Waypoint prediction method and apparatus, and waypoint prediction model training method and apparatus

Abstract

Disclosed in the embodiments of the present disclosure are a waypoint prediction method, a waypoint prediction model training method, and an apparatus. The waypoint prediction method comprises: acquiring a historical waypoint sequence of a navigated object, wherein the historical waypoint sequence comprises data records indexed by historical waypoints, and each data record comprises a historical waypoint and a corresponding historical scene feature sequence; acquiring a scene feature of a target navigation route of the navigated object and using same as the current scene feature; inputting the current scene feature and the historical waypoint sequence into a model, such that a recommendation representation module comprised in the model obtains, at least on the basis of the current scene feature and historical scene features in the historical waypoint sequence, recommendation representations of the historical waypoints corresponding to the historical scene features; a first scoring module comprised in the model obtaining a score of each historical waypoint on the basis of the recommendation representations of the historical waypoints; and at least on the basis of the score of each historical waypoint, determining a waypoint prediction result. The embodiments of the present disclosure can improve the prediction accuracy.

Description

Waypoint prediction method, waypoint prediction model training method and device

[0001] This disclosure claims priority to Chinese patent application filed on January 24, 2025, with application number 202510122021.9 and entitled "Method for predicting waypoints, method and apparatus for training waypoint prediction models", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This disclosure relates to the field of navigation technology, specifically to a waypoint prediction method, a waypoint prediction model training method, and an apparatus. Background Technology

[0003] Because map navigation services can provide users with efficient, low-cost travel routes and accurate route guidance, more and more applications are integrating map navigation-related service capabilities, such as travel applications, ride-hailing applications, and lifestyle service applications.

[0004] When users travel along or plan to travel along a route provided by a map navigation service, the aforementioned applications generally support users in searching for points of interest such as gas stations, charging stations, restaurants, subway stations, and rest areas along the travel route. This function can be called "search along the way" or "search along the route." After finding points of interest through "search along the way," users can select them as waypoints on their travel route.

[0005] While researching existing technologies, the inventors of this disclosure discovered that existing solutions for adding waypoints via route searching require users to first access a route search page, then manually enter or select search keywords, and finally choose a point of interest (POI) as a waypoint from the search results. This process requires multiple manual operations from the user, posing a safety hazard if the user is driving. To enable users to add waypoints more safely and conveniently, proactively recommending waypoints to users before or during navigation has become a hot research topic. How to accurately and proactively predict waypoints that meet the user's travel intentions is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] This disclosure provides a waypoint prediction method, a waypoint prediction model training method, and an apparatus.

[0007] In a first aspect, this disclosure provides a waypoint prediction method, wherein the method is used to predict waypoints for a target navigation route of a navigable object using a waypoint prediction model, the method comprising:

[0008] Obtain the historical waypoint sequence of the navigated object, wherein the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences;

[0009] Obtain the scene features of the target navigation route of the navigated object as the current scene features;

[0010] The current scene features and the historical waypoint sequence are input into the model. The model includes a recommendation representation module that, based at least on the current scene features and the historical scene features in the historical waypoint sequence, obtains a recommendation representation of the historical waypoints corresponding to the historical scene features.

[0011] The model includes a first scoring module that obtains a score for each historical route point based on the recommended representation of the historical route points.

[0012] The waypoint prediction result is determined based at least on the score of each historical waypoint.

[0013] Secondly, this invention provides a method for training a waypoint prediction model, comprising:

[0014] The system obtains the historical waypoint sequence of the navigable object, sample waypoints of the target's historical navigation route, and sample scene features of the sample waypoints; wherein, the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences, and the historical waypoints are historical waypoints in the historical navigation process before the target's historical navigation route;

[0015] The sample scene features and the historical waypoint sequence are input into the model. The recommendation representation module of the model obtains the recommendation representation of the historical waypoint corresponding to the historical scene features based at least on the sample scene features and the historical scene features in the historical waypoint sequence.

[0016] The recommended representation of the historical waypoints is input into the model, and the first scoring module of the model obtains a score for each historical waypoint based on the recommended representation of the historical waypoints.

[0017] The loss function adjusts the model parameters of the modules included in the model based at least on the score of each historical path point and the sample path points used as sample labels.

[0018] Thirdly, embodiments of the present invention provide a waypoint prediction device, comprising:

[0019] The first acquisition module is configured to acquire the historical waypoint sequence of the navigated object, wherein the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences;

[0020] The second acquisition module is configured to acquire scene features of the target navigation route of the navigated object as the current scene features;

[0021] The model includes a recommendation representation module that, based at least on the current scene features and the historical scene features in the historical waypoint sequence, obtains a recommendation representation of the historical waypoints corresponding to the historical scene features.

[0022] The model includes a first scoring module that obtains a score for each historical route point based on the recommended representation of the historical route points.

[0023] The first determining module is configured to determine the path point prediction result based at least on the score of each historical path point.

[0024] Fourthly, this invention provides a waypoint prediction model training device, comprising:

[0025] The third acquisition module is configured to acquire the historical waypoint sequence of the navigable object, the sample waypoints of the target historical navigation route, and the sample scene features of the sample waypoints; wherein, the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences, and the historical waypoints are historical waypoints in the historical navigation process before the target historical navigation route.

[0026] The model includes a recommendation representation module that, based at least on the sample scene features and the historical scene features in the historical waypoint sequence, obtains the recommendation representation of the historical waypoints corresponding to the historical scene features.

[0027] The model includes a first scoring module that obtains a score for each historical route point based on the recommended representation of the historical route points.

[0028] The loss function adjusts the model parameters of the modules included in the model, based at least on the score of each historical path point and the sample path points used as sample labels.

[0029] Fifthly, embodiments of this disclosure provide an electronic device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method described in any of the preceding aspects.

[0030] In a sixth aspect, embodiments of this disclosure provide a computer-readable storage medium for storing computer instructions used by any of the above-described devices, which, when executed by a processor, are used to implement the methods described in any of the above aspects.

[0031] In a seventh aspect, embodiments of this disclosure provide a computer program product comprising computer instructions which, when executed by a processor, are used to implement the methods described in any of the preceding aspects.

[0032] The technical solutions provided in this disclosure have the following beneficial effects:

[0033] In predicting waypoints, this embodiment obtains the historical scene feature sequence corresponding to each historical waypoint in the historically planned waypoint sequence of the navigable object, as well as the current scene features corresponding to the target navigation route of the navigable object. The current scene features and historical scene features are used as input to the recommendation representation module included in the pre-trained waypoint prediction model. The recommendation representation module outputs a recommended representation for each historical waypoint, which is then used as input to a first scoring module. The first scoring module outputs a score for each historical waypoint, and the waypoint prediction result is determined at least based on the scores of each historical waypoint. The technical solution provided by this embodiment utilizes the similarity between the historical scene features corresponding to historical waypoints and the current scene features to determine the waypoint prediction result of the target navigation route from historical waypoints. This achieves proactive recommendation of waypoints to the navigable object, avoiding the time-consuming process and potential safety hazards associated with manually inputting waypoint search terms in existing technologies.

[0034] In this embodiment, when training the waypoint prediction model, the historical waypoint sequence of the navigable object, sample waypoints of the target's historical navigation route, and sample scene features of the sample waypoints are obtained. The historical scene feature sequence and sample scene features are used as input to the recommendation representation module of the waypoint prediction model to be trained. The recommendation table module outputs the recommendation representation of each historical waypoint. This recommendation representation is then used as input to the first scoring module, which outputs a score for each historical waypoint. A loss function is constructed based on the scores and sample waypoints, and the model parameters of each module included in the waypoint prediction model are adjusted based on the constructed loss function. The technical solution provided in this embodiment utilizes the historical scene features and sample scene features corresponding to historical waypoints to train the model. This allows the model to learn the connection between the similarity between historical scene features and sample scene features and the navigable object's intention to select waypoints for route planning, thereby ensuring the accuracy of the trained model's waypoint prediction.

