Personalized path planning method and system based on federated learning and edge calculation

Abstract

The invention relates to the technical field of path planning, in particular to a personalized path planning method and system based on federated learning and edge calculation. Comprising a model training stage and a path planning stage. In a model training stage, a client quantifies travel experience of a user in an access region into a credibility score of a path, optimizes the credibility score of the path according to a path selection frequency in a user social network, and locally stores the credibility score of the path in a known region; the clients upload the path trust degree scores of the clients to the central server, global aggregation of the trust degree scores is carried out, and then the central server issues the aggregated trust degree scores to the clients; the invention discloses a personalized path planning method and system based on federal learning and edge calculation, provides an evaluation dimension of path trust degree for the personalized travel problem of a user, and can provide personalized path planning according to the personal preference of the user.

Description

Technical Field

[0001] The present invention relates to the field of path planning technology, and in particular to a personalized path planning method and system based on federated learning and edge computing. Background Art

[0002] With the development of society and the advancement of technology, people's demand for travel services is getting higher and higher. Users not only need basic navigation from point A to point B, but also hope to get the optimal route that takes into account personal habits, preferences and real-time traffic conditions.

[0003] In recent years, personalized route planning for drivers when traveling has become a major challenge in research. Existing research usually assumes that people share their sensitive travel data with cloud servers to find the route that best suits the user's personal preferences, but this is not ideal in practice due to concerns about user privacy. In addition, the existing work on route planning cannot effectively meet the needs of users for personalized travel. Therefore, how to provide users with personalized route planning while ensuring user privacy has become a problem that needs to be solved urgently in existing technologies. Summary of the invention

[0004] In order to solve the above-mentioned problems, the present invention provides a personalized path planning method and system based on federated learning and edge computing. Without involving any centralized personal privacy data, based on the travel experience of the user group, the route that best meets their personalized preferences under the travel conditions given by the user is found, while protecting the user's privacy information, providing users with personalized path planning. The efficiency of system transmission and data processing is improved by adding edge computing technology.

[0005] In a first aspect, the present invention provides a personalized path planning method based on federated learning and edge computing, which adopts the following technical solution: A personalized path planning method based on federated learning and edge computing, including a model training phase and a path planning phase; In the model training phase, the client quantifies the user's travel experience in the visited area into a path trust score, optimizes the path trust score based on the path selection frequency in the user's social network, and stores the trust score of the known area path locally; the client uploads its path trust score to the central server for global aggregation of the trust score, and then the central server sends the aggregated trust score to each client; In the path planning stage, the client uses the global path trust score of the user group and the path selection frequency of the target user's social network to screen out all target paths that meet the given travel conditions; then the client combines the user's personalized preferences and social network influencing factors to sort all possible target paths and output the sorted path recommendation results to the user.

[0006] Furthermore, the client quantifies the user's travel experience in the visited area as a trust score of the path, and optimizes the path trust score according to the path selection frequency in the user's social network. The specific steps are: determining the indicators used to evaluate the path trust, the client identifies the user's travel preference based on the user's historical travel data, and assigns a weight to each evaluation indicator; for all the user's historical travel paths, the client calculates the path trust score based on the score of each evaluation indicator and its weight, and uses the influence of social network data to further optimize the path trust score.

[0007] Furthermore, the indicators used to evaluate the path trustworthiness include path length, travel time, path safety and path congestion.

[0008] Furthermore, the client identifies the user's travel preference based on the user's historical travel data and assigns weights to each evaluation indicator. The specific steps are: the client analyzes the user's historical travel data to obtain the user's sensitivity to each evaluation indicator, assigns weights to each indicator, and ensures that the sum of all weights is 1.

[0009] Furthermore, the influence of social network data is used to further optimize the path trust score. The specific steps are: calculating the similarity between the user and other users in the social circle through a collaborative filtering algorithm, calculating the selection frequency of the path in the social network of the target user based on the similarity, and weighting the improvement of the path trust score based on the selection frequency of social circle members.

[0010] Furthermore, the client uses the global path trust score of the user group and the path selection frequency of the target user's social network to screen out all target paths that meet the travel conditions. The specific steps are: according to the travel conditions given by the user, the client preliminarily screens all paths that meet the user's travel needs based on the global path trust score and the path selection frequency of the social network.

