Federated learning based secure prediction method, platform, device and medium

Abstract

The application provides a secure prediction method, platform, device and medium based on federated learning, which comprises the following steps: constructing road network meteorological data by using meteorological information of a plurality of point positions of a map grid; converging route trajectory coordinates by using a map; performing gradient homogenization calculation on the route trajectory coordinates by using the road network meteorological data to obtain route trajectory meteorological data; performing intersection calculation on the route trajectory meteorological data and traffic flow data, and converging the traffic flow data and the route trajectory meteorological data in one time axis; the traffic flow data comprises vehicle position and speed information provided by a road public facility and road traffic information; constructing a safety prediction model based on Bayesian probability distribution by taking the converged data as label data; performing federated learning by using a distributed computing device to train the safety prediction model; and performing safety prediction by using the safety prediction model. The application provides a privacy calculation method for multi-source heterogeneous data, and realizes driving assistance.

Description

Technical Field

[0001] This invention relates to the field of traffic safety, and in particular to a safety prediction method, platform, device, and medium based on federated learning. Background Technology

[0002] Analysis and statistics on accident types during high-speed driving revealed that reduced driving control due to weather conditions, as well as incorrect driving practices such as excessive or insufficient speed, illegal parking, and reversing, lead to frequent accidents that affect the safe operation of highways and the protection of drivers' lives and property.

[0003] The meteorological information used in navigation systems is regional forecasting and cannot provide weather forecasts for the entire navigation trajectory. This involves not only weather prediction but also the privacy of vehicles and drivers. Therefore, weather information, traffic information, and individual vehicle trajectory information are all independent. To predict traffic flow and weather impacts along a driving trajectory and make advance assessments, the highway management center, lacking individual vehicle information, can only manage clusters and cannot provide precise command. To improve the safety and efficiency of highway operations, individual vehicle location and speed information are fundamental data that needs to be used in the research and application of new technologies. However, the methods of using this information and the methods of protecting privacy also require further research.

[0004] The privacy technology of the Internet of Vehicles is a high-end technology research and development product for the latest cars, which is highly integrated with the car and operation management platform. However, for vehicles that are still in operation, how can we also realize the need to improve the driver's safety experience by providing multiple functions such as weather and traffic safety flow prediction and safe navigation guidance under the condition of privacy protection technology?

[0005] Location privacy is a special type of information privacy, referring to location-related information that individuals do not wish to be known by outsiders, as well as personal information revealed by location information. Sensitive locations such as hospitals, bars, and home addresses are generally considered part of a user's location privacy. Users can decide when, how, and to what extent they share their location information with others. Users are generally less willing to reveal their current or future location, but protecting past location is also important because past location can help attackers understand who you are, where you live, and what you have been doing. Users are willing to share their location with friends knowingly, but do not want mobile applications to automatically share their location without their knowledge. Users prefer to reveal obfuscated areas rather than their true location.

[0006] Patent CN111445714A discloses a predictable method for analyzing severe weather on highways. This invention discloses a predictable method for analyzing severe weather on highways, with the following steps: Step 1: Establish multiple meteorological information collection terminals at current locations to collect weather and meteorological information on and near the highway section. Each current location meteorological information collection terminal includes a temperature and humidity sensor, a current wind direction sensor, a current wind speed sensor, a current air pressure sensor, and a total radiation sensor. These sensors are all connected to a data collection module. This predictable method for analyzing severe weather on highways utilizes multiple current location meteorological information collection terminals to collect information on weather changes, improving the accuracy of weather forecasting and analysis. Furthermore, it can promptly feed information back to highway toll stations and drivers, facilitating highway toll stations. Road closures and traffic control are implemented. Patent CN111489577A discloses a highway disaster weather adaptive intelligent early warning real-time speed limit system. This system includes a 5G highway monitoring and command center, a 5G network, roadside units, and vehicle-mounted receiving terminals. Through various meteorological modules installed on different sections of the highway, it can collect severe weather information affecting traffic in real time based on weather changes and instantly generate warnings and speed limits suitable for vehicle travel on the current road section, thus achieving automatic speed limiting. This avoids the risk of traffic accidents caused by speeding in foggy, rainy, snowy, and windy weather. When the weather improves, it can automatically restore the highway speed limit to 120 km / h, thereby improving the highway's capacity and operational efficiency. In the event of an emergency, the 5G highway monitoring and command center sends speed limit commands through the roadside units, thereby implementing traffic control for vehicles traveling on each section. The above patent utilizes meteorological observation points along the highway to conduct meteorological observations, then achieves weather forecasting, and finally uses the generated weather forecasts to publish a weather service.

