A driving behavior analysis method based on federated K-means clustering

Abstract

The application relates to a driving behavior analysis method based on federal K-means clustering, and belongs to the field of machine learning and traffic safety. In view of the privacy leakage risk, low high-dimensional data processing efficiency and insufficient driving behavior classification accuracy of traditional methods, the technical scheme comprises the following steps: selecting a networked vehicle / RSU as a client, fusing high-precision map to extract space-time features, performing dimension reduction through an automatic encoder, realizing distributed iterative optimization through federal K-means clustering (probability weighted updating of a local clustering center, server aggregation to generate a global center), and finally classifying driving behaviors according to dynamic thresholds such as speed change and steering rate. Under the premise of protecting data privacy, the classification accuracy of the application reaches 92.7%, the calculation efficiency is improved by 3.2 times, the communication cost is reduced by 82%, and the traffic safety management level is effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of machine learning and traffic safety, and relates to a driving behavior analysis method based on federated K-means clustering. Background Technology

[0002] With the rapid development of connected and autonomous vehicles (CAVs) and vehicle-road cooperative technologies, driving behavior analysis has become a core technical means to improve road traffic safety and optimize autonomous driving decisions. Traditional driving behavior analysis methods mainly rely on centralized data processing architectures, collecting vehicle trajectory, speed, acceleration, and other data through onboard sensors or roadside equipment, and uploading them to a central server for unified modeling and analysis. However, this model faces multiple challenges in the practical implementation of intelligent transportation systems, specifically in the following aspects:

[0003] (1) Centralized processing requires vehicles to upload raw driving data (such as location, driving route, and driving habits) to third-party servers, which poses a risk of sensitive information leakage. Although technologies such as differential privacy and homomorphic encryption have been introduced, these methods often come at the cost of data utility, resulting in a 20%-30% decrease in model accuracy.

[0004] (2) In intelligent transportation scenarios, roadside units (RSUs), edge computing nodes, and connected vehicles generate more than 1PB of heterogeneous data (including high-precision maps, millimeter-wave radar point clouds, video streams, etc.) every day. Traditional K-means clustering algorithms face problems of high computational resource consumption and high communication latency when processing such data in a centralized architecture. Studies show that when the data scale exceeds 10^6 samples, the clustering time of traditional methods increases exponentially and cannot adapt to the limited computing power of edge devices.

[0005] (3) Driving behavior analysis heavily relies on deep feature mining of multi-source heterogeneous data. Existing methods are usually based on manually designed feature engineering (such as average speed, number of emergency braking), ignoring the dynamic correlation between vehicle trajectory and high-precision map (such as lane offset, intersection turning trajectory), resulting in insufficient feature representation capabilities. In addition, factors such as vehicle sensor noise (GPS positioning error ±2 meters) and communication packet loss (packet loss rate >15%) introduce a large number of outliers, and existing preprocessing methods (such as mean imputation) are difficult to effectively preserve the data distribution characteristics. Experiments show that noise interference can reduce clustering accuracy by more than 40%.

[0006] (4) Existing classification models, such as those based on SVM or decision trees, use fixed thresholds to classify driving behavior types (e.g., aggressive / conservative), which cannot adapt to the dynamic changes in complex traffic scenarios. For example, in rainy or snowy weather or congested traffic conditions, the steering rate, following distance, and other features of the same driver may deviate significantly from the preset thresholds, leading to an increased misclassification rate. In addition, the differences in the interaction behaviors between autonomous vehicles and human-driven vehicles in mixed traffic environments have not been fully considered.

[0007] (5) Although Federated Learning (FL) provides a new approach to distributed data privacy protection, its combination with traditional clustering algorithms still faces technical bottlenecks. Existing federated clustering methods (such as FedAvg-Kmeans) directly aggregate local cluster centers on an average basis, ignoring the heterogeneity (Non-IID) of data distribution among different clients. For example, the vehicle trajectory features at urban intersections and highways differ significantly, and direct global aggregation can lead to cluster center shifts, reducing model convergence speed by more than 50%. In addition, existing methods have not solved the compatibility problem between high-dimensional feature embedding and federated training, resulting in severe information loss of low-dimensional features during transmission. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide a driving behavior analysis method based on federated K-means clustering.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A driving behavior analysis method based on federated K-means clustering includes the following steps:

[0011] S1: Select networked vehicles or roadside communication units as clients participating in federated learning, fuse vehicle driving trajectory data with high-precision map information to extract spatiotemporal feature set X, preprocess the driving data and then perform feature dimensionality reduction through an autoencoder to obtain low-dimensional feature representation H;

[0012] S2: Distributed clustering analysis of low-dimensional feature representation H is performed using the federated K-means clustering algorithm, including multiple iterative processes such as initializing global cluster centers, client calculating local clustering parameters, and server aggregating and generating new cluster centers, until the global cluster centers converge.

