Intelligent recommendation method and device based on driving behavior analysis

Abstract

According to the intelligent recommendation method and device based on driving behavior analysis provided by the embodiment of the invention, accurate analysis of user interests is realized by innovatively constructing a multi-dimensional feature fusion mechanism and integrating registration information, video data, driving data and scene perception data. And designing a trajectory prediction model based on a recurrent neural network, and establishing a feature weighted fusion strategy for intelligent matching in combination with a user interest modeling network of an attention mechanism. And an online learning mechanism is introduced, and model parameters are continuously optimized through clicking, staying duration and visit record data, so that dynamic adjustment of personalized recommendation contents is realized. According to the method, the defects of the traditional technology in the aspects of feature extraction, interest modeling, recommendation strategies and the like are effectively overcome, and the recommendation service level and the user experience in the driving scene are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to an intelligent recommendation method and device based on driving behavior analysis. Background Technology

[0002] Existing driving behavior analysis and recommendation methods have significant shortcomings. Traditional systems lack a systematic approach to feature extraction, making it difficult to effectively integrate registration information, video data, driving data, and scene-aware data, thus affecting the accuracy of user interest analysis.

[0003] Furthermore, existing technologies face bottlenecks in interest modeling. Most systems fail to comprehensively consider the weight differences of multi-dimensional features and lack feature fusion capabilities based on attention mechanisms, resulting in less than ideal recommendation accuracy.

[0004] Existing systems have technical shortcomings in recommendation strategies. They lack the ability to predict destinations and analyze arrival times, making it difficult to achieve personalized recommendations through distance scores and interest matching, thus impacting the service experience. Solving these problems is crucial for improving recommendation effectiveness. Summary of the Invention

[0005] To address the problems in existing technologies, this application provides an intelligent recommendation method and apparatus based on driving behavior analysis, which can effectively solve the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, and significantly improve the recommendation service level and user experience in driving scenarios.

[0006] To solve at least one of the above problems, this application provides the following technical solution:

[0007] Firstly, this application provides an intelligent recommendation method based on driving behavior analysis, including:

[0008] The system collects driver registration information data, vehicle camera video data, vehicle driving data, and scene perception data. It extracts user profile features from the registration information data, performs driving behavior recognition on the video data to obtain a behavior feature vector, extracts location information, driving duration, and driving trajectory from the vehicle driving data to construct a trajectory feature vector, and extracts time period identifiers, weather conditions, and road conditions from the scene perception data to construct a scene feature vector. It then constructs a trajectory prediction model based on a recurrent neural network and inputs the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time.

[0009] A user interest modeling network is constructed by inputting the user profile features, the behavioral feature vector, and the scene feature vector into the network. The features are weighted and fused based on an attention mechanism to obtain a fused feature vector. The fused feature vector is then input into a fully connected layer to generate a user interest vector. User click operation data, dwell time data, and visit record data for recommended content are collected. An online learning module is constructed based on the click operation data, dwell time data, and visit record data. The online learning module dynamically optimizes the parameters of the user interest modeling network.

[0010] Based on the destination coordinates, an interest point search range is constructed. Candidate location information within the search range is retrieved from the interest point database. The distance score between the candidate location and the destination coordinates is calculated. The interest matching score of the candidate location is calculated based on the user interest vector. The distance score and the interest matching score are weighted and fused to obtain a recommendation priority score. The location with the highest recommendation priority score is selected to generate a recommendation result. The push timing is determined based on the arrival time. The recommendation result is sent to the vehicle terminal display interface.

[0011] Furthermore, it also includes: obtaining driver registration information data and scene perception data from the vehicle terminal data interface; performing field parsing and structured processing on the driver registration information data; extracting driver identity information, vehicle information, and historical consumption data to construct user profile features; extracting the current time period identifier from the scene perception data; extracting weather condition information based on precipitation data and visibility data collected by the vehicle rain sensor; extracting road condition information based on road condition data and traffic flow data obtained by the vehicle communication module; and combining the time period identifier, the weather condition information, and the road condition information to construct a scene feature vector.

[0012] The system collects video data of the driver's actions using an in-vehicle camera, extracts frame sequences and detects human postures from the video data, and extracts parameters such as the driver's head rotation angle and body posture changes to construct a behavioral feature vector. It also obtains vehicle driving data from an in-vehicle positioning module and a driving recorder, performs spatiotemporal analysis on the vehicle driving data, and extracts vehicle position coordinate sequences, driving time data, and driving speed data. Based on the position coordinate sequences, it constructs a driving trajectory curve, and combines the driving trajectory curve, the driving time data, and the driving speed data to construct a trajectory feature vector.