[0035] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0036] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments, taken in conjunction with the accompanying drawings. In the drawings:

[0037] Figure 1 shows a flowchart of a waypoint prediction method provided in one embodiment of this disclosure.

[0038] Figure 2 shows a schematic diagram of the model structure of a waypoint prediction model provided in an embodiment of this disclosure.

[0039] Figure 3 shows a schematic diagram of the model structure of a waypoint prediction model provided in another embodiment of this disclosure.

[0040] Figure 4 shows a flowchart of a waypoint prediction model training method provided in one embodiment of this disclosure.

[0041] Figure 5 shows a structural block diagram of a waypoint prediction device provided in one embodiment of the present disclosure.

[0042] Figure 6 shows a structural block diagram of a waypoint prediction model training device provided in one embodiment of the present disclosure.

[0043] Figure 7 is a schematic diagram of the structure of a computer system suitable for implementing the waypoint prediction method and / or waypoint prediction model training method provided in an embodiment of the present disclosure. Detailed Implementation

[0044] In the following, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings to enable those skilled in the art to readily implement them. Furthermore, for clarity, portions unrelated to the description of the exemplary embodiments have been omitted from the drawings.

[0045] In this disclosure, it should be understood that terms such as “comprising” or “having” are intended to indicate the presence of features, figures, steps, behaviors, components, parts or combinations thereof disclosed in this specification, and do not preclude the possibility of the presence or addition of one or more other features, figures, steps, behaviors, components, parts or combinations thereof.

[0046] It should also be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0047] The user information (including but not limited to user device information such as location information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals shall be provided for users to choose to authorize or refuse.

[0048] The details of the embodiments of this disclosure are described in detail below through specific examples.

[0049] Figure 1 shows a flowchart of a waypoint prediction method provided in one embodiment of this disclosure. As shown in Figure 1, the method is used to predict waypoints for a target navigation route of a navigable object using a waypoint prediction model, and includes the following steps:

[0050] In step S101, the historical waypoint sequence of the navigated object is obtained, wherein the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences;

[0051] Each data record includes historical waypoints, which are points of interest (such as a supermarket, a gas station, or a restaurant) that participated in the historical navigation route planning of the navigable object. The historical scene feature sequence corresponding to a historical waypoint includes the historical scene features corresponding to each historical navigation route planning in which the historical waypoint participated.

[0052] For a historical navigation route planning process involving historical waypoints, a set of historical scene features can be obtained, including but not limited to the following features:

[0053] The characteristics of the origin and destination of this historical navigation route planning (such as origin and destination coordinates, type of point of interest at the destination, etc.), request time characteristics (that is, the time when the historical navigation route planning request was initiated), weather characteristics, and request date characteristics (that is, whether the day the historical navigation route planning request was initiated was a weekday or a holiday).

[0054] It is understandable that a historical waypoint may have only participated in one historical navigation route planning for the navigated object, or it may have participated in two or more historical navigation route plannings. Therefore, the historical scene feature sequence corresponding to a historical waypoint may include only one set of historical scene features, or it may include two or more sets of historical scene features arranged in order of request time features.

[0055] In step S102, the scene features of the target navigation route of the navigated object are obtained as the current scene features;

[0056] The current scene features can be obtained before the navigated object initiates the target navigation route planning request, that is, when the navigated object inputs the start and end points for requesting the target navigation route, the current scene features are obtained. Alternatively, the current scene features can be obtained after the navigated object initiates the target navigation route planning request (including before or after selecting navigation for the target navigation route). Regardless of at which stage of navigation the current scene features are obtained, the technical solution provided in this disclosure can be adopted.

[0057] In practical implementation, the current scene features may include origin and destination features used to obtain target navigation route planning (such as origin and destination coordinates, point of interest type of the destination, etc.), request time features (i.e., the time when the target navigation route planning request was initiated), weather features, and request date features (i.e., whether the day the target navigation route planning request was initiated was a weekday or a holiday). It is understood that the number of current scene features needs to be the same as the number of features used when training the waypoint prediction model.

[0058] In step S103, the current scene features and the historical waypoint sequence are input into the model. The model includes a recommendation representation module, which obtains a recommendation representation of the historical waypoints corresponding to the historical scene features based at least on the current scene features and the historical scene features in the historical waypoint sequence.

[0059] In practical implementation, the recommendation representation module can be implemented by a deep interest network model. The deep interest network model is mainly used to capture the user's interest preferences. In the pre-training stage, the recommendation representation module of this disclosure learns the differences in interest of the navigated object to different types of waypoints in different scenarios. Therefore, in the prediction stage, the degree of interest of the navigated object to the historical waypoints can be obtained by calculating the similarity between the current scene features and the features of each historical scene in the historical waypoint sequence. That is, the recommendation representation of the historical waypoints obtained in step S103 represents the degree of interest of the navigated object in the historical waypoints. The recommendation representation can be expressed in the form of a vector.

[0060] In step S104, the first scoring module of the model obtains a score for each historical waypoint based on the recommended representation of the historical waypoints;

[0061] The score of historical waypoints can be understood as the probability that the navigated object will choose that historical waypoint. The higher the score of the historical waypoints output by the first scoring module, the more likely the navigated object is to choose that historical waypoint. The lower the score, the less likely the navigated object is to choose that historical waypoint.

[0062] In step S105, the waypoint prediction result is determined based at least on the score of each historical waypoint.

[0063] In practical implementation, the waypoint prediction result is determined based at least on the score of each historical waypoint, which may specifically include:

[0064] The historical waypoints whose scores are higher than or equal to a set threshold output by the first scoring module are determined as the waypoint prediction results. Alternatively, a set number (e.g., 3 or 4) of historical waypoints are selected as the waypoint prediction results according to their scores from high to low.

[0065] The above is an embodiment of the waypoint prediction method provided in this disclosure. The solution provided in this embodiment determines the travel intention of the navigated object through a trained waypoint prediction model, and then predicts historical waypoints that match the travel intention from the historical waypoints of the navigated object as waypoint prediction results. Finally, the waypoint prediction results can be actively pushed to the navigated object, realizing the proactive recommendation of waypoints to the navigated object before or during navigation.

[0066] Using the method provided in the above embodiments, in scenarios requiring active prediction of waypoints, the predicted waypoints to be recommended to the navigated object can usually be obtained. However, in practice, in some scenarios, it may not be suitable to recommend predicted waypoints to the navigated object, or the predicted waypoints may not be waypoints that the navigated object is very interested in. Therefore, in order to improve the accuracy of prediction, this disclosure provides another embodiment. The difference between this embodiment and the previous embodiment is that the waypoint prediction model may further include: a second scoring module. This embodiment further includes, based on the method of the previous embodiment:

[0067] The second scoring module obtains a recommendation score based on the recommendation representation of the historical waypoints. The recommendation score is used to determine whether to recommend any historical waypoints to the navigated object.

[0068] When the waypoint prediction model further includes a second scoring module, step S105, namely, determining the waypoint prediction result based at least on the score of each historical waypoint, is implemented as follows:

[0069] Based on the recommended scores and the scores of each historical waypoint, the waypoint prediction results are determined.

[0070] In specific implementation, determining the waypoint prediction result based on the recommended score and the score of each historical waypoint may include:

[0071] The system compares the scores of each historical waypoint output by the first scoring module with the recommended scores output by the second scoring module. If the score of a historical waypoint output by the first scoring module is higher than the recommended score, then the historical waypoint corresponding to that score is used as the prediction result. If the recommended score output by the second scoring module is higher than the scores of all historical waypoints output by the first scoring module, then the waypoint prediction result can be empty. An empty waypoint prediction result means that no waypoints are recommended to the navigable object. For example, if the score of a historical waypoint pi is higher than the recommended score, then the prediction result is that historical waypoint pi, that is, historical waypoint pi is recommended. If the recommended score is higher than the scores of all historical waypoints, then the prediction result is empty, that is, no historical waypoints are recommended.