[0007] The academic paper "Research on Location Privacy Protection Based on Differential Privacy in VANETs," a master's thesis by Zhang Huijuan from Xi'an University of Electronic Science and Technology, discloses a location perturbation mechanism based on cooperation and caching that satisfies geographical indistinguishability for LBS (Location-Based Services) applications in VANETs (Vehicle Ad Hoc Networks). To address the issue of excessively rapid privacy budget consumption when using differential privacy in continuous LBS scenarios, this paper proposes a location perturbation mechanism that meets geographical indistinguishability requirements. By having requesting vehicles collaboratively construct groups, and selecting a group proxy to generate perturbed locations based on differential privacy and submit requests to the LSP (Location Service Provider) on behalf of the entire group, the paper achieves the goal of fulfilling the query needs of all members by consuming only the privacy budget of one member. This paper also utilizes differential privacy computation to solve the encryption problem of vehicle location services, thus protecting vehicle location and privacy information when VANETs provide location information for location services.

[0008] Existing navigation products or technical solutions are all based on a central hub with multiple sub-nodes, and customer information is unconditionally exposed to the product operation center. Moreover, it is a one-way functional display situation, which does not make good use of the relationships and needs between nodes. There is an urgent need for in-depth processing to create a service recommendation that runs for single-node customers. Summary of the Invention

[0009] The present invention aims to overcome at least one of the defects of the prior art and provide a secure prediction method, platform, device and medium based on federated learning.

[0010] In a first aspect, the present invention provides a secure prediction method based on federated learning, comprising:

[0011] Construct road network meteorological data using meteorological information from several points on a map grid;

[0012] Use the map to aggregate route trajectory coordinates;

[0013] Gradient homogenization calculation of road network meteorological data is performed on the route trajectory coordinates to obtain route trajectory meteorological data;

[0014] Intersection calculations are performed on route trajectory meteorological data and traffic flow data to merge traffic flow data and route trajectory meteorological data within the same time axis;

[0015] The traffic flow data includes vehicle location and speed information, as well as road traffic information, provided by road public facilities.

[0016] The aggregated data is used as labeled data to construct a security prediction model based on Bayesian probability distribution.

[0017] The security prediction model is trained using federated learning with distributed computing devices.

[0018] Security prediction is performed using a security prediction model.

[0019] Since meteorological data is not usually released according to traffic routes, the first step is to propose a method for aligning and predicting traffic trajectories with meteorological data. A meteorological prediction model based on route trajectories is then developed. This model uses road network information, meteorological forecast data, and a map grid to perform linear regression prediction based on meteorological forecasts from several nearby locations, ultimately generating meteorological data for the route trajectories.

[0020] By combining traffic data from the same timeline with meteorological and road network data, the changes in meteorological data within a time period can be stored in the traffic and road network database. This data can be arranged according to the time period, constructing a sequence mapping model between region and time. Traffic flow data can be aggregated into the data list of each road network segment. For example, it can be stored and managed according to a time period of 0-24 hours. Traffic flow changes within 24 hours are recorded in the database with a configured location ID. Meteorological data also cycles in 24-hour cycles, including precipitation, wind, and extreme weather changes. This is a cyclical storage repository. The addition of new information will pop the last piece of information and store it in the daily cycle repository. The daily cycle repository will then store this information monthly, and so on.