[0013] Furthermore, it also includes: constructing a bidirectional long short-term memory neural network as an encoder for the trajectory prediction model, wherein the input dimension of the encoder corresponds to the dimension of the trajectory feature vector, the hidden layer contains multiple memory units, each memory unit consists of an input gate, a forget gate, and an output gate, inputting historical labeled data into the encoder to extract trajectory temporal features, inputting the trajectory temporal features into the decoder to generate a prediction sequence, training the trajectory prediction model using a teacher-forced training strategy, and saving the trained model parameters to a prediction model library;

[0014] The trajectory feature vector is input into the trained trajectory prediction model. The encoder extracts the temporal feature representation of the current trajectory. The decoder generates a future trajectory sequence based on the temporal feature representation. The future trajectory sequence is subjected to endpoint clustering analysis to obtain a set of candidate destinations. The probability score of each candidate destination is calculated. The coordinates with the highest probability score are selected as the predicted destination coordinates. The path distance is calculated based on the vehicle's current position and the predicted destination coordinates. The estimated arrival time is calculated based on the path distance and the current driving speed.

[0015] Furthermore, it also includes: constructing a multi-head self-attention module as a feature extraction layer; performing linear transformations on the user profile features, the behavior feature vector, and the scene feature vector to obtain a query matrix, a key matrix, and a value matrix, respectively; performing matrix multiplication on the query matrix and the key matrix to obtain an attention score matrix; performing softmax normalization on the attention score matrix to obtain an attention weight matrix; performing matrix multiplication on the attention weight matrix and the value matrix to obtain a weighted feature matrix; and concatenating the outputs of multiple attention heads to obtain multi-head attention features.

[0016] A feature fusion layer is constructed, and a residual connection structure is used to enhance the multi-head attention features. The enhanced features are standardized through a layer normalization module. The standardized features are input into a feedforward neural network for nonlinear transformation. The transformed features are then subjected to a second layer normalization to obtain a fused feature vector. The fused feature vector is randomly deactivated using a dropout mechanism. The processed features are then input into the next layer of the user interest modeling network.

[0017] Furthermore, it also includes: constructing a three-layer fully connected network as the interest modeling layer; the first fully connected layer performs dimensionality reduction mapping on the fused feature vector to obtain an intermediate feature representation; a nonlinear transformation is introduced by using the ReLU activation function; the second fully connected layer learns the user's preference distribution on different service types based on the intermediate feature representation; the third fully connected layer transforms the preference distribution into a fixed-dimensional user interest vector; and a batch normalization method is used to numerically stabilize the user interest vector.

[0018] An online learning module based on gradient boosting trees is constructed. User click operation data, page dwell time data, and actual visit record data are collected from the vehicle terminal. The click operation data, the dwell time data, and the visit record data are used to construct sample labels. The latest training sample set is maintained using a sliding time window method. The importance score of the model parameters is calculated based on the training sample set. The parameter update step size is adjusted according to the importance score. The weights of the user interest modeling network are updated online using the stochastic gradient descent method.

[0019] Furthermore, it also includes: constructing a circular search range based on the destination coordinates, taking the destination coordinates as the center of the circle, determining the search radius according to the vehicle speed, retrieving place information within the search range from the point of interest database using a geospatial indexing method, classifying and filtering the retrieved place information to obtain a candidate place set, and extracting the geographical coordinate information, business hours information, and service type information of each place in the candidate place set;

[0020] A distance scoring model and an interest matching model are constructed. The Manhattan distance is used to calculate the straight-line distance between the candidate location and the destination coordinates. The actual driving path length is calculated by combining road network data. The straight-line distance and the path length are weighted to obtain the distance score. The service type information of the candidate location is converted into a location feature vector. The cosine similarity between the location feature vector and the user interest vector is calculated to obtain the interest matching score.

[0021] Furthermore, it also includes: constructing a multi-objective optimization model to calculate recommendation priority; performing numerical normalization on the distance score and the interest matching score based on the min-max normalization method; calculating the fusion weight coefficient using an adaptive weight allocation strategy; weighting and summing the distance score and the interest matching score according to the weight coefficient to obtain the recommendation priority score; sorting the candidate location set in descending order according to the recommendation priority score; and selecting the location identifier, location name, location location, and service type with the highest ranking to generate recommendation results.

[0022] Based on the arrival time, a push rule model is constructed, the remaining travel time from the vehicle's current location to the recommended location is calculated, the push priority interval is divided based on the remaining travel time, the optimal push time is determined by combining the location's business hours information, the recommendation result is encapsulated into a message packet, and the message packet is sent to the display control module through the vehicle terminal's message queue service. The display control module displays the recommended content in a designated area of ​​the terminal interface at the optimal push time.