[0072] The above is another embodiment provided by this disclosure. This embodiment can further filter the prediction results of waypoints by using the recommended score output by the second scoring module and the score of the historical waypoints output by the first scoring module. In particular, when the scores of the historical waypoints output by the first scoring module are not high, it can avoid recommending any waypoints to the navigated object, thus avoiding the negative experience brought to the navigated object by inaccurate prediction.

[0073] In practical implementation, the first scoring module and the second scoring module can be implemented by a multilayer perceptron, respectively. That is, the recommended representation of historical waypoints needs to be input into the first scoring module and the second scoring module respectively, and then the first scoring module outputs the score of each historical waypoint and the second scoring module outputs the recommended score.

[0074] It should be noted that the second scoring module needs to output a recommendation score by integrating the recommendation representations of all historical path points. To reduce the implementation complexity of the second scoring module, the recommendation representations of all historical path points can be fused before being input into the second scoring module. The fusion method can be to perform pooling operations on the corresponding feature values ​​in the recommendation representations of all historical path points.

[0075] Figure 2 shows a schematic diagram of the path point prediction model provided in an embodiment of this disclosure. As shown in Figure 2, this is an implementation structure of the path point prediction model provided in this disclosure. The deep interest network corresponds to the recommendation representation module. MLP1 is the first scoring module, implemented by a multilayer perceptron. MLP2 is the second scoring module, also implemented by a multilayer perceptron. Sum pooling is a feature pooling operation, which can be understood as adding the corresponding feature values ​​in the recommendation representations of all historical path points (i.e., the POI (Point of Interest) refinement representation shown in Figure 2). The scores corresponding to each historical path point output by the first scoring module (i.e., the POI refinement score shown in Figure 2) and the recommendation scores output by the second scoring module (i.e., the non-recommendation score shown in Figure 2) are concatenated (i.e., concat shown in Figure 2). After passing through a Softmax layer, the probability of refinement or non-recommendation is obtained, thus yielding the path point prediction result.

[0076] The following section explains the process of obtaining recommendation representations by examining the implementation structure of the waypoint prediction model.

[0077] In some embodiments, step S103, namely, obtaining the recommended representation of the historical route points corresponding to the historical scene features based at least on the current scene features and the historical scene features in the historical route point sequence, can be implemented as follows:

[0078] Based on the current scene features and the historical scene features, determine the similarity between the current scene features and the historical scene features;

[0079] Based on the similarity and the historical scene features, a recommended representation of the historical waypoints corresponding to the historical scene features is determined.

[0080] To facilitate understanding of the solution provided in this disclosure, the process of obtaining the recommended representation is illustrated below with specific examples. For instance, a historical waypoint sequence can be represented as follows:

[0081] in, Let p represent the set of historical waypoints in the historical waypoint sequence. n , Indicates the historical path point p n The historical navigation route planning records correspond to the historical scene feature sequences. This refers to the historical path point p. n The 1st, 2nd, ..., rth participants respectively n The historical navigation route planning records the corresponding historical scene characteristics, r nIndicates the historical path point p n The total number of times the historical navigation route planning was participated in.

[0082] The similarity between current scene features and historical scene features can be calculated using the following formula (i.e., the scene similarity matrix shown in Figure 2):

[0083] Among them, a n,r Represents the current scene feature C and the historical path point p. n The similarity of historical scene features in the corresponding historical scene feature sequence (i.e., the historical planning sequence of user waypoints (POIs) shown in Figure 2). These represent the current scene feature C and the historical scene feature sequence, respectively, and the scene representation obtained through the recommendation representation module.

[0084] After obtaining the similarity score, the similarity score a is further... n,r As a weighting coefficient for historical path points p n The corresponding historical navigation route planning record contains each historical scene feature in the corresponding historical scene feature sequence. After weighting and summing, we obtain the historical path point p. n Recommended representation u n The solution formula (i.e., the weighted summation of the historical sequences shown in Figure 2) is as follows:

[0085] The above are embodiments of the waypoint prediction method provided in this disclosure. These embodiments primarily utilize scene similarity to predict waypoints. To further improve the prediction accuracy of the waypoint prediction model, this disclosure also provides an embodiment that utilizes a self-attention mechanism to extract periodic representations of historical waypoints from historical arrival point sequences. This embodiment further includes, based on the method provided in the above embodiments, the following:

[0086] Obtain the historical arrival point sequence of the navigated object; wherein, the historical arrival point sequence includes: a set of historical arrival points and feature data corresponding to each historical arrival point in the set of historical arrival points, the historical arrival points in the set of historical arrival points are sorted according to the arrival time of the historical arrival points, and the feature data corresponding to each historical arrival point includes: historical scene features, the category of the historical arrival point, and the cumulative navigation distance between two adjacent historical arrival points; the historical arrival points include historical waypoints that have actually been reached in the past or historical navigation destinations that have actually been reached in the past.

[0087] The difference between the model in this embodiment and the previous embodiment is that the model further includes: a self-attention mechanism module and a periodic weight module, wherein the self-attention mechanism module obtains a periodic representation of each historical arrival point in the set of historical arrival points based on the historical arrival point sequence.

[0088] The periodic weighting module obtains the periodic weights of the historical transit points based on the periodic representation of the historical transit points belonging to the historical transit points in the historical transit point set.

[0089] The periodic weights of the historical waypoints are used as input to the recommendation representation module;

[0090] In this embodiment, the aforementioned step S103, namely, the step of obtaining the recommended representation of the historical path points corresponding to the historical scene features based at least on the current scene features and the historical scene features in the historical path point sequence, is implemented in the following manner:

[0091] Based on the periodic weights of the historical waypoints, the current scene features, and the historical scene features in the historical waypoint sequence, a recommended representation of the historical waypoints corresponding to the historical scene features is obtained.

[0092] Specifically, based on the periodic weights of the historical waypoints, the current scene features, and the historical scene features in the historical waypoint sequence, a recommended representation of the historical waypoints corresponding to the historical scene features is obtained, including:

[0093] Based on the current scene features and the historical scene features, the similarity between the current scene features and the historical scene features is obtained;

[0094] Based on the similarity, the periodic weight, and the historical scene features, a recommended representation of the historical waypoints is calculated.

[0095] In the above embodiments, the historical arrival point can be a historical transit point or historical destination that has actually been reached in the historical navigation route. The historical scene characteristics of the historical arrival point are similar to those of the historical transit point and may include, but are not limited to: the start and end point characteristics of the historical navigation route corresponding to the historical arrival point (such as the start and end point coordinates, the point of interest type of the end point, etc.), the request time characteristics (that is, the time when the historical navigation route planning request corresponding to the historical arrival point was initiated), the weather characteristics, and the request date characteristics (that is, whether the day when the historical navigation route planning request corresponding to the historical arrival point was initiated was a working day or a holiday).

[0096] The category of historical arrival points refers to the category of points of interest corresponding to historical transit points or historical destinations, such as gas stations, charging stations, and public toilets.

[0097] In this embodiment of the disclosure, the historical arrival point sequence is sorted according to the arrival time of the historical arrival points. The arrival time can be obtained from the historical navigation actual travel data of the historical navigation route (that is, the data actually formed when the navigated object is guided to travel according to the historical navigation route). The cumulative navigation distance between two adjacent historical arrival points is the cumulative distance actually navigated by the navigated object between the arrival time of the previous historical arrival point in the historical navigation route planning record of the previous historical arrival point and the arrival time of the next historical arrival point in the historical navigation route planning record of the next historical arrival point. For example, in the historical navigation route planning record of the previous historical arrival point, the arrival time of the previous historical arrival point is T1, and in the historical navigation route planning record of the next historical arrival point, the arrival time of the next historical arrival point is T2. According to the historical navigation actual data record, the navigation object was actually navigated 3 times between T1 and T2. The actual distance traveled in the first navigation was L1, the actual distance traveled in the second navigation was L2, and the actual distance traveled in the third navigation was L3. If no historical arrival point was found in these 3 navigations, then the cumulative navigation distance between the two adjacent historical arrival points is L1+L2+L3.