[0013] S3: Classify driving behavior based on driving feature indicators in the clustering results, and generate classification results for different driving behavior categories.

[0014] Furthermore, in S1, the selected client includes a trusted execution environment, driving data is stored locally in the vehicle or on an edge device, and the preprocessing includes processing missing values ​​using linear interpolation, removing outliers based on the 3σ criterion, processing noise using moving average filtering, and eliminating dimensional differences using Z-score normalization and Min-Max normalization.

[0015] Furthermore, the extraction of the spatiotemporal feature set X includes: spatiotemporally matching the position, speed, and direction change data of the vehicle trajectory with the lane lines and traffic signs of the high-precision map to generate the position offset relative to the road elements, the slope of the speed change curve before the stop line, the standard deviation of the steering wheel angle, and the Pearson correlation coefficient between the vehicle distance and the speed.

[0016] Furthermore, the autoencoder comprises a three-layer encoder and a three-layer decoder. The encoder uses the ReLU activation function and the decoder uses the Sigmoid activation function. The reconstruction loss function is minimized using the Adam optimizer. The minimized reconstruction loss function is:

[0017]

[0018] in Indicates the losses incurred during reconstruction. Represents the original data. represents the reconstructed data; g(·) represents the decoding function; f(·) represents the encoding function.

[0019] Furthermore, the federated K-means clustering algorithm in S2 specifically includes:

[0020] The server initializes and broadcasts k cluster centers to the clients;

[0021] Each client calculates the probability of a data point belonging to its cluster and updates its local cluster center. The probability calculation formula is as follows:

[0022]

[0023] in Representing data points i Belongs to clustering j The probability, Indicates the client m data points i , Indicates the client m Cluster center j ;

[0024] After performing local K-means clustering, the client uploads the cluster centers and the number of samples.

[0025] The server generates new global cluster centers using a weighted average, calculated using the following formula:

[0026]

[0027] in express t+1 After round of iterations j A global cluster center, Indicates the client m The middle belongs to the cluster center j Quantity, Indicates the client m Local cluster centers j .

[0028] Furthermore, the client uses a probability-weighted calculation when updating the local cluster center:

[0029]

[0030] in Indicates the client m of j One updated cluster center; Let m be the total number of samples for client m.

[0031] Furthermore, in S3, the driving behavior classification criteria include:

[0032] Speed ​​change pattern classification: If the slope of the speed curve 5 seconds before the stop line is greater than -0.5m / s², it is considered aggressive behavior; if it is less than -1.0m / s², it is considered conservative behavior.

[0033] Steering rate classification: Steering wheel angle standard deviation less than 15° is considered aggressive driving, and greater than 25° is considered conservative driving;

[0034] Following vehicle characteristic classification: The correlation coefficient between following distance and speed is greater than 0.7, which is judged as conservative driving.

[0035] Furthermore, during the federated learning process, clustering parameters are transmitted between the client and the server using AES encryption, and the server only stores the aggregated global cluster centers.

[0036] Furthermore, the local K-means clustering is achieved by minimizing the clustering loss function:

[0037]

[0038] in Clustering loss; As a low-dimensional feature representation, It serves as the cluster center.

[0039] Furthermore, the high-precision map information includes lane topology, traffic sign locations, and three-dimensional intersection layout data, with the spatiotemporal matching error controlled within ±0.5 meters.

[0040] The beneficial effects of this invention are as follows:

[0041] (1) Based on the federated learning architecture, the original driving data (such as vehicle trajectory and driver ID) is always kept on the local client (connected vehicle / RSU), and only the cluster center parameters (such as probability-weighted feature vectors) are transmitted, thus avoiding the risk of data leakage from the system design level. By combining the Trusted Execution Environment (TEE) and AES encrypted transmission technology, the requirements of data privacy regulations such as GDPR and CCPA are met, thereby reducing the risk of data leakage.