[0023] Secondly, this application provides an intelligent recommendation device based on driving behavior analysis, comprising:

[0024] The trajectory prediction module is used to collect driver registration information data, vehicle camera video data, vehicle driving data, and scene perception data. It extracts user profile features from the registration information data, performs driving behavior recognition on the video data to obtain a behavior feature vector, extracts location information, driving duration, and driving trajectory from the vehicle driving data to construct a trajectory feature vector, extracts time period identifiers, weather conditions, and road conditions from the scene perception data to construct a scene feature vector, constructs a trajectory prediction model based on a recurrent neural network, and inputs the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time.

[0025] An interest modeling module is used to construct a user interest modeling network. The user profile features, behavioral feature vectors, and scene feature vectors are input into the user interest modeling network. Based on an attention mechanism, the features are weighted and fused to obtain a fused feature vector. This fused feature vector is then input into a fully connected layer to generate a user interest vector. User click operation data, dwell time data, and visit record data for recommended content are collected. Based on the click operation data, dwell time data, and visit record data, an online learning module is constructed. This online learning module dynamically optimizes the parameters of the user interest modeling network.

[0026] The intelligent recommendation module is used to construct a search range of points of interest based on the destination coordinates, retrieve candidate location information within the search range from the point of interest database, calculate the distance score between the candidate location and the destination coordinates, calculate the interest matching score of the candidate location based on the user's interest vector, weight and fuse the distance score and the interest matching score to obtain a recommendation priority score, select the location with the highest recommendation priority score to generate a recommendation result, determine the push timing based on the arrival time, and send the recommendation result to the vehicle terminal display interface.

[0027] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the intelligent recommendation method based on driving behavior analysis.

[0028] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the intelligent recommendation method based on driving behavior analysis.

[0029] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the intelligent recommendation method based on driving behavior analysis.

[0030] As described above, this application provides an intelligent recommendation method and device based on driving behavior analysis. By innovatively constructing a multi-dimensional feature fusion mechanism, it integrates registration information, video data, driving data, and scene perception data to achieve accurate analysis of user interests. A trajectory prediction model based on a recurrent neural network is designed, combined with a user interest modeling network based on an attention mechanism, to establish a feature-weighted fusion strategy for intelligent matching. An online learning mechanism is introduced to continuously optimize model parameters using click, dwell time, and visit record data, enabling dynamic adjustment of personalized recommendation content. This method effectively addresses the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, significantly improving the recommendation service level and user experience in driving scenarios. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating the intelligent recommendation method based on driving behavior analysis in an embodiment of this application.

[0033] Figure 2 This is a structural diagram of the intelligent recommendation device based on driving behavior analysis in the embodiments of this application;

[0034] Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0035] Figure label:

[0036] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0039] To address the shortcomings of existing technologies, this application provides an intelligent recommendation method and device based on driving behavior analysis. By innovatively constructing a multi-dimensional feature fusion mechanism, it integrates registration information, video data, driving data, and scene perception data to achieve accurate analysis of user interests. A trajectory prediction model based on a recurrent neural network is designed, combined with a user interest modeling network based on an attention mechanism, to establish a feature-weighted fusion strategy for intelligent matching. An online learning mechanism is introduced to continuously optimize model parameters using click, dwell time, and visit record data, enabling dynamic adjustment of personalized recommendation content. This method effectively solves the deficiencies of traditional technologies in feature extraction, interest modeling, and recommendation strategies, significantly improving the recommendation service level and user experience in driving scenarios.

[0040] To effectively address the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, and to significantly improve recommendation service levels and user experience in driving scenarios, this application provides an embodiment of an intelligent recommendation method based on driving behavior analysis. See [link to embodiment]. Figure 1 The intelligent recommendation method based on driving behavior analysis specifically includes the following:

[0041] Step S101: Collect driver registration information data, vehicle camera video data, vehicle driving data, and scene perception data. Extract user profile features from the registration information data. Perform driving behavior recognition on the video data to obtain a behavior feature vector. Extract location information, driving duration, and driving trajectory from the vehicle driving data to construct a trajectory feature vector. Extract time period identifiers, weather conditions, and road conditions from the scene perception data to construct a scene feature vector. Construct a trajectory prediction model based on a recurrent neural network. Input the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time.

[0042] Optionally, this embodiment addresses the problems of incomplete feature extraction, inaccurate behavior analysis, and imprecise trajectory prediction in traditional in-vehicle recommendation systems by innovatively designing a driving behavior analysis scheme based on multi-dimensional perception. This embodiment first constructs a multi-source data acquisition network to obtain driver registration information through the data interface of the in-vehicle terminal. The system employs a data synchronization mechanism to integrate information from different sources: Feature_Score = w1User + w2Behavior + w3Trajectory + w4Scene, where each item represents a score for user features, behavior features, trajectory features, and scene features, respectively. w1 to w4 are dynamically adjusted weight coefficients, enabling comprehensive analysis of driving features.