[0098] In this embodiment, a self-attention mechanism is used to extract the implicit periodic representations from the historical arrival point sequence. These periodic representations can be used to represent the periodic behavior of the navigated object, such as how often it visits a gas station or a supermarket. The self-attention mechanism is a sequence modeling method that automatically learns the relationships between historical arrival points by calculating the association weights between different historical arrival points in the historical arrival point sequence.

[0099] In some embodiments, as described above, after inputting the current scene features and historical scene features into the recommendation representation module, the recommendation representation module can calculate the similarity between the historical scene features and the current scene features. This similarity can be used as a weighted weight of the historical scene features when calculating the recommendation representation of historical waypoints. After introducing the periodic behavior of the navigated object, the periodic weight of the historical waypoints will be used as another weight of the historical scene features, and together with the similarity, the historical scene features will be weighted to obtain the recommendation representation of the historical waypoints.

[0100] Figure 3 shows a schematic diagram of the model structure of a waypoint prediction model provided in another embodiment of this disclosure. As shown in Figure 3, this is a waypoint prediction model that introduces periodic behavior. The historical arrival point sequence is processed by a self-attention mechanism module in the waypoint prediction model to obtain a periodic representation of the historical arrival points. If a historical waypoint appears in the historical arrival point sequence, its periodic representation is obtained from the periodic representation of the historical arrival point. This periodic representation is then concatenated with the current scene features and processed by a multilayer perceptron (Multilayer Perceptron). After implementing the periodic weight module using Perceptron (MLP), the periodic weights of historical path points can be obtained. These periodic weights are input into the recommendation representation module (i.e., the deep interest network in Figure 3), where they are combined with similarity (i.e., the scene similarity matrix in Figure 3) to weight the historical scene features (i.e., the weighted summation of the historical sequence shown in Figure 3) to obtain the recommendation representation of historical path points (i.e., the POI refined recommendation representation shown in Figure 3). After the recommendation representation of historical path points is processed by the path point prediction model, which includes the first scoring module, the second scoring module, the self-attention mechanism module, and the periodic weight module, the path point prediction result can be obtained. This part has been described above and will not be repeated here.

[0101] The historical arrival point sequence can be represented as follows:

[0102] Among them, set This represents the set of historical arrival points in the historical arrival point sequence. Indicates the historical arrival point p n Historical scene characteristics and historical arrival point p n Category, d n This indicates that the object being navigated to has reached point p in the history. n-1 and historical arrival point p n The cumulative navigation distance between them.

[0103] As mentioned earlier, historical arrival points refer to historical waypoints or historical destinations that the navigated object has actually reached. Therefore, in order to ensure that the historical arrival points of the navigated object that are discovered are indeed historical arrival points that the navigated object has actually reached, in some embodiments, the step of obtaining the historical arrival point sequence of the navigated object can be implemented in the following manner:

[0104] From the historical navigation route planning records of the navigated object, obtain the historical waypoint planning records, historical destination planning records, and corresponding navigation trajectory records of the navigated object;

[0105] Based on the historical waypoint planning records and the historical destination planning records, the candidate arrival points of the navigated object are determined;

[0106] The candidate arrival point is matched with the road segment corresponding to the navigation trajectory record. If the candidate arrival point matches the road segment corresponding to the navigation trajectory record, the candidate arrival point is used as the historical arrival point of the navigated object.

[0107] If the candidate arrival point does not match the road segment corresponding to the navigation trajectory record, but the matching distance from the candidate arrival point to the road segment corresponding to the navigation trajectory record is less than a set distance threshold, then the candidate arrival point is used as the historical arrival point of the navigated object.

[0108] Based on the historical arrival points, the historical waypoint planning records, and the historical destination planning records, a historical arrival point sequence is determined.

[0109] Among them, the historical waypoint planning record can include historical planning data of the waypoints that the navigated object has planned in the past, the historical destination planning record can include all historical planning data of the navigated object in the past, and the navigation trajectory record refers to the trajectory record generated when the navigated object actually navigates according to the historical navigation route.

[0110] Furthermore, when obtaining historical waypoint sequences, the issue of duplicate waypoint records also needs to be considered. For example, if the navigated object plans a navigation route for its target origin and destination (but has not yet started navigation), this navigation route planning record is represented by id_1; if the navigated object does not change its target origin and destination, but only adds waypoint A and replans the route, this historical navigation route planning record is represented by id_2; if the navigated object continues to not change its target origin and destination or waypoint A, but only adds waypoint B and plans the route, this historical navigation route planning record is represented by id_3. In this case, the historical navigation route planning records sorted by planning time will contain two adjacent historical navigation route planning records (id_2 and id_3) including waypoint A and one historical navigation route planning record including waypoint B (id_3). If no processing is done on this record, the constructed historical waypoint sequence will include two data records indexed by waypoint A, but these two records express the same travel intention. This situation is considered a duplicate record of waypoint A. To solve this problem, step S101, namely, obtaining the historical waypoint sequence of the navigated object, can be implemented as follows:

[0111] From the historical navigation route planning records of the navigated object, obtain the waypoints that participated in the historical navigation route planning records and the historical scene features of the historical navigation route planning;

[0112] For the same waypoint, if the waypoint appears in more than two historical navigation route planning records, it is determined whether there are more than two records with the same planning start and end point in the more than two historical navigation route planning records. If so, it is determined whether the planning time difference between the more than two records with the same planning start and end point exceeds a threshold and whether there are other planning records between the more than two records.

[0113] If the planning time difference does not exceed the threshold and no other records appear between two or more records, then for the two or more records with the same planning start and end point, the historical scene characteristics of any one of the records will be retained.

[0114] If the time difference exceeds the threshold or other records appear between two or more records, the historical scene characteristics of each record are retained.

[0115] If the waypoint appears in more than two historical navigation route planning records, but the start or end point of the historical navigation route planning records is different, then the historical scene features corresponding to each record are retained.

[0116] Using the aforementioned waypoints as indexes and all the preserved historical scene features corresponding to each waypoint as keys, a historical waypoint sequence is generated.

[0117] For planning records with the same origin and destination, performing the above steps ensures that historical navigation route planning records expressing different travel intentions of the navigated object are retained, reducing redundant data and improving data processing efficiency.

[0118] For example:

[0119] Historical navigation route planning record 1: From O1 to D1, including historical waypoint A, time is 1;

[0120] Historical navigation route planning record 2: From O1 to D1, there are no historical waypoints, and the time is 2.

[0121] Historical navigation route planning record 3: From O1 to D1, including historical stop A, time is 3;

[0122] Historical navigation route planning record 4: From O1 to D1, including historical stops A and B, the time is 8;

[0123] For historical waypoint A, there are three historical navigation route planning records: Plan 1, Plan 3, and Plan 4. Without any processing, the historical waypoint planning data for historical waypoint A is as follows:

[0124] Historical waypoint A: (Historical navigation route planning record 1, time 1), (Historical navigation route planning record 3, time 3), (Historical navigation route planning record 4, time 8)

[0125] Among them, historical waypoint planning data 1 and 3, and historical navigation route planning records 3 and 4 are two adjacent historical navigation route planning records of historical waypoint A. Assuming that the time interval mentioned in the above condition (1) is set to 3, for planning historical navigation route planning records 1 and 3, although their time interval is 2, in the full record of historical navigation route planning, we can see that there is another historical navigation route planning record 2 between historical navigation route planning records 1 and 3. Therefore, historical navigation route planning records 1 and 3 are both retained. For historical navigation route planning records 3 and 4, the time interval is 5. Therefore, historical navigation route planning records 3 and 4 are also retained.