[0042] By employing a probabilistic weighted federated clustering center aggregation strategy, the model bias problem caused by the distribution of Non-IID data is effectively mitigated. Experiments show that, with 100 heterogeneous clients (mixed urban / highway scenarios), the clustering accuracy (ARI index) reaches 0.892.

[0043] (2) By matching vehicle trajectories with high-precision maps in time and space, fine-grained features such as lane offset (error ±0.3 meters), slope of the speed curve before the stop line, and standard deviation of steering angle are extracted, expanding the feature dimension from the traditional 6 dimensions to 15 dimensions, significantly improving the classification recall rate. A three-layer autoencoder is used for feature compression, which reduces the feature dimension from 15 dimensions to 2 dimensions while retaining 95% of the original information, improving computational efficiency by 3.2 times (tested single inference time on Tesla T4 GPU <2ms). Combined with moving average filtering and 3σ outlier removal, the clustering purity (Purity index) can still be maintained at 88.6% under 20% Gaussian noise interference.

[0044] (3) Adaptively adjust the classification criteria based on the clustering results, for example:

[0045] Speed ​​variation pattern: Based on the distribution range of the slope of the speed curve 5 seconds before the stop line, the classification accuracy is improved.

[0046] Steering behavior recognition: The classification threshold is dynamically set by using the cluster center value of the standard deviation of the steering wheel angle, which reduces the false judgment rate.

[0047] In a test set that includes both human-driven and autonomous vehicles, heterogeneous driving patterns were captured through federated clustering, and the classification F1-score reached 0.916.

[0048] (4) By using autoencoder dimensionality reduction (2D feature transmission) and parameter encryption compression (AES-256), the amount of communication data per round for a single client is only 2.7KB, which reduces bandwidth usage compared to the traditional centralized method (transmitting 15-dimensional original features). The local K-means clustering task consumes an average of 8.3W of power on RSU devices (such as NVIDIA Jetson Xavier), supports concurrent processing of 50 clients, and meets the real-time requirements of edge computing. It supports a dynamic client join / exit mechanism, and at a scale of 1000 clients, the number of communication rounds required for global cluster center convergence only increases by 18% (from 150 rounds to 177 rounds).

[0049] (5) The clustering results of the aggressive driving category have a Spearman correlation coefficient of 0.78 with historical accident data, which can provide early warning for high-risk road sections. By analyzing the following characteristics of conservative drivers (vehicle distance-speed correlation coefficient > 0.7), a safe following strategy library is generated, which reduces the frequency of emergency braking of autonomous vehicles in mixed traffic scenarios. Based on the clustering results, a regional driving behavior heat map is generated to assist traffic management departments in optimizing traffic light timing schemes, and the measured intersection traffic efficiency is improved.

[0050] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0051] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0052] Figure 1 This is a flowchart of a driving behavior analysis method based on federated K-means clustering;

[0053] Figure 2 This is a structural diagram of a driving behavior analysis method based on federated K-means clustering;

[0054] Figure 3 This provides a visualization of the clustering results and driving behavior classification results for an instance of a driving behavior analysis method based on federated K-means clustering. Detailed Implementation

[0055] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0056] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, 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.

[0057] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0058] like Figure 1 As shown, the present invention describes a driving behavior analysis method based on federated K-means clustering, comprising:

[0059] S1, Data Acquisition. Select a connected vehicle or roadside unit (RSU) as the client to collect local vehicle driving data via vehicle-to-everything (V2X) communication.

[0060] S2, Data Preprocessing: This involves data type conversion, data cleaning, standardization, and normalization of the collected data to ensure its integrity and accuracy. Extracted features X are obtained, including information on vehicle position, speed, and direction over time, as well as detailed road information obtained from high-precision maps, such as lane lines, traffic signs, and intersection layouts.

[0061] S3, Autoencoder obtains low-dimensional representation. The extracted features X are initially extracted, dimensionality reduced, and denoised using an autoencoder to obtain the low-dimensional representation features H in the embedding space.

[0062] Specifically: The original data is mapped to a low-dimensional embedding space using the encoding function f(·), and then the original data is reconstructed from this embedding using the decoding function g(·). Optimization is achieved by minimizing the reconstruction loss, which is calculated using the following formula:

[0063]

[0064] in Indicates the losses incurred during reconstruction. Represents the original data. This indicates the reconstruction of data.

[0065] S4, Federated K-means Clustering, uses the Federated K-means clustering algorithm to perform cluster analysis on driving behavior data within the federated learning framework.

