calculation targets for each route segment corresponding to each grid. For each grid, the multiple route calculation targets are sorted in ascending order according to their weights, and a preset number of route segments are selected as the route segments of the multi-target navigation route to be recalled corresponding to that grid. The weights of the route segments of the multi-target navigation route to be recalled corresponding to all grids from the starting point to the end point are summed to obtain all possible multi-target navigation routes to be recalled from the starting point to the end point. The multi-target navigation routes to be recalled are sorted in ascending order according to their weights, and a preset number of routes are selected from all possible multi-target navigation routes to be recalled as the multi-target navigation routes.

[0075] In a specific embodiment, after the road network is pre-divided into multiple grids, when K sets of single-path calculation target weight combinations are used to perform route recall, for each route calculation request, the grids where the starting point and end point of the route calculation request are located are first determined, and then all grids passed from the starting point to the end point are determined; for each set of single-path calculation target weight combinations in the K sets of single-path calculation target weight combinations, the weights of each single-path calculation target in the single-path calculation target weight combination are used to weight the attribute values ​​of the corresponding targets in each route segment corresponding to each grid passed from the starting point to the end point, respectively, to obtain the multi-path calculation target weights of each route segment corresponding to each grid, and for each grid, the Dijkstra algorithm is used to calculate the weights of the target attributes in the route segment corresponding to each grid. The method is used to screen out a shortest route segment; the weights of the shortest route segments corresponding to all grids from the starting point to the end point are summed to obtain all the multi-target navigation routes to be recalled from the starting point to the end point, and the route with the smallest weight is selected from all the multi-target navigation routes to be recalled as the multi-target navigation route corresponding to the set of single-route target weight combinations. Then, K sets of single-route target weight combinations can recall K multi-route target navigation routes. However, considering that there may be more than two sets of single-route target weight combinations recalling the same route, the final number of multi-target navigation routes obtained is less than or equal to K, where K is a positive integer greater than or equal to 1.

[0076] Example 2

[0077] In a specific embodiment, the process described in step S103 above of sorting the navigation routes with multiple route calculation targets and the navigation routes with a single route calculation target to determine a recommended navigation route to be pushed to the user includes:

[0078] Inputting the at least one multi-route target navigation route and the at least one single-route target navigation route into a trained ranking model, and outputting ranking scores for the at least one multi-route target navigation route and the at least one single-route target navigation route;

[0079] Sort by ranking score and get the recommended navigation route pushed to the user.

[0080] In this embodiment of the present invention, the ranking scores described herein may be calculated by a ranking model for the at least one multi-route target navigation route and the at least one single-route target navigation route, and outputting a ranking score for each route. The ranking model then sorts the routes from high to low based on the ranking scores, and selects a preset number of navigation routes with the highest ranking scores from the ranking results as recommended navigation routes to be pushed to the user.

[0081] In a specific embodiment, the ranking model described in the above embodiment is pre-trained in the following manner:

[0082] According to the preset navigation target combination, the loss function is set using the linear rectification ReLU function;

[0083] Based on multiple historical route calculation requests, a Pareto route set is determined using a selected multi-objective algorithm, and K or fewer multi-route calculation objective navigation routes are screened from the Pareto route set based on K sets of single route calculation objective weight combinations initially set or obtained during training.

[0084] Recalling at least one single-route-calculation-target navigation route according to the determined single-route-calculation targets determined by the plurality of historical route calculation requests;

[0085] Inputting the obtained route set consisting of K or fewer multi-route calculation target navigation routes and at least one single-route calculation target navigation route into a sorting model, and determining the optimal route in the route set and the optimal route in the at least one single-route calculation target navigation route;

[0086] According to the attribute value of the optimal route in the route set and the attribute value of the optimal route in the single-route target navigation route, the loss value of the loss function is determined, and the weights of each single-route target in each set of single-route target weight combinations and the parameters of the machine learning model are adjusted according to the loss value of the loss function. The above process is repeated until K sets of single-route target weight combinations and machine learning model parameters are obtained when the loss function is minimized.