[0126] The above describes a method for predicting waypoints using a waypoint prediction model. This method, when predicting waypoints, obtains the historical scene feature sequence corresponding to each historical waypoint in the historically planned waypoint sequence of the navigable object, as well as the current scene features corresponding to the target navigation route of the navigable object. The current scene features and historical scene features are used as input to the recommendation representation module included in the pre-trained waypoint prediction model. The recommendation representation module outputs a recommended representation for each historical waypoint, which is then used as input to a first scoring module. The first scoring module outputs a score for each historical waypoint, and the waypoint prediction result is determined at least based on the scores of each historical waypoint. The technical solution provided in this disclosure utilizes the similarity between the historical scene features corresponding to historical waypoints and the current scene features to determine the waypoint prediction result of the target navigation route from historical waypoints. This achieves proactive recommendation of waypoints to the navigable object, avoiding the time-consuming process and potential safety issues associated with manually inputting waypoint search terms required by existing technologies.

[0127] The training method of the waypoint prediction model of this disclosure will be described in detail below with reference to the accompanying drawings.

[0128] Figure 4 shows a flowchart of a waypoint prediction model training method provided in one embodiment of this disclosure. As shown in Figure 4, the training method includes the following steps:

[0129] In step S401, the historical waypoint sequence of the navigable object, the sample waypoints of the target historical navigation route, and the sample scene features of the sample waypoints are obtained; wherein, the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences, and the historical waypoints are historical waypoints in the historical navigation process before the target historical navigation route;

[0130] The specific implementation process of obtaining the historical waypoint sequence of the navigated object during the training model phase can also be applied to the prediction phase. The same content will not be repeated here to save space, and should not be regarded as the training method not including the aforementioned specific implementation process of obtaining the historical waypoint sequence of the navigated object.

[0131] In this embodiment, the target historical navigation route refers to the navigation route that the navigated object has historically planned. For the target historical navigation route, the waypoints selected by the navigated object are known, and these known waypoints are the sample waypoints. Therefore, specific details such as the historical waypoint sequence, the target historical navigation route, the sample waypoints, the sample scene features, and the historical scene features can be found in the description above, and will not be repeated here.

[0132] In step S402, the sample scene features and the historical waypoint sequence are input into the model. The model includes a recommendation representation module, which obtains the recommendation representation of the historical waypoints corresponding to the historical scene features based at least on the sample scene features and the historical scene features in the historical waypoint sequence.

[0133] The recommendation representation module can be implemented using a deep interest network model. For details on the specific implementation of the recommendation representation module and recommendation representation, please refer to the relevant sections above. It will not be repeated here.

[0134] In step S403, the recommended representation of the historical waypoints is input into the model, and the first scoring module of the model obtains a score for each historical waypoint based on the recommended representation of the historical waypoints.

[0135] The first scoring module can be implemented using a multilayer perceptron (MLP). For details regarding the first scoring module and the scoring of historical waypoints, please refer to the description of the waypoint prediction method above, which will not be repeated here.

[0136] In step S404, the loss function adjusts the model parameters of the modules included in the model based at least on the score of each historical path point and the sample path point as a sample label.

[0137] In this step, a loss function can be constructed using the scores and sample labels of each historical waypoint, and the model parameters of each module included in the waypoint prediction model can be adjusted based on the loss function, as shown in Figures 2 and 3 as CE loss.

[0138] The following example illustrates a cross-entropy loss function that can be used for single-label multi-class classification problems in this disclosure. It should be understood that the loss functions that can be used in this disclosure are not limited to the example shown below, and those skilled in the art can construct loss functions of any form based on actual circumstances.

[0139] The cross-entropy loss function for a single-label multi-class classification problem is expressed as follows:

[0140] in, The set represents the training sample set. Each training sample includes the historical waypoint sequence of the navigated object and the sample scene features corresponding to the historical navigation route of the target. s,n This represents the sample label corresponding to the training sample s. This represents the predicted value of the waypoint prediction model, i.e., whether it is the predicted value of recommending the nth historical waypoint for the training sample s.

[0141] As described in the previous embodiment, in order to avoid actively recommending any waypoints when the scores of historical waypoints output by the first scoring module are all low, the waypoint prediction model may further include a second scoring module, which outputs a recommendation score for whether to recommend historical waypoints. Therefore, the method further includes:

[0142] The second scoring module obtains a recommendation score based on the recommendation representation of the historical waypoints. The recommendation score is used to determine whether to recommend any historical waypoints to the navigated object.

[0143] In the case of including the second scoring module, step S404, which is the step of adjusting the model parameters of the modules included in the model based at least on the score of each historical path point and the sample path point as the sample label, can be implemented as follows:

[0144] The loss function adjusts the model parameters of the modules included in the model based on the recommendation score, the score of each historical path point, and the sample path points.

[0145] For technical details regarding the second scoring module, please refer to the relevant sections above, which will not be repeated here.

[0146] It should be noted that the construction of sample labels needs to be adapted in this case. Specifically, sample labels can be constructed as the label values ​​corresponding to all historical waypoints in the historical waypoint sequence, as well as the label values ​​of unrecommended waypoints.

[0147] In practical applications, step S402, namely the recommendation representation module included in the model, obtains the recommendation representation of the historical path points corresponding to the historical scene features based at least on the sample scene features and the historical scene features in the historical path point sequence, specifically including:

[0148] Based on the input of the sample scene features and the historical scene features into the recommendation representation module of the waypoint prediction model, the similarity between the sample scene features and the historical scene features is determined.

[0149] Based on the similarity and the historical scene features, a recommended representation of the historical waypoints is determined.

[0150] Among them, the recommended representation of historical waypoints is obtained based on the sample scene features and the historical scene features corresponding to the historical waypoints. For the implementation process of obtaining the recommended representation of historical waypoints based on the current scene features and the historical scene features corresponding to the historical waypoints, please refer to the above description. The relevant details can also be found in the above description, and will not be repeated here.

[0151] Furthermore, to further improve the prediction accuracy of the waypoint prediction model, embodiments of this disclosure also propose using a self-attention mechanism to extract periodic representations of historical waypoints from historical arrival sequence. Therefore, in some embodiments, the above-mentioned training method of this disclosure further includes:

[0152] Obtain the historical arrival point sequence of the navigated object; wherein, the historical arrival point sequence includes a set of historical arrival points and feature data corresponding to each historical arrival point in the set of historical arrival points, the historical arrival points in the set of historical arrival points are sorted according to the arrival time of the historical arrival points, and the feature data corresponding to each historical arrival point includes: historical scene features, the category of the historical arrival point, and the cumulative navigation distance between two adjacent historical arrival points; the historical arrival points include historical waypoints that have actually been reached in the past or historical navigation destinations that have actually been reached in the past.

[0153] The specific implementation process of obtaining the historical arrival point sequence of the navigated object during the training model stage can also be applied to the prediction stage. The same content will not be repeated here to save space, and should not be regarded as the training method not including the content related to the specific implementation process of obtaining the historical arrival point sequence of the navigated object provided in the foregoing embodiments.

[0154] The model includes a self-attention mechanism module that obtains a periodic representation of the historical arrival points based on the historical arrival point sequence.

[0155] The model further includes a self-attention mechanism module and a periodic weight module, wherein the self-attention mechanism module obtains a periodic representation of each historical arrival point in the historical arrival point set based on the historical arrival point sequence.

[0156] The periodic weighting module obtains the periodic weights of the historical transit points based on the periodic representation of the historical transit points belonging to the historical transit points in the historical transit point set.

[0157] The periodic weights of the historical waypoints are used as input to the recommendation representation module;

[0158] When the model incorporates periodic representations, the recommendation representation module included in the model, based at least on the sample scene features and the historical scene features in the historical waypoint sequence, obtains the recommended representations of the historical waypoints corresponding to the historical scene features, which can be implemented as follows:

[0159] The recommendation representation module obtains the recommendation representation of the historical path points corresponding to the historical scene features based on the periodic weights, the sample scene features, and the historical scene features in the historical path point sequence.