[0011] Furthermore, the client sorts all possible target paths in combination with the user's personalized preferences and social network influencing factors. The specific steps are as follows: for the screened target paths, the client recalculates the path trust according to the user's personalized preferences and social network influencing factors, and arranges all target paths in descending order based on the calculation results.

[0012] The second aspect is a personalized path planning system based on federated learning and edge computing, including a model training module and a path planning module; The model training module is configured so that the client quantifies the user's travel experience in the visited area into a trust score of the path, and optimizes the path trust score according to the frequency of path selection in the user's social network. The user stores the trust score of the path in the known area on the local client; all clients upload their path trust scores to the central server for global aggregation of the trust scores, and then the central server sends the aggregated global trust scores to each client. The path trust transmission process in this stage is realized through the edge server; The route planning module is configured such that the client uses the global route trust score of the user group and the route selection frequency of the target user's social network to screen out all target routes that meet the given travel conditions; the client combines the user's personalized preferences and social network influencing factors to sort all possible target routes and output the sorted route recommendation results to the user.

[0013] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device for a personalized path planning method based on federated learning and edge computing.

[0014] In a fourth aspect, the present invention provides a terminal device including a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being suitable for being loaded by the processor and executing the personalized path planning method based on federated learning and edge computing.

[0015] In summary, the present invention has the following beneficial technical effects: The present invention discloses a personalized path planning method and system based on federated learning and edge computing. Aiming at the personalized travel problem of users, an evaluation dimension of path trust is proposed, and personalized path planning can be provided according to the personal preferences of users. Aiming at the data privacy protection of travelers, the present invention proposes to model sensitive travel data as a trust score of the path, and share the trust score among users, which can not only protect user privacy, but also use the group intelligence of users to assist personalized decision-making. In addition, when optimizing the path trust score based on the travel experience of similar users in the user's social network, the sensitive travel data of other users is not involved, and only their public non-sensitive travel information is used. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a framework diagram of the model training phase in the first embodiment of the present invention; Figure 2This is a framework diagram of the path planning stage in the first embodiment of the present invention; Figure 3 This is a comparison chart of the recall rates of various path planning methods in Example 1 of the present invention. DETAILED DESCRIPTION

[0017] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0018] Example 1 For mobile navigation software, the client quantifies the user's travel experience into a path trust score and optimizes it based on the frequency of path selection in the user's social network. The central server aggregates the path trust scores of all users and sends the global path trust score to each client. When the user enters his or her travel needs into the software, the client uses the global path trust score and the path selection frequency of the target user's social network to filter out all paths that meet the requirements. Then, the trust scores of all the initially selected paths are recalculated based on the user's personal preferences and social network factors, and are sorted in descending order and output to the user.

[0019] Reference Figure 1 In this embodiment, a personalized path planning method based on federated learning and edge computing is proposed. Aiming at the personalized travel problem of users, the method proposes an evaluation dimension of path trust and can provide personalized path planning according to the personal preferences of users. Aiming at the data privacy protection problem of travelers, it is proposed to model sensitive travel data as the trust score of the path, and share the trust score among users, which can not only protect user privacy, but also use the group intelligence of users to assist personalized decision-making. In addition, the process of optimizing the path trust score according to the user's social network does not involve the sensitive travel data of other users, but only uses their public non-sensitive travel information.

[0020] The method includes a model training phase and a path planning phase, respectively. Figure 1 and Figure 2 As shown. The main purpose of the model training phase is to use the path trust of the user group to assist the user's personalized path planning through distributed data processing and global model aggregation. The model training phase can be divided into three steps: local path trust calculation, trust upload and global aggregation, and global trust distribution. Local path trust calculation: The client quantifies the user's travel experience in the visited area as a trust score of the path, and optimizes it according to the frequency of path selection in the user's social network, and stores the trust score of the path in the known area locally. Trust upload and global aggregation: All clients upload their path trust scores to the central server for global aggregation of trust scores. Global trust distribution: The central server distributes the aggregated trust scores to each client.