[0087] In a specific embodiment, the navigation objective combination includes an optimization objective and at least one objective serving as a constraint condition, and the optimization objective and the objective serving as a constraint condition are selected from the following objectives:

[0088] Estimated travel time, travel distance, tolls, congestion distance, side roads distance, clear road distance, number of traffic lights and number of turns.

[0089] Example 3:

[0090] For a route calculation request, selecting a single route calculation target and calculating the route using the Dijkstra algorithm does not satisfy the user's navigation needs when planning a route. This is because users often have multiple requirements for routes when planning navigation routes in travel apps. For example, they may consider traffic congestion, driving distance, driving time, and tolls. Suppose there are three routes p1, p2, and p3 from a starting point s to a destination t: route p1 = {time = 20 minutes, distance = 10 kilometers}, route p2 = {time = 12 minutes, distance = 11 kilometers}, and route p3 = {time = 10 minutes, distance = 18 kilometers}. If, when recalling a route, the single route calculation target with the shortest distance is selected and calculated using the Dijkstra algorithm, route p1 can theoretically be recalled. Similarly, the single route calculation target with the shortest time and the Dijkstra algorithm can theoretically be recalled for route p3. However, using the Dijkstra algorithm for route recall based on the current single-objective route calculation objective only allows recall of routes P1 and P3, but not P2. However, P2 is the preferred navigation route based on both time and distance, making it the user's preferred route during actual route planning. The inventors of the present invention have discovered that, when performing route calculation using both distance and time as the two objectives, and utilizing a conventional multi-objective shortest path algorithm, it is possible to recall P2.

[0091] Specifically, for a given route calculation request, a Pareto route set corresponding to the at least two route calculation objectives can be obtained using an existing multi-objective algorithm based on the start and end points of the route calculation request and a combination of multiple route calculation objectives including at least two route calculation objectives. Assume that the Pareto route set = {p1, p2, ...pm}, where p1, p2, ...pm are candidate routes in the Pareto route set, and m is the number of candidate routes in the Pareto route set. The Pareto route set obtained using the multi-objective algorithm is the optimal route set corresponding to the routing request. No alternative route in the Pareto route set can dominate any other alternative route in the same Pareto route set (for example, assuming that the distance of route p1 is shorter than that of route p2, and the estimated arrival time of route p2 is shorter than the estimated arrival time of route p1, then p1 cannot dominate p2, and p2 cannot dominate p1). In addition, for a given routing request, other routes that do not belong to the Pareto route set will be dominated by at least one alternative route in the Pareto set (for example, assuming that a route pn is obtained based on the start and end points of the routing request, if the distance and estimated arrival time of route px are both greater than the distance and estimated arrival time of route p2 in the Pareto route set, then route px is dominated by route p2, so route px does not belong to the Pareto route set).

[0092] In a specific embodiment, if there are four routes for a route calculation request: p1 = {distance = 10 kilometers, time = 20 minutes}, p2 = {distance = 12 kilometers, time = 12 minutes}, p3 = {distance = 20 kilometers, time = 10 minutes}, p4 = {distance = 15 minutes, time = 15 minutes}, according to the distance and time attribute values ​​of each route, it can be seen that: route p1 has the shortest distance, route p3 has the shortest time, and route p2 has a shorter time than p1 and a shorter distance than p3, that is, routes p1, p2 and p3 cannot dominate each other, and the time and distance of route p4 are both greater than route p2, so route p4 is dominated by route p2. If the Pareto route set is obtained by route calculation based on the multi-objective algorithm, then the Pareto route set = {p1, p2, p3}.