[0066] Specifically, the central server first initializes cluster centers and broadcasts them to each client (i.e., connected vehicles or roadside units). Each client calculates the probability that a data point belongs to that cluster center based on the received cluster centers, and uses this probability to update its local cluster centers.

[0067] The formula for calculating the probability that a data point belongs to the cluster center is:

[0068]

[0069] in Representing data points i Belongs to clustering j The probability, Indicates the client m data points i , Indicates the client m Cluster center j .

[0070] The formula for updating local cluster centers using probability is:

[0071]

[0072] in Indicates the client m of j An updated cluster center.

[0073] Subsequently, each client performs K-means clustering on its local data points to obtain local cluster centers and the corresponding number of samples.

[0074] For each data point, the client calculates its distance to the k cluster centers and assigns it to the nearest cluster. The K-means clustering algorithm is optimized by minimizing the clustering loss function. The formula for calculating the K-means clustering algorithm is:

[0075]

[0076] in This represents the clustering loss.

[0077] These local results are uploaded to the central server, which receives and aggregates the local results from each client, performs a weighted average, and calculates the new global cluster centers.

[0078] The new formula for calculating global cluster centers is:

[0079]

[0080] in express t+1 After round of iterations j A global cluster center, Indicates the client m The middle belongs to the cluster center j Quantity, Indicates the client m Local cluster centers j .

[0081] The new global cluster center is then distributed to each client for further local iterative optimization until the global cluster center converges or the predetermined number of iterations is reached.

[0082] S5: Based on the clustering results, driving behaviors are categorized into different types. The categorization criteria include driving characteristics such as speed changes before reaching the stop line, turning rate, and the correlation between the number of vehicles, vehicle distances, and vehicle speeds. For example, failing to reduce speed to zero before reaching the stop line is considered aggressive behavior; driving at a low speed until a complete stop is considered conservative behavior. Aggressive drivers have lower turning rates and tend to change lanes earlier to save time; while conservative drivers have higher turning rates and tend to drive in a straight line.

[0083] Based on the classification results of driving behavior, suggestions and measures are provided to improve traffic safety.

[0084] The specific implementation steps are as follows:

[0085] S1: Data Collection and Preprocessing

[0086] Step 1: Deploy 3 RSU clients at an intersection to collect trajectory data of 500 vehicles during red light periods (sampling rate 10Hz). Extract 6-dimensional features including speed reduction rate before the stop line, steering angle fluctuation, and distance to the vehicle in front.

[0087] Step 2: Clean the collected driving data, removing missing values, outliers, and noise. Then, standardize and normalize the data to eliminate dimensional differences between different features.

[0088] Step 3: Use an autoencoder to perform preliminary feature extraction, dimensionality reduction, and noise reduction on the cleaned features.

[0089] Specifically, the original data is mapped to a low-dimensional embedding space using the encoding function f(·), and then the original data is reconstructed from this embedding using the decoding function g(·). Optimization is achieved by minimizing the reconstruction loss, which is calculated using the following formula:

[0090]

[0091] in Indicates the losses incurred during reconstruction. Represents the original data. This indicates the reconstruction of data.

[0092] The autoencoder was trained for 500 epochs with a learning rate of 0.001.

[0093] S2: Federated K-means Clustering

[0094] Step 1: The central server initializes k=3 cluster centers and broadcasts them to each client.

[0095] Step 2: Each client calculates the probability that a data point belongs to a cluster center based on the received cluster center, and uses this probability to update the local cluster center.

[0096] The formula for calculating the probability that a data point belongs to a cluster center is as follows:

[0097]

[0098] in Representing data points i Belongs to clustering j The probability, Indicates the client m data points i , Indicates the client m Cluster center j .

[0099] The local cluster centers are updated using probability, calculated as follows:

[0100]

[0101] in Indicates the client m of jAn updated cluster center.

[0102] Step 3: Each client performs K-means clustering on local data points to obtain local cluster centers and the corresponding number of samples.

[0103] For each data point, the client calculates its distance to the k cluster centers and assigns it to the nearest cluster. This is optimized by minimizing the clustering loss function, calculated using the following formula:

[0104]

[0105] in This represents the clustering loss.

[0106] Step 4: These local results are uploaded to the central server, which receives and aggregates the local results from each client, performs a weighted average, and calculates the new global cluster centers.

[0107] The new formula for calculating global cluster centers is:

[0108]

[0109] in express t+1 After round of iterations j A global cluster center, Indicates the client m The middle belongs to the cluster center j Quantity, Indicates the client m Local cluster centers j .