[0021] The path planning stage is divided into two steps: preliminary screening of target paths and personalized path sorting. Preliminary screening of target paths: The client uses the global path trust score of the user group and the path selection frequency of the target user's social network to screen out all target paths that meet the given travel conditions. Personalized path sorting: The client sorts all possible target paths based on the user's personalized preferences and social network factors, and outputs the sorted path recommendation results to the user.

[0022] The specific steps of the above two stages are as follows: During the model training phase, the transmission process of all path trust scores is implemented through the edge server. The specific steps are as follows: Step 1: The client quantifies the user's travel experience in the visited area as a trust score of the path, and stores the trust scores of the paths in its known areas locally.

[0023] Step 2: All clients upload their path trust scores to the central server for global aggregation of trust scores.

[0024] Step 3: The central server sends the aggregated trust score to each client.

[0025] The path planning phase includes the following steps: Step 4: The client uses the global path trust score of the user group and the path selection frequency of the target user's social network to filter out all target paths that meet the given travel conditions.

[0026] Step 5: The client ranks all possible target paths based on the user's personalized preferences and social network factors, and outputs the ranked path recommendation results to the user.

[0027] In step 1, the client quantifies the user's travel experience in the visited area into the path trust score. The specific steps are as follows: determine the indicators used to evaluate the path trust. The client identifies the user's travel preferences based on the user's historical travel data and assigns weights to each evaluation indicator. For all users' historical travel paths, the client calculates the path trust score based on the score of each evaluation indicator and its weight. On this basis, we increase the influence of social network data and use collaborative filtering algorithms to further optimize the path trust score.

[0028] The indicators selected in this method include path length , travel time , path security , path congestion C. In order to meet the personalized needs of each user, based on the sensitivity shown in the user's historical travel data Assigning indicator weights , reflecting the importance users place on path characteristics. The sensitivity of each indicator It can be calculated by counting the percentage of selection times related to this indicator in the user's historical selection records: Indicates the number of user selections under the influence of a certain indicator. represents the total number of user selections. Therefore, the weight The specific calculation formula is: To ensure the rationality of all weights, the constraints must be met: By calculating the similarity between the target user and other users in the social circle, the selection frequency of social circle members will affect the trust score of the path. For example, if multiple social circle members choose a path, the trust score of the path will be weighted and improved. The similarity between the user and other users in the social circle is calculated through the collaborative filtering algorithm. The similarity calculation method uses cosine similarity. The formula is as follows: in, Indicates user and The similarity between Indicates user On the path Ratings, For users Average rating of . Based on similarity To calculate the selection frequency of the path in the target user's social network , It reflects the popularity of the path in the social circle. The specific calculation formula is as follows: in, Indicates user 's social circle members, Is a user Is the path selected? The indicator function of Selected the path ,but , otherwise 0.

[0029] Trust score is the final quantitative result of user path data. Multiply and sum the corresponding weights, and consider the influence of the target user's social network to get the path A comprehensive trust score : This score reflects both the objective performance of the path and incorporates the user's subjective preferences.

[0030] Afterwards, the client builds a personalized path trust score model for the user based on the path characteristics and the user's historical travel data, which is used for the trust score calculation in subsequent path planning. The model consists of a multi-layer perceptron (MLP) with the following structure: The input layer receives the feature vector of the path , d is the path feature dimension. There are two hidden layers, and the output of the first hidden layer is , the output of the second hidden layer is , ReLU is the activation function. The output layer uses a linear activation function to generate the predicted path trust score ,in is the weight matrix, For bias.

[0031] The model uses mean squared error (MSE) during training to evaluate the prediction quality of the model: in Score the trustworthiness of the real path, is the model prediction value, and m is the number of historical paths of the user. During the training process, the Adam optimizer is used to update the weight matrix ,in is the learning rate.

[0032] In step 2, each client completes the path trust score locally After the calculation, the score data will be uploaded to the central server through the edge server for subsequent global aggregation. To further enhance privacy protection, each client will score the trust before uploading data. Encryption or adding noise, specifically, adding Gaussian noise through differential privacy technology to prevent the leakage of rating data, can be expressed as: in is Gaussian noise.