[0093] The inventors of the present invention also discovered through experiments that while a multi-objective algorithm, which generates a Pareto route set based on a route calculation request and a preset combination of multiple route calculation targets, can generate at least one navigation route with multiple route calculation targets, the performance of the computer system used to recall the navigation route slows down as the number of individual route calculation targets in the preset combination increases. Therefore, this method cannot be directly applied to real-time interactive systems and is not suitable for real-time navigation route recall. This is because the multi-objective problem solved using the multi-objective method is a non-deterministic polynomial problem (NP). This complex algorithmic problem, which cannot be solved in polynomial time, is time-consuming and severely consumes computer system resources. Furthermore, the larger the number of individual route calculation targets, the more alternative routes are generated, and the larger the Pareto route set is. This increases the resource consumption of the computing system, resulting in slower system performance and failing to meet the response requirements of real-time interactive systems. Therefore, the multi-objective algorithm cannot be applied to navigation route recall on electronic maps.

[0094] In order to overcome the problem that directly using a multi-target algorithm cannot be applied to real-time route recall on electronic maps, the inventors of the present invention have improved the route recall method by pre-setting a multi-route target combination containing at least two single-route targets, and using a machine learning model for training to obtain K sets of single-route target weight combinations corresponding to the multi-route target combination, where K is a positive integer greater than or equal to 1. Route calculation is performed based on the starting and end points in the route calculation request, and each set of single-route target weight combinations in the K sets of single-route target weight combinations, to obtain less than or equal to K multi-route target navigation routes. This navigation route recall method can be applied to a real-time interactive system for map data, and the recalled multi-route target navigation routes meet the user's expectation of simultaneously considering multiple single-route targets when planning routes.

[0095] Before training K sets of single-route target weight combinations through a machine learning model, the size of the K value can be determined based on the actual needs of navigation route recall. It should be noted that the larger the K value, the more multi-route target navigation routes may be obtained during route recall. However, the larger the K value, the greater the computational workload of the computer system, the longer it will take, the more computer resources will be occupied, and the higher the performance requirements for the computer system of the map data service. Therefore, in order to meet the time requirements of the real-time interactive system, the K value cannot be set too large.

[0096] In one embodiment, referring to Figure 4 As shown, the machine learning model in this embodiment is a ranking model, which is pre-trained in the following way:

[0097] S401: According to the preset navigation target combination, a loss function is set using a linear rectification ReLU function; obtaining;

[0098] S402: Determine a Pareto route set based on multiple historical route calculation requests using a selected multi-objective algorithm; the Pareto route set includes multiple candidate routes that meet preset multi-objective combination requirements; and recall at least one multi-route calculation objective navigation route from the Pareto route set based on at least one set of single route calculation objective weight combinations initially set or obtained during training.

[0099] S403: Recalling at least one single-route-calculation-target navigation route according to the multiple historical route calculation requests and the determined single-route-calculation target;

[0100] S404: Inputting the obtained route set of at least one multi-route calculation target navigation route and at least one single-route calculation target navigation route into a sorting model, and determining the optimal route in the route set and the optimal route in the at least one single-target navigation route;

[0101] S405: Determine the loss value of the loss function based on the attribute value of the optimal route in the route set and the attribute value of the optimal route in the single-route target navigation route, adjust the weights of each single-route target in each set of single-route target weight combinations and the parameters of the sorting model based on the loss value of the loss function, and repeat the above process until at least one set of single-route target weight combinations and the parameters of the sorting model are obtained when the loss function is minimized.

[0102] In a specific embodiment, assuming a preset multiple route calculation target combination, including five single route calculation targets: fastest conventional route, shortest distance, shortest estimated time, fewest number of traffic lights, and fewest number of turning maneuvers, the attribute values ​​of each alternative route in the corresponding Pareto route set include a congestion avoidance attribute value, a time attribute value, a distance attribute value, a traffic light number attribute value, and a turning maneuver number attribute value. When K sets of single route calculation target weight combinations are initially set, where K is a positive integer greater than or equal to 1, each set of single route calculation target weight combinations includes weights corresponding to the five single route calculation targets: fastest conventional route, shortest distance, shortest estimated time, fewest number of traffic lights, and number of turning maneuvers. During the sorting model training process, the K sets of single route calculation target weight combinations are adjusted, that is, the weights of each single route calculation target in each set of single route calculation target weight combinations in the K sets of single route calculation target weight combinations are adjusted, where the sum of the weights of each single route calculation target in each set of single route calculation target weight combinations is 1.