[0110] The new global cluster center is distributed to each client for further local iterative optimization. Steps 2 to 4 are repeated until the global cluster center converges or the predetermined number of iterations is reached.

[0111] The federated K-means clustering algorithm is set to 20 local iterations and 150 communication rounds.

[0112] S3: Driving behavior classification, as shown in Table 1.

[0113] Table 1

[0114]

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A driving behavior analysis method based on federated K-means clustering, characterized in that: Includes the following steps: S1: Select networked vehicles or roadside communication units as clients participating in federated learning, fuse vehicle driving trajectory data with high-precision map information to extract spatiotemporal feature set X, preprocess the driving data and then perform feature dimensionality reduction through an autoencoder to obtain low-dimensional feature representation H; The extraction of the spatiotemporal feature set X includes: spatiotemporally matching the position, speed, and direction change data of the vehicle trajectory with the lane lines and traffic signs of the high-precision map to generate the position offset relative to the road elements, the slope of the speed change curve before the stop line, the standard deviation of the steering wheel angle, and the Pearson correlation coefficient between the vehicle distance and the speed. The preprocessing includes eliminating dimensional differences by using Z-score standardization and Min-Max normalization. S2: Distributed clustering analysis of low-dimensional feature representation H is performed using the federated K-means clustering algorithm, including multiple iterative processes of initializing global cluster centers, client calculating local clustering parameters, and server aggregating and generating new cluster centers, until the global cluster centers converge; The federated K-means clustering algorithm specifically includes: The server initializes and broadcasts k cluster centers to the clients; Each client calculates the probability of a data point belonging to its cluster and updates its local cluster center. The probability calculation formula is as follows: in Representing data points i Belongs to clustering j The probability, Indicates the client m data points i , Indicates the client m Cluster center j ; The client uses a probability-weighted calculation when updating the local cluster center: in Indicates the client m of j One updated cluster center; For the client m The total number of samples; After performing local K-means clustering, the client uploads the cluster centers and the number of samples. The server generates new global cluster centers using a weighted average, calculated using the following formula: in express t+1 After round of iterations j A global cluster center, Indicates the client m The middle belongs to the cluster center j Quantity, Indicates the client m Local cluster centers j ; S3: Classify driving behaviors based on driving characteristic indicators in the clustering results, generating classification results for different driving behavior categories; driving behavior classification criteria include: Speed ​​change pattern classification: If the slope of the speed curve 5 seconds before the stop line is greater than -0.5m / s², it is considered aggressive behavior; if it is less than -1.0m / s², it is considered conservative behavior. Steering rate classification: Steering wheel angle standard deviation less than 15° is considered aggressive driving, and greater than 25° is considered conservative driving; Following vehicle characteristic classification: The correlation coefficient between following distance and speed is greater than 0.7, which is judged as conservative driving.

2. The driving behavior analysis method based on federated K-means clustering according to claim 1, characterized in that: In S1, the selected client includes a trusted execution environment, driving data is stored locally in the vehicle or on an edge device, and the preprocessing also includes processing missing values ​​using linear interpolation, removing outliers based on the 3σ criterion, and processing noise using moving average filtering.

3. The driving behavior analysis method based on federated K-means clustering according to claim 1, characterized in that: The autoencoder comprises a three-layer encoder and a three-layer decoder. The encoder uses the ReLU activation function and the decoder uses the Sigmoid activation function. The reconstruction loss function is minimized using the Adam optimizer. The minimized reconstruction loss function is: in Indicates the losses incurred during reconstruction. Represents the original data. represents the reconstructed data; g(·) represents the decoding function; f(·) represents the encoding function.

4. The driving behavior analysis method based on federated K-means clustering according to claim 1, characterized in that: During the federated learning process, clustering parameters are transmitted between the client and the server using AES encryption, and the server only stores the aggregated global cluster centers.

5. The driving behavior analysis method based on federated K-means clustering according to claim 1, characterized in that: The local K-means clustering is achieved by minimizing the clustering loss function: in Clustering loss; As a low-dimensional feature representation, It serves as the cluster center.

6. The driving behavior analysis method based on federated K-means clustering according to claim 1, characterized in that: The high-precision map information includes lane topology, traffic sign locations, and three-dimensional intersection layout data, with the spatiotemporal matching error controlled within ±0.5 meters.

Images