[0021] This invention establishes a safety prediction model based on the fusion of multi-source heterogeneous data. It can incorporate vehicle location and speed information, vehicle data from ETC and highway gantries, etc., to establish a traffic flow data prediction model. Since this patent does not aim to provide a new modeling method, but rather to provide a new reference point in terms of data acquisition, it adopts a relatively mature modeling method. The modeling method uses traffic flow prediction based on Bayesian probability distribution to achieve traffic safety prediction.

[0022] Traffic flow data can originate from highway ETC, highway gantries, and location and speed information provided by in-vehicle safety navigation devices. By segmenting highways using data from different locations and incorporating meteorological data as an independent feature, along with historical congestion data, traffic accident data, and other feature information, traffic flow is predicted based on a Bayesian probability distribution algorithm using federated learning. This invention can establish a comprehensive traffic flow prediction model and a weather-based safety prediction model through a virtual central node based on overall traffic flow and weather. This model can provide drivers with information on the traffic flow ahead and guide them to safe driving speeds, reminding the leading vehicle to accelerate and the following vehicles to maintain a safe distance. When weather conditions are poor, based on continuous weather forecasts along the highway trajectory, it can push weather information within a 1-kilometer radius, a 5-kilometer weather radius, and a 10-kilometer safety radius, providing guidance and prompts to drivers to ensure safe driving distance and speed.

[0023] The system can utilize navigation devices equipped with distributed computing capabilities within vehicles, employing a distributed computing approach based on federated learning and heterogeneous multi-source data fusion. Safety navigation devices placed within moving vehicles participate in federated learning computations, with local vehicle position and speed parameters serving as participating data. The trained prediction model then makes predictions based on real-time state parameters and incorporates weather information pushed along the high-speed driving trajectory to provide congestion and safety index alerts. Under the premise of privacy and security, it recommends safe driving strategies, including weather-based safe driving alerts, traffic flow prediction alerts, service area rest recommendations, and vehicle refueling / charging maintenance recommendations, among other safety-aiding recommendations. Furthermore, privacy technologies enable high-speed dispatching.

[0024] Furthermore, the use of distributed computing devices for federated learning includes:

[0025] Obtain user samples from several distributed computing devices, wherein the user samples come from at least two user groups, and each user group has a security prediction model;

[0026] Feature alignment is performed on overlapping user samples from different user groups;

[0027] Calculate intermediate gradient results for different user groups separately, and use public keys to encrypt and interact with the intermediate gradient results.

[0028] Each user group uses encrypted gradient values ​​for calculation, while simultaneously calculating the loss value of their respective label data;

[0029] The calculation results are summarized to calculate the total gradient value;

[0030] The total gradient value of the decryption is sent back to all user groups to complete federated learning.

[0031] Federated learning can consist of three parts. The first part is encrypted sample alignment. Since the user groups of two companies are not completely overlapping, the system uses encrypted user sample alignment technology to identify the common users of A and B without A and B disclosing their respective data, and without exposing non-overlapping users, so as to jointly model the features of these users.

[0032] Part Two: Encrypted Model Training. After identifying the shared user group, this data can be used to train a machine learning model. To ensure data confidentiality during training, encrypted training is required with the help of a third-party collaborator, C. Taking a linear regression model as an example, the training process can be divided into the following four steps:

[0033] Step 1: Collaborator C distributes the public key to A and B to encrypt the data that needs to be exchanged during the training process.

[0034] Step 2: A and B exchange intermediate results for gradient calculation in an encrypted manner.

[0035] Step 3: A and B calculate the gradient value based on the encrypted gradient value, while B calculates the loss based on its label data and summarizes the results to C. C calculates the total gradient value based on the summarized results and decrypts it.

[0036] Step 4: C sends the decrypted gradient back to A and B respectively, and A and B update the parameters of their respective models based on the gradient.

[0037] Iterate the above steps until the loss function converges, thus completing the entire training process. During sample alignment and model training, the data of both A and B are stored locally, and data interaction during training does not lead to data privacy leaks. Therefore, with the help of federated learning, the two parties are able to collaboratively train the model.