[0103] In one embodiment, based on K sets of single-path calculation target weight combinations initially set or obtained during training, no more than K navigation routes with multiple path calculation targets are selected from the Pareto route set. In a specific implementation, for example, each set of single-path calculation target weight combinations in the K sets of single-path calculation target weight combinations can be weighted with each alternative route in the Pareto route set, that is, the weight of each single-path calculation target in each set of single-path calculation target weight combinations is multiplied by the target attribute value corresponding to each alternative route in the Pareto route set, and then the sum is calculated to obtain a comprehensive weight for each alternative route in the Pareto route set. All alternative routes in the Pareto route set are sorted according to the size of the comprehensive weight, and the route with the smallest comprehensive weight is selected as the navigation route with multiple path calculation targets selected by the set of single-path calculation target weight combinations. Since there may be a situation where two sets of single-path calculation objective weight combinations filter out the same multi-path calculation objective navigation route, when using K sets of single-path calculation objective weight combinations to filter out multi-path calculation objective navigation routes from the Pareto route set, less than or equal to K multi-path calculation objective navigation routes can be filtered out.

[0104] In a specific embodiment, when setting the loss function, the navigation objective combination used includes an optimization objective and at least one objective as a constraint condition, and the optimization objective and the objective as a constraint condition are any one or more of the following:

[0105] Estimated travel time, travel distance, tolls, congestion distance, number of traffic lights and turns along the route.

[0106] In a specific embodiment, the navigation objective combination can be, for example, an optimization objective of estimated travel time, with constraints of travel distance and toll, to ensure that the route distance and toll do not exceed the single-route target navigation route obtained using conventional methods, while also reducing the estimated travel time. In this case, the loss function can be set as Loss = feature_eta + lam_dis * tf.nn.relu (feature_dis - online_dis) + lam_toll * tf.nn.relu (feature_toll - online_toll). In the above formula, feature_eta is the time attribute value of the optimal route in the route set consisting of multi-route target navigation routes and single-route target navigation routes, feature_dis is the distance attribute value of the optimal route in the route set consisting of multi-route target navigation routes and single-route target navigation routes, feature_toll is the charging attribute value of the optimal route in the route set consisting of multi-route target navigation routes and single-route target navigation routes, online_dis is the distance attribute value of the optimal route in the single-route target navigation route, and online_toll is the charging attribute value of the optimal route in the single-route target navigation route; lam_dis*tf.nn.relu(feature_dis-online_dis) represents the ReLU function set according to the distance, which is used to compare feature_dis and online The size of ne_dis. If feature_dis-online_dis is greater than 0, the function value is the true value of feature_dis-online_dis. If feature_dis-online_dis is less than or equal to 0, the function value is 0. lam_toll*tf.nn.relu(feature_toll-online_toll) represents the ReLU function set according to the charge setting, which is used to compare the sizes of feature_toll and online_toll. If feature_toll-online_toll is greater than 0, the function value is the true value of feature_toll-online_toll. If feature_toll-online_toll is less than or equal to 0, the function value is 0.

[0107] Since K sets of single-path target weight combinations are trained according to the loss function set for the navigation target combination, the determination of the weights of each single-path target in each set of single-path target weight combinations conforms to the expected effect of the navigation target combination. When recalling multi-path target navigation routes through K sets of single-path target weight combinations, the recalled navigation routes have comprehensive multi-path targets that are better than the single-path target navigation routes recalled according to the single-path target recall strategy of the prior art.

[0108] In one embodiment, during the sorting model training process, the process of determining the optimal route in a route set consisting of at least one multi-route calculation target navigation route and at least one single-route calculation target navigation route in the sorting model, and the optimal route in at least one single-target navigation route, is as follows:

[0109] When the route set consisting of the obtained multi-route calculation target navigation routes and the single-route calculation target navigation routes is input into the sorting model, the sorting model determined according to the parameters of the sorting model can output the sorting scores of all the navigation routes in the route set, and all the navigation routes in the route set are sorted according to the size of the sorting scores. The navigation route with the highest sorting score is the optimal route in the route set. At the same time, the route with the highest sorting score among at least one single-route calculation target navigation route is the optimal route among the single-route calculation target navigation routes.