[0033] Through the global aggregation process, the system can integrate the path trust scores scattered across various clients into a global path trust score while effectively protecting user privacy. This process uses the Federal Average Method (FedAvg), and its core calculation formula is: in is the number of clients participating in the aggregation, Indicates The amount of data per client, Indicates The path trust score of each client.

[0034] In step 4, the client needs to score the user's travel conditions and global path trustworthiness. and the frequency of path selection in social networks Perform a preliminary screening of candidate paths to generate a set of target paths that meet current requirements The core of this screening process is to set a confidence threshold. , user-specified travel conditions, and thresholds for social network path frequencies , filter and constrain the path set, and select the path set with high scores and meeting user needs. The screening formula is: in, is a set of travel conditions specified by the user, Indicates the path Meet the travel conditions specified by the user by setting the trust threshold , can effectively eliminate paths with low scores and potentially unreliable quality, thereby improving the overall quality of the target path set, Used to control the effect of social network selection frequency on path screening.

[0035] In step 5, the client needs to filter the target path set Perform personalized sorting. The core of sorting is to recalculate the trust score of each path. , we use the personalized path trust score model of the user in step 1 to calculate. This score not only considers the global attributes of the path, but also incorporates the user's personalized preferences. Specifically, after the user enters the path features, the model's predicted personalized trust score formula is: in, is the weight matrix. The weight matrix of each user is different, reflecting the user's attention to different path characteristics (such as path length, time, security, etc.).

[0036] After completing the personalized calculation of path trust, the system will The path in Sort in descending order. The sorting process is designed to prioritize routes that better meet user preferences and help users quickly find the most suitable travel plan.

[0037] After the path planning is completed, the user's feedback on the recommended path is collected, and the weight of the user's personalized trust scoring model is updated based on the feedback for the next path planning. The formula is as follows: in is the updated feature weight, is the feature weight before updating, f is the user feedback score, is the learning rate, and the scoring criteria of f are The calculation standard is consistent.

[0038] In order to verify the effectiveness of the method of this embodiment, the personalized path planning method based on federated learning and edge computing was experimented with a real data set, where the real data set comes from existing public data and is agreed by the information owner.

[0039] Experimental verification of personalized path planning: The experimental figure shows the recall rate of the proposed method and other comparison algorithms. Recall rate is an important indicator to measure the completeness of recommendation, which is defined as the proportion of all actual optimal paths that are successfully recommended.

[0040] In the experiment, three traditional path planning algorithms were compared with the proposed method. The results are as follows: Figure 3 As shown. Figure 3 It can be seen that as the number of training rounds increases, the recall rates of all algorithms are improved to varying degrees. The recall rate of the method in this embodiment is very close to that of the traditional algorithm, or even slightly higher than that of other algorithms. In addition, the method in this embodiment does not centrally collect user privacy data, which can protect user information to a certain extent, while other algorithms perform personalized path planning for users based on the centralized collection of user information. The method in this embodiment can not only achieve accurate path planning, but also ensure user privacy.

[0041] In summary, the present invention provides a personalized path planning method based on federated learning and edge computing, which realizes the integration of data sharing and group wisdom through the federated learning mechanism, avoids the privacy leakage and security risks that may be caused by traditional centralized data processing, and reduces the computing load of the central server with the help of the distributed processing capability of the edge server, significantly improving the stability of the upload process and the system operation efficiency. Through personalized path sorting, the system, on the basis of making full use of the global score, further combines the user's preference information and the influence of the user's social network factors, realizes the organic combination of global optimization and personalized needs, significantly improves the user experience, enables the system to adapt to the diverse user needs more flexibly, and provides strong support for intelligent path planning.

[0042] Example 2 This embodiment provides a personalized path planning system based on federated learning and edge computing, including a model training module and a path planning module; The model training module is configured so that the client quantifies the user's travel experience in the visited area into a trust score of the path, and optimizes the path trust score according to the frequency of path selection in the user's social network. The user stores the trust score of the path in the known area on the local client; all clients upload their path trust scores to the central server for global aggregation of the trust scores, and then the central server sends the aggregated global trust scores to each client. The path trust transmission process in this stage is realized through the edge server; The route planning module is configured such that the client uses the global route trust score of the user group and the route selection frequency of the target user's social network to screen out all target routes that meet the given travel conditions; the client combines the user's personalized preferences and social network influencing factors to sort all possible target routes and output the sorted route recommendation results to the user.