[0160] In its specific implementation, the recommendation representation module obtains the recommendation representation of the historical path points corresponding to the historical scene features based on the periodic weights, the sample scene features, and the historical scene features in the historical path point sequence, including:

[0161] The recommendation representation module obtains the similarity between the current scene features and the historical scene features based on the current scene features and the historical scene features;

[0162] The recommendation representation module in the model also calculates the recommended representation of the historical waypoints based on the similarity, the periodic weight, and the historical scene features.

[0163] Similarly, the process of obtaining the periodic weights of historical waypoints by utilizing the historical arrival point sequence, and then combining the periodic weights of historical waypoints with the current scene features and historical scene features to obtain the recommended representation of historical waypoints, can be found in the relevant description of waypoint prediction methods above, and will not be repeated here.

[0164] In this embodiment, during model training, by inputting the category of historical arrival points, the historical scene features of historical arrival points, and the cumulative navigation distance between two adjacent historical arrival points into the self-attention mechanism module, the model can learn periodic representations such as the relationships between the categories of historical arrival points, the periodic features of the historical scene features of different historical arrival points, the periodic features between the arrival times of different historical arrival points, and the influence of navigation distance between different historical arrival points. After these periodic representations are processed by the periodic weight module of the waypoint prediction model, the periodic weights of historical waypoints can be obtained. The periodic weights, together with the current scene features and historical scene features, are used to calculate the recommendation representation of historical waypoints. This allows the recommendation representation using historical waypoints to include the influencing factors of the periodic behavior of the navigable object, thereby improving the model's prediction accuracy and personalization level.

[0165] In training the waypoint prediction model, this embodiment acquires the historical waypoint sequence of the navigable object, sample waypoints of the target's historical navigation routes, and sample scene features of the sample waypoints. The historical scene feature sequence and sample scene features are used as input to the recommendation representation module of the waypoint prediction model to be trained. The recommendation table module outputs the recommendation representation of each historical waypoint, which is then used as input to the first scoring module. The first scoring module outputs a score for each historical waypoint. A loss function is constructed based on the scores and sample waypoints, and the model parameters of each module included in the waypoint prediction model are adjusted based on the constructed loss function. The technical solution provided by this embodiment utilizes the historical scene features and sample scene features corresponding to historical waypoints to train the model. This allows the model to learn the connection between the similarity between historical scene features and sample scene features and the navigable object's intention to select waypoints for route planning, thereby ensuring the accuracy of the trained model's waypoint prediction.

[0166] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein.

[0167] Figure 5 shows a structural block diagram of a waypoint prediction device according to an embodiment of this disclosure. This device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. As shown in Figure 5, the waypoint prediction device includes:

[0168] The first acquisition module 501 is configured to acquire the historical waypoint sequence of the navigated object, wherein the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences;

[0169] The second acquisition module 502 is configured to acquire the scene features of the target navigation route of the navigated object as the current scene features;

[0170] The model includes a recommendation representation module 503, which obtains a recommendation representation of the historical path points corresponding to the historical scene features based at least on the current scene features and the historical scene features in the historical path point sequence.

[0171] The model includes a first scoring module 504, which obtains a score for each historical route point based on the recommended representation of the historical route points.

[0172] The first determining module 505 is configured to determine the path point prediction result based at least on the score of each historical path point.

[0173] The device further includes: a second scoring module included in the waypoint prediction model, which obtains a recommendation score based on the recommendation representation of the historical waypoints, and the recommendation score is used to determine whether to recommend any historical waypoints to the navigated object;

[0174] The first determining module can be implemented in the following manner:

[0175] Based on the recommended scores and the scores of each historical waypoint, the waypoint prediction results are determined.

[0176] The recommended representation module can be implemented in the following manner:

[0177] Based on the current scene features and the historical scene features, determine the similarity between the current scene features and the historical scene features;

[0178] Based on the similarity and the historical scene features, a recommended representation of the historical waypoints corresponding to the historical scene features is determined.

[0179] The device further includes:

[0180] The first arrival point acquisition module is configured to acquire the historical arrival point sequence of the navigated object; wherein, the historical arrival point sequence includes: a set of historical arrival points and feature data corresponding to each historical arrival point in the set of historical arrival points, the historical arrival points in the set of historical arrival points are sorted according to their arrival time, and the feature data corresponding to each historical arrival point includes: historical scene features, the category of the historical arrival point, and the cumulative navigation distance between two adjacent historical arrival points; the historical arrival points include historical waypoints that have actually been reached or historical navigation endpoints that have actually been reached.

[0181] The model includes a self-attention mechanism module that, based on the historical arrival point sequence, obtains a periodic representation of each historical arrival point in the historical arrival point set.

[0182] The model includes a periodic weight module, which obtains the periodic weights of the historical transit points based on the periodic representation of the historical transit points belonging to the historical transit points in the historical transit point set.

[0183] The first input module is configured to use the periodic weights of the historical waypoints as input to the recommendation representation module;

[0184] The recommended representation module can be implemented in the following manner:

[0185] Based on the periodic weights of the historical waypoints, the current scene features, and the historical scene features in the historical waypoint sequence, a recommended representation of the historical waypoints corresponding to the historical scene features is obtained.

[0186] The recommended representation of historical path points corresponding to the historical scene features is obtained based on the periodic weights of the historical path points, the current scene features, and the historical scene features in the historical path point sequence. This can be implemented in the following manner:

[0187] Based on the current scene features and the historical scene features, the similarity between the current scene features and the historical scene features is obtained;

[0188] Based on the similarity, the periodic weight, and the historical scene features, a recommended representation of the historical waypoints is calculated.

[0189] The first acquisition module can be implemented in the following manner:

[0190] From the historical navigation route planning records of the navigated object, obtain the waypoints that participated in the historical navigation route planning records and the historical scene features of the historical navigation route planning;

[0191] For the same waypoint, if the waypoint appears in more than two historical navigation route planning records, it is determined whether there are more than two records with the same planning start and end point in the more than two historical navigation route planning records. If so, it is determined whether the planning time difference between the more than two records with the same planning start and end point exceeds a threshold and whether there are other planning records between the more than two records.

[0192] If the planning time difference does not exceed the threshold and no other records appear between two or more records, then for the two or more records with the same planning start and end point, the historical scene characteristics of any one of the records will be retained.

[0193] If the time difference exceeds the threshold or other records appear between two or more records, the historical scene characteristics of each record are retained.

[0194] If the waypoint appears in more than two historical navigation route planning records, but the start or end point of the historical navigation route planning records is different, then the historical scene features corresponding to each record are retained.

[0195] Using the aforementioned waypoints as indexes and all the preserved historical scene features corresponding to each waypoint as keys, a historical waypoint sequence is generated.

[0196] The first arrival point acquisition module can be implemented as follows:

[0197] From the historical navigation route planning records of the navigated object, obtain the historical waypoint planning records, historical destination planning records, and corresponding navigation trajectory records of the navigated object;

[0198] Based on the historical waypoint planning records and the historical destination planning records, the candidate arrival points of the navigated object are determined;

[0199] The candidate arrival point is matched with the road segment corresponding to the navigation trajectory record. If the candidate arrival point matches the road segment corresponding to the navigation trajectory record, the candidate arrival point is used as the historical arrival point of the navigated object.

[0200] If the candidate arrival point does not match the road segment corresponding to the navigation trajectory record, but the matching distance from the candidate arrival point to the road segment corresponding to the navigation trajectory record is less than a set distance threshold, then the candidate arrival point is used as the historical arrival point of the navigated object.

[0201] Based on the historical arrival points, the historical waypoint planning records, and the historical destination planning records, a historical arrival point sequence is determined.

[0202] The specific implementation methods of the execution steps of each module of the above-mentioned waypoint prediction device correspond to the specific implementation methods of the corresponding steps of the waypoint prediction method provided above. For details, please refer to the description of the waypoint prediction method above, which will not be repeated here.