[0110] In one specific embodiment, the ranking model is a deep neural network (DNN) model. For example, the DNN model can be a multidimensional neural network that connects neurons and activation functions. The specific structure of the ranking model can be described in the prior art and is not specifically limited in the embodiments of the present invention.

[0111] The following is a specific example to illustrate the training process of the sorting model:

[0112] Reference Figure 5 As shown, a sorting model is established, including a route calculation request acquisition module 501, a Pareto set acquisition module 502, a multi-target route recall module 503, a single-target route recall module 504, and a sorting model to be trained 505. For multiple historical route calculation requests, such as multiple route calculation requests during the morning rush hour in Beijing, a multi-target navigation combination including five goals, namely, the fastest conventional route, the shortest distance, the shortest estimated time, the fewest number of traffic lights, and the fewest number of turning actions, is preset. The preset navigation goal combination is: to achieve the goal of shortening the estimated driving time while keeping the driving distance and the toll no higher than the single-target navigation route obtained by the existing single-target route calculation recall strategy. The loss function is set to Loss = feature_eta + lam_dis*tf.nn.relu(feature_dis-online_dis) + lam_toll*tf.nn.relu(feature_toll-online_toll). The specific training process of the sorting model is as follows:

[0113] The route calculation request acquisition module 501 acquires multiple historical route calculation requests during the morning rush hour in Beijing from the historical route calculation request data;

[0114] The Pareto set acquisition module 502 uses a multi-objective algorithm to obtain a Pareto route set that meets the preset multiple route calculation objective combinations for each historical route calculation request. Each candidate route in the Pareto route set is represented by attribute values ​​corresponding to the five route calculation objectives.

[0115] The multi-target route recall module 503 multiplies the weights of each single-route target in each of the K initialized single-route target weight combinations by the attribute values ​​of the corresponding targets of all the alternative routes in the Pareto set, and then sums the results to obtain a comprehensive weight for each alternative route. The comprehensive weights of all the alternative routes are sorted from low to high, so that a route with the lowest comprehensive weight can be obtained for each single-route target weight combination, i.e., the multi-route target navigation route. Because it is possible that several single-route target weight combinations may recall the same multi-route target navigation route, using K sets of single-route target weight combinations will obtain less than or equal to K multi-route target navigation routes.

[0116] While determining K or fewer multi-route-target navigation routes using the multi-target algorithm and K sets of single-route-target weight combinations, the single-target route recall module 504 recalls at least one single-route-target navigation route for each historical route calculation request and the determined single-route-target using the single-route-target recall strategy.

[0117] The obtained route set consisting of K or fewer multi-route calculation target navigation routes and at least one single-route calculation target navigation route is input into a sorting model 505 determined based on the initialized sorting model parameters. The sorting scores of all navigation routes in the route set are output, and the routes are sorted in descending order of the sorting scores. The optimal route in the route set and the optimal route among the single-route calculation target navigation routes are determined. The time attribute value, distance attribute value, and charge attribute value of the optimal route in the route set, as well as the distance attribute value and charge attribute value of the optimal route among the single-route calculation target navigation routes, are obtained. The loss value of the loss function is calculated. Based on the loss value of the loss function, the weights of each single-route calculation target in each of the K sets of single-route calculation target weight combinations and the parameters of the sorting model are adjusted.

[0118] Using the same method as in the above process, the multi-target route recall module 503 again obtains K or fewer multi-target navigation routes. This route set, consisting of the K or fewer multi-target navigation routes and at least one single-target navigation route, is input into the ranking model 505, which is determined based on the trained ranking model parameters. This module then obtains the optimal route for the corresponding route set and the optimal route among the single-target navigation routes. The loss function is then calculated, and the individual weights for each of the K sets of single-target weight combinations, as well as the ranking model parameters, are readjusted based on the loss function. This process continues until the K sets of single-target weight combinations and ranking model parameters that minimize the loss function are obtained, and the training of the ranking model 505 is terminated.