[0043] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for a personalized path planning method based on federated learning and edge computing.

[0044] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the personalized path planning method based on federated learning and edge computing.

[0045] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A personalized path planning method based on federated learning and edge computing, characterized in that: Includes model training phase and path planning phase; In the model training phase, the client quantifies the user's travel experience in the visited area into a path trust score, optimizes the path trust score based on the path selection frequency in the user's social network, and stores the trust score of the known area path locally; the client uploads its path trust score to the central server for global aggregation of the trust score, and then the central server sends the aggregated trust score to each client; In the path planning stage, the client uses the global path trust score of the user group and the path selection frequency of the target user's social network to screen out all target paths that meet the given travel conditions; then the client combines the user's personalized preferences and social network influencing factors to sort all possible target paths and output the sorted path recommendation results to the user.

2. According to claim 1, a personalized path planning method based on federated learning and edge computing is characterized in that: The client quantifies the user's travel experience in the visited area into a path trust score, and optimizes the path trust score according to the path selection frequency in the user's social network. The specific steps are: determining an indicator for evaluating the path trust, the client identifies the user's travel preference based on the user's historical travel data, and assigns a weight to each evaluation indicator; for all the user's historical travel paths, the client calculates the path trust score according to the score and weight of each evaluation indicator, and uses the influence of social network data to further optimize the path trust score.

3. The personalized path planning method based on federated learning and edge computing according to claim 2, characterized in that: The indicators used to evaluate the path trustworthiness include path length, travel time, path safety and path congestion.

4. The personalized path planning method based on federated learning and edge computing according to claim 3 is characterized in that: The client identifies the user's travel preference based on the user's historical travel data and assigns weights to each evaluation indicator. The specific steps are: the client analyzes the user's historical travel data to obtain the user's sensitivity to each evaluation indicator, assigns weights to each indicator, and ensures that the sum of all weights is 1.

5. The personalized path planning method based on federated learning and edge computing according to claim 4, characterized in that: The specific steps of further optimizing the path trust score by utilizing the influence of social network data are as follows: calculating the similarity between the user and other users in the social circle through a collaborative filtering algorithm, calculating the selection frequency of the path in the social network of the target user based on the similarity, and weighting and improving the path trust score based on the selection frequency of the social circle members.

6. The personalized path planning method based on federated learning and edge computing according to claim 5, characterized in that: The client uses the global path trust score of the user group and the path selection frequency of the target user's social network to screen out all target paths that meet the travel conditions. The specific steps are: according to the travel conditions given by the user, the client preliminarily screens all paths that meet the user's travel needs based on the global path trust score and the path selection frequency of the social network.

7. The personalized path planning method based on federated learning and edge computing according to claim 6, characterized in that: The client sorts all possible target paths in combination with the user's personalized preferences and social network influencing factors. The specific steps are as follows: for the screened target paths, the client recalculates the path trust according to the user's personalized preferences and social network influencing factors, and sorts all target paths in descending order based on the calculation results.

8. A personalized path planning system based on federated learning and edge computing, executing a personalized path planning method based on federated learning and edge computing as described in any one of claims 1 to 7, characterized in that: It includes model training module and path planning module; among them, The model training module is configured such that the client quantifies the user's travel experience in the visited area into the trust score of the path, and optimizes the path trust score according to the frequency of path selection in the user's social network. The user stores the trust score of the path in the known area on the local client; all clients upload their path trust scores to the central server for global aggregation of the trust scores, and then the central server sends the aggregated global trust scores to each client. The path trust transmission process in this stage is realized through the edge server; The route planning module is configured such that the client uses the global route trust score of the user group and the route selection frequency of the target user's social network to screen out all target routes that meet the given travel conditions; the client combines the user's personalized preferences and social network influencing factors to sort all possible target routes, and outputs the sorted route recommendation results to the user.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the method according to claim 1 .

10. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the method as claimed in claim 1 .

Images