[0203] Figure 6 shows a structural block diagram of a waypoint prediction model training device provided in one embodiment of this disclosure. This device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. As shown in Figure 6, the waypoint prediction model training device includes:

[0204] The third acquisition module 601 is configured to acquire the historical waypoint sequence of the navigable object, the sample waypoints of the target historical navigation route, and the sample scene features of the sample waypoints; wherein, the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences, and the historical waypoints are historical waypoints in the historical navigation process before the target historical navigation route.

[0205] The model includes a recommendation representation module 602, which obtains a recommendation representation of the historical path points corresponding to the historical scene features based at least on the sample scene features and the historical scene features in the historical path point sequence.

[0206] The model includes a first scoring module 603, which obtains a score for each historical route point based on the recommended representation of the historical route points.

[0207] The loss function 604 adjusts the model parameters of the modules included in the model based at least on the score of each historical path point and the sample path points used as sample labels.

[0208] The device further includes:

[0209] The waypoint prediction model includes a second scoring module, which obtains a recommendation score based on the recommendation representation of the historical waypoints. The recommendation score is used to determine whether to recommend any historical waypoints to the navigated object.

[0210] The loss function can be implemented as follows:

[0211] Based on the recommended scores, the scores of each historical path point, and the sample path points, the model parameters of the modules included in the model are adjusted.

[0212] The device further includes:

[0213] The second arrival point acquisition module is configured to acquire the historical arrival point sequence of the navigated object; wherein, the historical arrival point sequence includes a set of historical arrival points and feature data corresponding to each historical arrival point in the set of historical arrival points, the historical arrival points in the set of historical arrival points are sorted according to their arrival time, and the feature data corresponding to each historical arrival point includes: historical scene features, the category of the historical arrival point, and the cumulative navigation distance between two adjacent historical arrival points; the historical arrival points include historical waypoints that have actually been reached or historical navigation destinations that have actually been reached.

[0214] The model includes a self-attention mechanism module that obtains a periodic representation of the historical arrival points based on the historical arrival point sequence.

[0215] The model includes a self-attention mechanism module that, based on the historical arrival point sequence, obtains a periodic representation of each historical arrival point in the historical arrival point set.

[0216] The model includes a periodic weight module, which obtains the periodic weights of the historical transit points based on the periodic representation of the historical transit points belonging to the historical transit points in the historical transit point set.

[0217] The second input module is configured to use the periodic weights of the historical waypoints as input to the recommendation representation module;

[0218] The recommended representation module can be implemented in the following manner:

[0219] Based on the periodic weights, the sample scene features, and the historical scene features in the historical waypoint sequence, a recommended representation of the historical waypoints corresponding to the historical scene features is obtained.

[0220] The step of obtaining the recommended representation of historical route points corresponding to the historical scene features based on the periodic weights, the sample scene features, and the historical scene features in the historical route point sequence can be implemented in the following manner:

[0221] Based on the current scene features and the historical scene features, the similarity between the current scene features and the historical scene features is obtained;

[0222] Based on the similarity, the periodic weight, and the historical scene features, a recommended representation of the historical waypoints is calculated.

[0223] The third acquisition module can be implemented in the following manner:

[0224] From the historical navigation route planning records of the navigated object, obtain the waypoints that participated in the historical navigation route planning records and the historical scene features of the historical navigation route planning;

[0225] For the same waypoint, if the waypoint appears in more than two historical navigation route planning records, it is determined whether there are more than two records with the same planned start and end points. If so, it is determined whether the planning time difference between the more than two records with the same planned start and end points exceeds a threshold and whether there are other records between the more than two records. If the planning time difference does not exceed the threshold and there are no other records between the more than two records, the historical scene features of one of the more than two records with the same planned start and end points are retained; or if the planning time difference exceeds the threshold or other records appear between the more than two records, the historical scene features of each record are retained; if the waypoint appears in records with different start and end points, the historical scene features corresponding to each record are retained.

[0226] If the waypoint appears in more than two historical navigation route planning records, but the start or end point of the historical navigation route planning records is different, then the historical scene features corresponding to each record are retained.

[0227] A historical waypoint sequence is generated using the waypoints as indexes and all the preserved historical scene features corresponding to the waypoints as keys.

[0228] The second arrival point acquisition module can be implemented as follows:

[0229] From the historical navigation route planning records of the navigated object, obtain the historical waypoint planning records, historical destination planning records, and corresponding navigation actual travel records of the navigated object;

[0230] Based on the historical waypoint planning records and the historical destination planning records, the candidate arrival points of the navigated object are determined;

[0231] The candidate arrival point is matched with the road segment corresponding to the navigation trajectory record. If the candidate arrival point matches the road segment corresponding to the navigation trajectory record, the candidate arrival point is used as the historical arrival point of the navigated object.

[0232] If the candidate arrival point does not match the road segment corresponding to the navigation trajectory record, but the matching distance from the candidate arrival point to the road segment corresponding to the navigation trajectory record is less than a set distance threshold, then the candidate arrival point is used as the historical arrival point of the navigated object.

[0233] Based on the historical arrival points, the historical waypoint planning records, and the historical destination planning records, a historical arrival point sequence is determined.

[0234] The historical waypoints refer to points of interest that participate in the historical navigation route planning of the navigated object as waypoints, and the historical scene feature sequence includes the historical scene features corresponding to each historical navigation route planning in which the points of interest participate as waypoints.

[0235] The specific implementation methods of the execution steps of each module of the above-mentioned waypoint prediction model training device correspond to the specific implementation methods of the corresponding steps of the training method and prediction method provided above. For details, please refer to the above description, which will not be repeated here.

[0236] The functions implemented by the above-described device can be achieved through hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the aforementioned functions.

[0237] In one possible design, the aforementioned device includes a memory and a processor. The memory stores one or more computer instructions that enable the device to execute the corresponding functions, and the processor is configured to execute the computer instructions stored in the memory. The device may also include a communication interface for communicating with other devices or communication networks.

[0238] Figure 7 is a schematic diagram of the structure of a computer system suitable for implementing the waypoint prediction method and / or waypoint prediction model training method provided in an embodiment of the present disclosure.

[0239] As shown in Figure 7, the computer system 700 includes a processing unit 701, which can be implemented as a CPU (Central Processing Unit), GPU (Graphics Processing Unit), FPGA (Field-Programmable Gate Array), NPU (Neural Processing Unit), or other processing units. The processing unit 701 can execute various processes according to any of the above-described embodiments of the method disclosed herein, based on a program stored in read-only memory (ROM) 702 or a program loaded from storage portion 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the computer system 700. The processing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0240] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0241] In particular, according to embodiments of this disclosure, any of the methods described above in the embodiments of this disclosure can be implemented as a computer software program. For example, embodiments of this disclosure include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program containing program code for performing any of the methods in the embodiments of this disclosure. In such an embodiment, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711.

[0242] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0243] The units or modules described in the embodiments of this disclosure can be implemented in software or hardware. The described units or modules can also be located in a processor, and the names of these units or modules do not necessarily constitute a limitation on the unit or module itself.

[0244] In another aspect, this disclosure also provides a computer-readable storage medium, which may be a computer-readable storage medium included in the apparatus described in the above embodiments; or it may be a standalone computer-readable storage medium not assembled into a device. The computer-readable storage medium stores one or more programs that are used by one or more processors to perform the methods described in this disclosure.

[0245] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

Claims

1. A waypoint prediction method, wherein, The method for predicting waypoints for a target navigation route of a navigable object using a waypoint prediction model includes: Obtain the historical waypoint sequence of the navigated object, wherein the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences; Obtain the scene features of the target navigation route of the navigated object as the current scene features; The current scene features and the historical waypoint sequence are input into the model. The model includes a recommendation representation module that, based at least on the current scene features and the historical scene features in the historical waypoint sequence, obtains a recommendation representation of the historical waypoints corresponding to the historical scene features. The model includes a first scoring module that obtains a score for each historical route point based on the recommended representation of the historical route points. The waypoint prediction result is determined based at least on the score of each historical waypoint.