[0119] In an embodiment of the present invention, the training indicator of the sorting model is to reduce the loss value of the loss function by adjusting K sets of single-path calculation target weight combinations and the parameters of the sorting model, and find the optimal K sets of single-path calculation target weight combinations and sorting model parameters when minimizing the loss function. In other words, during the sorting model training process, the loss value of the loss function is optimized by adjusting the parameters of the K sets of single-path calculation target weight combinations and the sorting model, and the parameters of the K sets of single-path calculation target weight combinations and the sorting model when minimizing the loss function are obtained. When a route calculation request is received, the route is calculated using the obtained K sets of single-path calculation target weight combinations based on the start and end point information carried in the route calculation request. Based on the route calculation results, at least one multi-path calculation target navigation route corresponding to each set of single-path calculation target weight combinations can be determined; then, using the sorting model determined based on the parameters of the trained sorting model, the at least one multi-path calculation target navigation route and at least one single-path calculation target navigation route input are sorted, and finally at least one recommended navigation route is determined and pushed to the user.

[0120] Based on the same inventive concept, an embodiment of the present invention further provides a navigation route acquisition device, system, related storage medium and equipment. Since the principles of the problems solved by these devices, systems, related storage media and equipment are similar to those of the aforementioned route recall method, the implementation of this method, device, system, related storage medium and equipment can refer to the implementation of the aforementioned route recall method, and the repeated parts will not be repeated.

[0121] Based on the same inventive concept, this embodiment provides a navigation route acquisition device, which can be installed in a local server or a cloud server. By executing the above navigation route acquisition method, at least one navigation route is obtained and pushed to the navigation client for route planning and navigation. Figure 6 As shown, the navigation route acquisition device includes:

[0122] The multi-target recall module 601 is configured to retrieve a navigation route with multiple route calculation targets based on a route calculation request and at least one set of single route calculation target weight combinations, wherein the set of single route calculation target weight combinations includes weights corresponding to at least two single route calculation targets.

[0123] A single target recall module 602 is configured to recall a single target navigation route according to the route calculation request and the determined single target;

[0124] The sorting module 603 is configured to sort the navigation routes with multiple route calculation targets and the navigation routes with a single route calculation target, and determine a recommended navigation route to be pushed to the user.

[0125] In one embodiment, the multi-target recall module 601 may be further configured to determine a single route calculation target corresponding to the route preference indicator according to the route preference indicator carried in the route calculation request;

[0126] If more than two single route calculation targets are determined, the step of recalling navigation routes with multiple route calculation targets is performed.

[0127] The route preference indicators include: shortest travel time, shortest driving distance, least number of traffic lights, least turning movements and avoiding congestion.

[0128] In one embodiment, the multi-target recall module 601 is specifically configured to determine the route segments of the multi-target navigation route to be recalled based on the start and end point information carried in the route calculation request and using each set of single route calculation target weight combinations;

[0129] Weighting the weights of the route segments included in each multi-target navigation route to be recalled to obtain the route weight of the corresponding multi-target navigation route to be recalled;

[0130] A preset number of routes are selected from the multi-target navigation routes to be recalled in ascending order of route weights as multi-route target navigation routes.

[0131] In one embodiment, the multi-target recall module 601 is specifically configured to divide the road network into a plurality of grids in advance;

[0132] Based on the start and end point information carried in the route calculation request, determine all the grids passed from the start point to the end point;

[0133] Using the weights of each single path calculation target in each set of single path calculation target weight combinations, the attribute values ​​of each path calculation target in each route segment corresponding to each grid passed from the starting point to the end point are weighted to obtain the weight of each route segment corresponding to each grid;

[0134] For each grid, a preset number of route segments are selected in order of weight from small to large as the route segments of the multi-target navigation route to be recalled.