[0038] Furthermore, the gradient homogenization calculation of the route trajectory coordinates based on road network meteorological data includes:

[0039] Use SVM (Support Vector Machine) to calculate the road network meteorological data with the nearest coordinate distance for all route trajectories;

[0040] The nearest road network weather point is selected as the basic weight, and the weights of the weather points around the nearest road network weather point are calculated to achieve weight homogenization.

[0041] Furthermore, the route trajectory coordinates are selected by using slices of 0.5-5 kilometers to select the nearest road network weather points.

[0042] Furthermore, road traffic information includes historical congestion data and / or traffic accident data;

[0043] The safety prediction model is constructed using the historical congestion data and / or traffic accident data as labeled data.

[0044] The safety prediction model predicts safe driving speed and / or safe distance based on route trajectory coordinates.

[0045] A complete prediction model requires strong data support. Traffic flow data, meteorological data based on route trajectories, and vehicle-specific data can be aggregated into a complete data chain. The invention first addresses the meteorological aggregation problem by using gradient homogenization and support vector machines to predict meteorological data. Secondly, a device with autonomous positioning and distributed computing capabilities can be placed in the vehicle to train the federated learning model, keeping local data within the domain and enabling unified management. This achieves the fusion of multi-source heterogeneous information based on traffic flow data, positioning data, motion planning information, and meteorological forecast information, constructing a prediction model characterized by safety, efficiency, and economy.

[0046] Secondly, based on the same inventive concept, the present invention provides a federated learning-based safety prediction platform, including an in-vehicle mobile navigation device and a safety prediction control system.

[0047] The vehicle-mounted mobile navigation device is used to train a safety prediction model and to make safety predictions using the safety prediction model.

[0048] The safety predictive control system is used for safety predictive data control, including:

[0049] Construct road network meteorological data using meteorological information from several points on a map grid;

[0050] Use the map to aggregate route trajectory coordinates;

[0051] Gradient homogenization calculation of road network meteorological data is performed on the route trajectory coordinates to obtain route trajectory meteorological data;

[0052] Intersection calculations are performed on route trajectory meteorological data and traffic flow data to merge traffic flow data and route trajectory meteorological data within the same time axis;

[0053] The traffic flow data includes vehicle location and speed information, as well as road traffic information, provided by road public facilities.

[0054] The aggregated data is used as labeled data to build a secure prediction model based on Bayesian probability distribution.

[0055] Based on the same inventive concept, the present invention also provides a safety prediction system based on federated learning, including: a road network meteorological data construction module, used to construct road network meteorological data using meteorological information from several points on a map grid;

[0056] The coordinate aggregation module is used to aggregate route trajectory coordinates using a map;

[0057] The homogenization calculation module is used to perform gradient homogenization calculation on the road network meteorological data of the route trajectory coordinates to obtain the route trajectory meteorological data.

[0058] The intersection calculation module is used to perform intersection calculations on route trajectory meteorological data and traffic flow data, and to merge traffic flow data and route trajectory meteorological data within the same time axis;

[0059] A security prediction model, built on Bayesian probability distribution, is used for security prediction.

[0060] The federated learning module is used to train the security prediction model by performing federated learning using distributed computing devices.

[0061] Furthermore, the federated learning module includes:

[0062] The sample acquisition module is used to acquire user samples from several distributed computing devices. The user samples come from at least two user groups, and each user group has a security prediction model.

[0063] The feature alignment module is used to align features of overlapping user samples from different user groups;

[0064] The encryption module is used to calculate intermediate gradient results for different user groups and to encrypt and interact with the intermediate gradient results using a public key.

[0065] The calculation module is used to calculate the encrypted gradient values ​​for each user group, while also calculating the loss value of their respective label data.

[0066] The summary module is used to summarize the calculation results and calculate the total gradient value.

[0067] The backhaul module is used to send the total decrypted gradient values ​​back to all user groups to complete federated learning.

[0068] Based on the same inventive concept, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor implements the aforementioned federated learning-based security prediction method when executing the computer program.

[0069] Based on the same inventive concept, the present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the aforementioned federated learning-based security prediction method.