2. The method according to claim 1, wherein, The waypoint prediction model further includes: a second scoring module, and the method further includes: The second scoring module obtains a recommendation score based on the recommendation representation of the historical waypoints. The recommendation score is used to determine whether to recommend any historical waypoints to the navigated object. The determination of the waypoint prediction result based at least on the scores of each historical waypoint specifically includes: Based on the recommended scores and the scores of each historical waypoint, the waypoint prediction results are determined.

3. The method according to claim 1 or 2, wherein, Based at least on the current scene features and the historical scene features in the historical waypoint sequence, a recommended representation of the historical waypoints corresponding to the historical scene features is obtained, specifically including: Based on the current scene features and the historical scene features, determine the similarity between the current scene features and the historical scene features; Based on the similarity and the historical scene features, a recommended representation of the historical waypoints corresponding to the historical scene features is determined.

4. The method according to any one of claims 1-3, wherein, The method further includes: Obtain the historical arrival point sequence of the navigated object; wherein, the historical arrival point sequence includes: a set of historical arrival points and feature data corresponding to each historical arrival point in the set of historical arrival points, the historical arrival points in the set of historical arrival points are sorted according to the arrival time of the historical arrival points, and the feature data corresponding to each historical arrival point includes: historical scene features, the category of the historical arrival point, and the cumulative navigation distance between two adjacent historical arrival points; the historical arrival points include historical waypoints that have actually been reached in the past or historical navigation destinations that have actually been reached in the past. The model further includes a self-attention mechanism module and a periodic weight module, wherein the self-attention mechanism module obtains a periodic representation of each historical arrival point in the historical arrival point set based on the historical arrival point sequence. The periodic weighting module obtains the periodic weights of the historical transit points based on the periodic representation of the historical transit points belonging to the historical transit points in the historical transit point set. The periodic weights of the historical waypoints are used as input to the recommendation representation module; The step of obtaining the recommended representation of historical routes corresponding to the historical scene features based at least on the current scene features and historical scene features in the historical waypoint sequence specifically includes: Based on the periodic weights of the historical waypoints, the current scene features, and the historical scene features in the historical waypoint sequence, a recommended representation of the historical waypoints corresponding to the historical scene features is obtained.

5. The method according to claim 4, wherein, Based on the periodic weights of the historical waypoints, the current scene features, and the historical scene features in the historical waypoint sequence, a recommended representation of the historical waypoints corresponding to the historical scene features is obtained, including: Based on the current scene features and the historical scene features, the similarity between the current scene features and the historical scene features is obtained; Based on the similarity, the periodic weight, and the historical scene features, a recommended representation of the historical waypoints is calculated.

6. The method according to any one of claims 1-5, wherein, Obtaining the historical waypoint sequence of the navigated object includes: From the historical navigation route planning records of the navigated object, obtain the waypoints that participated in the historical navigation route planning records and the historical scene features of the historical navigation route planning; For the same waypoint, if the waypoint appears in more than two historical navigation route planning records, it is determined whether there are more than two records with the same planning start and end point in the more than two historical navigation route planning records. If so, it is determined whether the planning time difference between the more than two records with the same planning start and end point exceeds a threshold and whether there are other planning records between the more than two records. If the planning time difference does not exceed the threshold and no other records appear between two or more records, then for the two or more records with the same planning start and end point, the historical scene characteristics of any one of the records will be retained. If the time difference exceeds the threshold or other records appear between two or more records, the historical scene characteristics of each record are retained. If the waypoint appears in more than two historical navigation route planning records, but the start or end point of the historical navigation route planning records is different, then the historical scene features corresponding to each record are retained. Using the aforementioned waypoints as indexes and all the preserved historical scene features corresponding to each waypoint as keys, a historical waypoint sequence is generated.

7. The method according to claim 4 or 5, wherein, Obtaining the historical arrival point sequence of the navigated object includes: From the historical navigation route planning records of the navigated object, obtain the historical waypoint planning records, historical destination planning records, and corresponding navigation trajectory records of the navigated object; Based on the historical waypoint planning records and the historical destination planning records, the candidate arrival points of the navigated object are determined; The candidate arrival point is matched with the road segment corresponding to the navigation trajectory record. If the candidate arrival point matches the road segment corresponding to the navigation trajectory record, the candidate arrival point is used as the historical arrival point of the navigated object. If the candidate arrival point does not match the road segment corresponding to the navigation trajectory record, but the matching distance from the candidate arrival point to the road segment corresponding to the navigation trajectory record is less than a set distance threshold, then the candidate arrival point will be used as the historical arrival point of the navigated object. Based on the historical arrival points, the historical waypoint planning records, and the historical destination planning records, a historical arrival point sequence is determined.

8. A method for training a waypoint prediction model, wherein, include: The system obtains the historical waypoint sequence of the navigable object, sample waypoints of the target's historical navigation route, and sample scene features of the sample waypoints; wherein, the historical waypoint sequence includes data records indexed by historical waypoints, and each data record includes: historical waypoints and corresponding historical scene feature sequences, and the historical waypoints are historical waypoints in the historical navigation process before the target's historical navigation route; The sample scene features and the historical waypoint sequence are input into the model. The recommendation representation module of the model obtains the recommendation representation of the historical waypoint corresponding to the historical scene features based at least on the sample scene features and the historical scene features in the historical waypoint sequence. The recommended representation of the historical waypoints is input into the model, and the first scoring module of the model obtains a score for each historical waypoint based on the recommended representation of the historical waypoints. The loss function adjusts the model parameters of the modules included in the model based at least on the score of each historical path point and the sample path points used as sample labels.

9. The method according to claim 8, wherein, The waypoint prediction model further includes a second scoring module, and the method further includes: The second scoring module obtains a recommendation score based on the recommendation representation of the historical waypoints. The recommendation score is used to determine whether to recommend any historical waypoints to the navigated object. The loss function adjusts the model parameters of the modules included in the model based at least on the score of each historical path point and the sample path points used as sample labels, specifically including: The loss function adjusts the model parameters of the modules included in the model based on the recommendation score, the score of each historical path point, and the sample path points.

10. The method according to any one of claims 8-9, wherein, The method further includes: Obtain the historical arrival point sequence of the navigated object; wherein, the historical arrival point sequence includes a set of historical arrival points and feature data corresponding to each historical arrival point in the set of historical arrival points, the historical arrival points in the set of historical arrival points are sorted according to the arrival time of the historical arrival points, and the feature data corresponding to each historical arrival point includes: historical scene features, the category of the historical arrival point, and the cumulative navigation distance between two adjacent historical arrival points; the historical arrival points include historical waypoints that have actually been reached in the past or historical navigation destinations that have actually been reached in the past. The model includes a self-attention mechanism module that obtains a periodic representation of the historical arrival points based on the historical arrival point sequence. The model further includes a self-attention mechanism module and a periodic weight module, wherein the self-attention mechanism module obtains a periodic representation of each historical arrival point in the historical arrival point set based on the historical arrival point sequence. The periodic weighting module obtains the periodic weights of the historical transit points based on the periodic representation of the historical transit points belonging to the historical transit points in the historical transit point set. The periodic weights of the historical waypoints are used as input to the recommendation representation module; The recommendation representation module, based at least on the sample scene features and the historical scene features in the historical waypoint sequence, obtains the recommendation representation of the historical waypoints corresponding to the historical scene features, specifically implemented as follows: The recommendation representation module obtains the recommendation representation of the historical path points corresponding to the historical scene features based on the periodic weights, the sample scene features, and the historical scene features in the historical path point sequence.

11. An electronic device, wherein, The method includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method of any one of claims 1-10.

12. A computer-readable storage medium, wherein, The computer-readable storage medium stores computer instructions that, when executed by a processor, are used to implement the method described in any one of claims 1-10.

13. A computer program product comprising computer instructions, wherein, When executed by a processor, the computer instructions are used to implement the method described in any one of claims 1-10.

Images