[0135] In one embodiment, the ranking module 603 may be specifically configured to input the at least one multi-route target navigation route and the at least one single-route target navigation route into a trained ranking model, and output ranking scores of the at least one multi-route target navigation route and the at least one single-route target navigation route;

[0136] Sort by ranking score and get the recommended navigation route pushed to the user.

[0137] In one embodiment, the navigation route acquisition device may further include a training module 600 for obtaining the ranking model by:

[0138] According to the preset navigation target combination, the loss function is set using the linear rectification ReLU function;

[0139] Determine a Pareto route set based on multiple historical route calculation requests using a selected multi-objective algorithm, and recall at least one multi-route calculation objective navigation route from the Pareto route set based on at least one set of single route calculation objective weight combinations initially set or obtained during training;

[0140] Recalling at least one single-route-calculation-target navigation route according to the multiple historical route calculation requests and the determined single-route-calculation target;

[0141] Inputting the obtained route set consisting of at least one multi-route calculation target navigation route and at least one single-route calculation target navigation route into a sorting model, and determining the optimal route in the route set and the optimal route in the at least one single-target navigation route;

[0142] According to the attribute value of the optimal route in the route set and the attribute value of the optimal route in the single-route target navigation route, the loss value of the loss function is determined, and the weights of each single-route target in each set of single-route target weight combinations and the parameters of the sorting model are adjusted according to the loss value of the loss function. The above process is repeated until at least one set of single-route target weight combinations and the parameters of the sorting model are obtained when the loss function is minimized.

[0143] The navigation objective combination includes an optimization objective and at least one objective serving as a constraint condition, and the optimization objective and the objective serving as a constraint condition are selected from the following objectives:

[0144] Estimated travel time, travel distance, tolls, congestion distance, side roads distance, clear road distance, number of traffic lights and number of turns.

[0145] Based on the same inventive concept, this embodiment also provides a navigation route acquisition system, referring to Figure 6 As shown, it includes: a server 1 and at least one client 2;

[0146] Client 2 is used to send a route calculation request and receive a recommended navigation route pushed by the server 1;

[0147] Server 1 is configured to recall navigation routes with multiple route calculation targets based on a route calculation request and at least one set of single route calculation target weight combinations, wherein one set of single route calculation target weight combinations includes weights corresponding to at least two single route calculation targets respectively; recall navigation routes with single route calculation targets based on the route calculation request and the determined single route calculation target; sort the navigation routes with multiple route calculation targets and the navigation routes with single route calculation targets, determine a recommended navigation route to be pushed to the user, and push the recommended navigation route to the client 2.

[0148] In an embodiment of the present invention, the server 1 is provided with the navigation route acquisition device described in the above embodiment, which is used to execute the above navigation route acquisition method, obtain at least one recommended navigation route to be pushed to the user, and push it to the client for route planning and navigation.

[0149] Based on the same inventive concept, this embodiment further provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the navigation route acquisition method as described in the above embodiment.

[0150] Based on the same inventive concept, this embodiment further provides a computer device, comprising: a processor and a memory for storing processor-executable commands; wherein the processor is configured to execute the navigation route acquisition method as described in the above embodiment.

[0151] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0152] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0155] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for obtaining a navigation route, wherein: include: Recalling a multi-path calculation target navigation route based on a path calculation request and at least one set of single path calculation target weight combinations, wherein the set of single path calculation target weight combinations includes weights corresponding to at least two single path calculation targets respectively; Recalling a navigation route of the single route calculation target according to the route calculation request and the determined single route calculation target; The multi-route calculation target navigation routes and the single-route calculation target navigation routes are sorted to determine a recommended navigation route to be pushed to the user.

2. The navigation route acquisition method according to claim 1, wherein: The method further comprises: Determine a single route calculation target corresponding to the route preference indicator according to the route preference indicator carried in the route calculation request; If more than two single route calculation targets are determined, the step of recalling navigation routes with multiple route calculation targets is performed.