[0070] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0071] (1) Implement node computing for federated learning, and be able to achieve communication, payment, and integrated functions such as voice calling. Implement data communication and join the federated learning computing cluster as a single node.

[0072] (2) To achieve the prediction of meteorological data.

[0073] (3) Provide privacy calculations for multi-source heterogeneous meteorological traffic flow and vehicle driving parameters through a safe prediction model, provide driving assistance, and make predictions in combination with the real-time status of vehicles.

[0074] (4) Realize safe driving prediction based on real-time status parameters and recommend safe driving schemes under the premise of privacy and security. Attached Figure Description

[0075] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention.

[0076] Figure 2 This is a block diagram of the vehicle-mounted mobile navigation device according to Embodiment 2 of the present invention.

[0077] Figure 3 This is a block diagram of the control system of Embodiment 2 of the present invention. Detailed Implementation

[0078] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. To better illustrate the following embodiments, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0079] Example 1

[0080] like Figure 1 As shown, this embodiment provides a secure prediction method based on federated learning, including:

[0081] S1. Construct road network meteorological data using meteorological information from several points on the map grid;

[0082] S2. Use the map to gather the coordinates of the route trajectory;

[0083] S3. Perform gradient homogenization calculation on the road network meteorological data for the route trajectory coordinates to obtain the route trajectory meteorological data;

[0084] S4. Perform intersection calculations on the route trajectory meteorological data and traffic flow data to merge the traffic flow data and route trajectory meteorological data within the same time axis;

[0085] The traffic flow data includes vehicle location and speed information, as well as road traffic information, provided by road public facilities.

[0086] S5. Use the aggregated data as labeled data to construct a security prediction model based on Bayesian probability distribution;

[0087] S6. Use distributed computing devices to perform federated learning and train the security prediction model;

[0088] S7. Use a security prediction model to make security predictions.

[0089] Since meteorological data is not usually released according to traffic routes, the first step is to propose a method for aligning and predicting traffic trajectories with meteorological data. A meteorological prediction model based on route trajectories is then developed. This model uses road network information, meteorological forecast data, and a map grid to perform linear regression prediction based on meteorological forecasts from several nearby locations, ultimately generating meteorological data for the route trajectories.

[0090] By combining traffic data from the same timeline with meteorological and road network data, the changes in meteorological data within a time period can be stored in the traffic and road network database. This data can be arranged according to the time period, constructing a sequence mapping model between region and time. Traffic flow data can be aggregated into the data list of each road network segment. For example, it can be stored and managed according to a time period of 0-24 hours. Traffic flow changes within 24 hours are recorded in the database with a configured location ID. Meteorological data also cycles in 24-hour cycles, including precipitation, wind, and extreme weather changes. This is a cyclical storage repository. The addition of new information will pop the last piece of information and store it in the daily cycle repository. The daily cycle repository will then store this information monthly, and so on.

[0091] This invention establishes a safety prediction model based on the fusion of multi-source heterogeneous data. It can incorporate vehicle location and speed information, vehicle data from ETC and highway gantries, etc., to establish a traffic flow data prediction model. Since this patent does not aim to provide a new modeling method, but rather to provide a new reference point in terms of data acquisition, it adopts a relatively mature modeling method. The modeling method uses traffic flow prediction based on Bayesian probability distribution to achieve traffic safety prediction.

[0092] Traffic flow data can originate from highway ETC, highway gantries, and location and speed information provided by in-vehicle safety navigation devices. By segmenting highways using data from different locations and incorporating meteorological data as an independent feature, along with historical congestion data, traffic accident data, and other feature information, traffic flow is predicted based on a Bayesian probability distribution algorithm using federated learning. This invention can establish a comprehensive traffic flow prediction model and a weather-based safety prediction model through a virtual central node based on overall traffic flow and weather. This model can provide drivers with information on the traffic flow ahead and guide them to safe driving speeds, reminding the leading vehicle to accelerate and the following vehicles to maintain a safe distance. When weather conditions are poor, based on continuous weather forecasts along the highway trajectory, it can push weather information within a 1-kilometer radius, a 5-kilometer weather radius, and a 10-kilometer safety radius, providing guidance and prompts to drivers to ensure safe driving distance and speed.

[0093] The system can utilize navigation devices equipped with distributed computing capabilities within vehicles, employing a distributed computing approach based on federated learning and heterogeneous multi-source data fusion. Safety navigation devices placed within moving vehicles participate in federated learning computations, with local vehicle position and speed parameters serving as participating data. The trained prediction model then makes predictions based on real-time state parameters and incorporates weather information pushed along the high-speed driving trajectory to provide congestion and safety index alerts. Under the premise of privacy and security, it recommends safe driving strategies, including weather-based safe driving alerts, traffic flow prediction alerts, service area rest recommendations, and vehicle refueling / charging maintenance recommendations, among other safety-aiding recommendations. Furthermore, privacy technologies enable high-speed dispatching.

[0094] Preferably, the federated learning using distributed computing devices includes:

[0095] Obtain user samples from several distributed computing devices, wherein the user samples come from at least two user groups, and each user group has a security prediction model;

[0096] Feature alignment is performed on overlapping user samples from different user groups;

[0097] Calculate intermediate gradient results for different user groups separately, and use public keys to encrypt and interact with the intermediate gradient results.

[0098] Each user group uses encrypted gradient values ​​for calculation, while simultaneously calculating the loss value of their respective label data;

[0099] The calculation results are summarized to calculate the total gradient value;

[0100] The total gradient value of the decryption is sent back to all user groups to complete federated learning.

[0101] Federated learning can consist of three parts. The first part is encrypted sample alignment. Since the user groups of two companies are not completely overlapping, the system uses encrypted user sample alignment technology to identify the common users of A and B without A and B disclosing their respective data, and without exposing non-overlapping users, so as to jointly model the features of these users.

[0102] Part Two: Encrypted Model Training. After identifying the shared user group, this data can be used to train a machine learning model. To ensure data confidentiality during training, encrypted training is required with the help of a third-party collaborator, C. Taking a linear regression model as an example, the training process can be divided into the following four steps:

[0103] Step 1: Collaborator C distributes the public key to A and B to encrypt the data that needs to be exchanged during the training process.

[0104] Step 2: A and B exchange intermediate results for gradient calculation in an encrypted manner.

[0105] Step 3: A and B calculate the gradient value based on the encrypted gradient value, while B calculates the loss based on its label data and summarizes the results to C. C calculates the total gradient value based on the summarized results and decrypts it.

[0106] Step 4: C sends the decrypted gradient back to A and B respectively, and A and B update the parameters of their respective models based on the gradient.

[0107] Iterate the above steps until the loss function converges, thus completing the entire training process. During sample alignment and model training, the data of both A and B are stored locally, and data interaction during training does not lead to data privacy leaks. Therefore, with the help of federated learning, the two parties are able to collaboratively train the model.

[0108] Preferably, the gradient homogenization calculation of the route trajectory coordinates based on road network meteorological data includes:

[0109] Use SVM (Support Vector Machine) to calculate the road network meteorological data with the nearest coordinate distance for all route trajectories;

[0110] The nearest road network weather point is selected as the basic weight, and the weights of the weather points around the nearest road network weather point are calculated to achieve weight homogenization.

[0111] Preferably, the route trajectory coordinates are selected by using slices of 0.5-5 kilometers to select the nearest road network weather points.

[0112] Preferably, the road traffic information includes historical congestion data and / or traffic accident data;

[0113] The safety prediction model is constructed using the historical congestion data and / or traffic accident data as labeled data.

[0114] The safety prediction model predicts safe driving speed and / or safe distance based on route trajectory coordinates.

[0115] A complete prediction model requires strong data support. Traffic flow data, meteorological data based on route trajectories, and vehicle-specific data can be aggregated into a complete data chain. The invention first addresses the meteorological aggregation problem by using gradient homogenization and support vector machines to predict meteorological data. Secondly, a device with autonomous positioning and distributed computing capabilities can be placed in the vehicle to train the federated learning model, keeping local data within the domain and enabling unified management. This achieves the fusion of multi-source heterogeneous information based on traffic flow data, positioning data, motion planning information, and meteorological forecast information, constructing a prediction model characterized by safety, efficiency, and economy.