3. The navigation route acquisition method according to claim 2, wherein: The step of recalling a navigation route with multiple path calculation targets based on the path calculation request and at least one set of single path calculation target weight combinations includes: According to the start and end point information carried in the route calculation request, the route segments included in the multi-target navigation route to be recalled are determined using each set of single route calculation target weight combinations; Weighting the weights of the route segments included in each multi-target navigation route to be recalled to obtain the route weight of the corresponding multi-target navigation route to be recalled; A preset number of routes are selected from the multi-target navigation routes to be recalled in ascending order of route weights as multi-route target navigation routes.

4. The navigation route acquisition method according to claim 3, wherein: The method of determining the route segments of the multi-target navigation route to be recalled based on the start and end point information carried in the route calculation request and using each set of single route calculation target weight combinations includes: Divide the road network into multiple grids in advance; Based on the start and end point information carried in the route calculation request, determine all the grids passed from the start point to the end point; Using the weights of each single-route target in each set of single-route target weight combinations, weight the attribute values ​​of each single-route target in each route segment corresponding to each grid from the start point to the end point, and obtain the weight of each route segment corresponding to each grid; For each grid, a preset number of route segments are selected in order of weight from small to large as the route segments of the multi-target navigation route to be recalled.

5. The navigation route acquisition method according to claim 2, wherein: The route preference indicators include: Shortest travel time, shortest driving distance, least number of traffic lights, least turning movements and avoidance of congestion.

6. The navigation route acquisition method according to claim 1, wherein: The step of sorting the navigation routes with multiple route calculation targets and the navigation routes with a single route calculation target to determine a recommended navigation route to be pushed to the user includes: Inputting at least one multi-route target navigation route and at least one single-route target navigation route into a trained ranking model, and outputting ranking scores of the at least one multi-route target navigation route and the at least one single-route target navigation route; Sort by ranking score and get the recommended navigation route pushed to the user.

7. The navigation route acquisition method according to claim 6, wherein: The ranking model is obtained in the following way: According to the preset navigation target combination, the loss function is set using the linear rectification ReLU function; Determine a Pareto route set based on multiple historical route calculation requests using a selected multi-objective algorithm, and recall at least one multi-route calculation objective navigation route from the Pareto route set based on at least one set of single route calculation objective weight combinations initially set or obtained during training; Recalling at least one single-route-calculation-target navigation route according to the multiple historical route calculation requests and the determined single-route-calculation target; Inputting the obtained route set consisting of at least one multi-route calculation target navigation route and at least one single-route calculation target navigation route into a sorting model, and determining the optimal route in the route set and the optimal route in the at least one single-target navigation route; According to the attribute value of the optimal route in the route set and the attribute value of the optimal route in the single-route target navigation route, the loss value of the loss function is determined, and the weights of each single-route target in each set of single-route target weight combinations and the parameters of the sorting model are adjusted according to the loss value of the loss function. The above process is repeated until at least one set of single-route target weight combinations and the parameters of the sorting model are obtained when the loss function is minimized.

8. The navigation route acquisition method according to claim 7, wherein: The navigation objective combination includes an optimization objective and at least one objective serving as a constraint condition, wherein the optimization objective and the objective serving as a constraint condition are selected from the following objectives: Estimated travel time, travel distance, tolls, congestion distance, side roads distance, clear road distance, number of traffic lights and number of turns.

9. A navigation route acquisition device, wherein: include: A multi-target recall module is used to retrieve a navigation route with multiple route calculation targets based on a route calculation request and at least one set of single route calculation target weight combinations, wherein the set of single route calculation target weight combinations includes weights corresponding to at least two single route calculation targets respectively; A single target recall module, configured to recall a single target navigation route according to the route calculation request and the determined single target; The sorting module is used to sort the navigation routes with multiple route calculation targets and the navigation routes with a single route calculation target, and determine a recommended navigation route to be pushed to the user.

10. A computer-readable storage medium having computer instructions stored thereon, wherein: When the instruction is executed by the processor, the navigation route acquisition method according to any one of claims 1 to 8 is implemented.

11. A computer device, wherein: include: A processor and a memory for storing processor-executable commands; wherein the processor is configured to execute the navigation route acquisition method according to any one of claims 1 to 8.

Images