[0116] Example 2

[0117] This embodiment provides a safety prediction platform based on federated learning, including an in-vehicle mobile navigation device and a safety prediction control system;

[0118] The vehicle-mounted mobile navigation device is used to train a safety prediction model and to make safety predictions using the safety prediction model.

[0119] The safety predictive control system is used for safety predictive data control, including:

[0120] Construct road network meteorological data using meteorological information from several points on a map grid;

[0121] Use the map to aggregate route trajectory coordinates;

[0122] Gradient homogenization calculation of road network meteorological data is performed on the route trajectory coordinates to obtain route trajectory meteorological data;

[0123] Intersection calculations are performed on route trajectory meteorological data and traffic flow data to merge traffic flow data and route trajectory meteorological data within the same time axis;

[0124] The traffic flow data includes vehicle location and speed information, as well as road traffic information, provided by road public facilities.

[0125] The aggregated data is used as labeled data to build a secure prediction model based on Bayesian probability distribution.

[0126] like Figure 2As shown, the vehicle-mounted mobile navigation device aligns meteorological data, combines it with vehicle operating parameters and highway traffic flow data, exchanges data using a communication module, and uses a node computing unit, combined with a positioning and information processing module, to train a safety prediction model and then use the safety prediction model to make safety predictions. A human-machine interface is used to interact and assist the driver in making safety predictions.

[0127] like Figure 3 As shown, the safety prediction and control system matches and aligns high-speed navigation location data and meteorological data, aggregates high-speed traffic data, slices it by time period and location, predicts meteorological forecast data by high-speed location, associates meteorological and traffic safety characteristics, verifies the information, adds vehicle operation parameter data, and uses the multi-dimensional heterogeneous data fusion features to train the safety prediction model.

[0128] Example 3

[0129] This embodiment provides a safe prediction system based on federated learning, including: a road network meteorological data construction module, used to construct road network meteorological data using meteorological information from several points on a map grid;

[0130] The coordinate aggregation module is used to aggregate route trajectory coordinates using a map;

[0131] The homogenization calculation module is used to perform gradient homogenization calculation on the road network meteorological data of the route trajectory coordinates to obtain the route trajectory meteorological data.

[0132] The intersection calculation module is used to perform intersection calculations on route trajectory meteorological data and traffic flow data, and to merge traffic flow data and route trajectory meteorological data within the same time axis;

[0133] A security prediction model, built on Bayesian probability distribution, is used for security prediction.

[0134] The federated learning module is used to train the security prediction model by performing federated learning using distributed computing devices.

[0135] Preferably, the federated learning module includes:

[0136] The sample acquisition module is used to acquire user samples from several distributed computing devices. The user samples come from at least two user groups, and each user group has a security prediction model.

[0137] The feature alignment module is used to align features of overlapping user samples from different user groups;

[0138] The encryption module is used to calculate intermediate gradient results for different user groups and to encrypt and interact with the intermediate gradient results using a public key.

[0139] The calculation module is used to calculate the encrypted gradient values ​​for each user group, while also calculating the loss value of their respective label data.

[0140] The summary module is used to summarize the calculation results and calculate the total gradient value.

[0141] The backhaul module is used to send the total decrypted gradient values ​​back to all user groups to complete federated learning.

[0142] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A secure prediction method based on federated learning, characterized in that, include: Construct road network meteorological data using meteorological information from several points on a map grid; Use the map to aggregate route trajectory coordinates; Gradient homogenization calculation of road network meteorological data is performed on the route trajectory coordinates to obtain route trajectory meteorological data; Intersection calculations are performed on route trajectory meteorological data and traffic flow data to merge traffic flow data and route trajectory meteorological data within the same time axis; The traffic flow data includes vehicle location and speed information, as well as road traffic information, provided by road public facilities. The aggregated data is used as labeled data to construct a security prediction model based on Bayesian probability distribution. The security prediction model is trained using federated learning with distributed computing devices. Security prediction is performed using the aforementioned security prediction model; The gradient homogenization calculation of the route trajectory coordinates based on road network meteorological data includes: Use SVM (Support Vector Machine) to calculate the road network meteorological data with the nearest coordinate distance for all route trajectories; The nearest road network meteorological point is selected as the basic weight, and the meteorological forecasts issued by the meteorological points around the nearest road network meteorological point are used to perform linear regression prediction to form meteorological data of the route trajectory.

2. The secure prediction method based on federated learning according to claim 1, characterized in that, The use of distributed computing devices for federated learning includes: Obtain user samples from several distributed computing devices, wherein the user samples come from at least two user groups, and each user group has a security prediction model; Feature alignment is performed on overlapping user samples from different user groups; Calculate intermediate gradient results for different user groups separately, and use public keys to encrypt and interact with the intermediate gradient results. Each user group uses encrypted gradient values ​​for calculation, while simultaneously calculating the loss value of their respective label data; The calculation results are summarized to calculate the total gradient value; The total gradient value of the decryption is sent back to all user groups to complete federated learning.

3. The secure prediction method based on federated learning according to claim 1, characterized in that, The route trajectory coordinates are selected by using slices of 0.5-5 kilometers to select the nearest road network weather points.

4. The secure prediction method based on federated learning according to claim 1, characterized in that, Road traffic information includes historical congestion data and / or traffic accident data; The safety prediction model is constructed using the historical congestion data and / or traffic accident data as labeled data. The safety prediction model predicts safe driving speed and / or safe distance based on route trajectory coordinates.

5. A security prediction platform based on federated learning, applied to the security prediction method based on federated learning as described in claim 1, characterized in that, This includes in-vehicle mobile navigation devices and safety predictive control systems; The vehicle-mounted mobile navigation device is used to train a safety prediction model and to make safety predictions using the safety prediction model. The safety predictive control system is used for safety predictive data control, including: Construct road network meteorological data using meteorological information from several points on a map grid; Use the map to aggregate route trajectory coordinates; Gradient homogenization calculation of road network meteorological data is performed on the route trajectory coordinates to obtain route trajectory meteorological data; Intersection calculations are performed on route trajectory meteorological data and traffic flow data to merge traffic flow data and route trajectory meteorological data within the same time axis; The traffic flow data includes vehicle location and speed information, as well as road traffic information, provided by road public facilities. The aggregated data is used as labeled data to build a secure prediction model based on Bayesian probability distribution.

6. A security prediction system based on federated learning, applied to the security prediction method based on federated learning as described in claim 1, characterized in that, include: The road network meteorological data construction module is used to construct road network meteorological data using meteorological information from several points on the map grid. The coordinate aggregation module is used to aggregate route trajectory coordinates using a map; The homogenization calculation module is used to perform gradient homogenization calculation on the road network meteorological data of the route trajectory coordinates to obtain the route trajectory meteorological data. The intersection calculation module is used to perform intersection calculations on route trajectory meteorological data and traffic flow data, and to merge traffic flow data and route trajectory meteorological data within the same time axis; A security prediction model, built on Bayesian probability distribution, is used for security prediction. The federated learning module is used to train the security prediction model by performing federated learning using distributed computing devices.

7. The security prediction system based on federated learning according to claim 6, characterized in that, The federated learning module includes: The sample acquisition module is used to acquire user samples from several distributed computing devices. The user samples come from at least two user groups, and each user group has a security prediction model. The feature alignment module is used to align features of overlapping user samples from different user groups; The encryption module is used to calculate intermediate gradient results for different user groups and to encrypt and interact with the intermediate gradient results using a public key. The calculation module is used to calculate the encrypted gradient values ​​for each user group, while also calculating the loss value of their respective label data. The summary module is used to summarize the calculation results and calculate the total gradient value. The backhaul module is used to send the total decrypted gradient values ​​back to all user groups to complete federated learning.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the security prediction method based on federated learning as described in any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the security prediction method based on federated learning as described in any one of claims 1 to 4.

Images