[0043] This embodiment deeply optimizes the user profile construction mechanism. Driver registration information is structured, extracting multi-dimensional features including identity information, vehicle information, and consumption habits. The system processes unstructured text using field parsing technology and extracts key attributes using entity recognition methods. Particular attention is paid to users' time and location preferences; for example, historical consumption data is analyzed to predict user activity patterns at different times, and location check-in data is used to mine the types of places users frequently visit. These features provide foundational data support for subsequent interest modeling.

[0044] This embodiment innovatively implements a driving behavior analysis strategy. It captures driver action video using an in-vehicle high-definition camera and extracts key point coordinates using an improved human pose estimation algorithm. The system focuses on features such as head rotation angle, eye gaze direction, and hand gestures, which directly reflect the driver's attention distribution and operational intentions. Temporal analysis methods are used to identify abnormal driving behaviors; for example, frequent head turns may indicate searching for a target location, a behavior pattern often related to destination selection.

[0045] This embodiment deeply optimizes the trajectory feature extraction mechanism. Precise location and driving data are acquired from the vehicle positioning module and dashcam. The system employs a sliding window technique to process continuous trajectory data, extracting features including location coordinate sequences, driving duration, and average speed. Particular attention is paid to trajectory inflection points and stop points, as these feature points often contain important semantic information. A smooth trajectory curve is constructed using curve fitting methods, providing reliable input data for subsequent destination prediction.

[0046] This embodiment innovatively designs a scene feature analysis scheme. It collects comprehensive scene data based on the onboard environmental perception system, including time period information, weather conditions, and road condition information. The system assesses weather conditions using onboard rain and visibility sensors and obtains real-time road condition information via V2X communication. These environmental features are closely related to driving behavior and destination selection; for example, rainy days may increase the probability of choosing indoor locations, and congested road sections may lead to changes in route preferences.

[0047] This embodiment constructs a trajectory prediction model using recurrent neural network technology. A bidirectional LSTM network is employed to process trajectory sequence data, and a gating mechanism is used to capture the long-term and short-term dependencies of the trajectory. The system employs a hierarchical feature extraction structure, with the lower layer focusing on local trajectory patterns and the higher layer extracting global motion trends. An attention mechanism dynamically adjusts the feature weights at different time steps to improve the ability to identify key trajectory segments. The model output includes the predicted destination coordinates and estimated arrival time, providing spatiotemporal constraints for subsequent recommendation services.

[0048] This embodiment's innovative design not only solves the feature extraction and behavior analysis problems of traditional methods but also establishes a continuously optimizeable trajectory prediction framework. Through the fusion analysis of multi-dimensional features, the system can accurately understand the driver's behavioral intentions and destination preferences, providing a reliable decision-making basis for intelligent recommendations. This deep learning-based prediction mechanism ensures that the system maintains efficient analytical capabilities and reliable prediction results even when facing complex driving scenarios.

[0049] This embodiment achieves a precise understanding of driving behavior by establishing a complete data acquisition, feature extraction, and predictive analysis chain. The system can dynamically predict destinations based on real-time observed multi-dimensional features, avoiding the limitations of traditional fixed rules. Through deep learning and multimodal analysis, the accuracy and real-time performance of trajectory prediction are significantly improved, providing reliable spatiotemporal support for subsequent location recommendations. This intelligent analysis mechanism demonstrates strong scene adaptability and predictive accuracy in practical applications.

[0050] Step S102: Construct a user interest modeling network. Input the user profile features, the behavior feature vector, and the scene feature vector into the user interest modeling network. Based on the attention mechanism, perform weighted fusion of the features to obtain a fused feature vector. Input the fused feature vector into a fully connected layer to generate a user interest vector. Collect user click operation data, dwell time data, and visit record data for recommended content. Based on the click operation data, dwell time data, and visit record data, construct an online learning module. The online learning module dynamically optimizes the parameters of the user interest modeling network.

[0051] Optionally, this embodiment addresses the problems of insufficient feature fusion, inaccurate interest expression, and weak model adaptability in traditional user interest modeling by innovatively designing a dynamic interest modeling scheme based on deep learning. This embodiment first constructs a multi-layered feature processing network and uses an attention mechanism to achieve adaptive feature fusion: Attention_Score = softmax(QK^T / sqrt(d))V, where Q, K, and V represent the query vector, key vector, and value vector, respectively, and d is the feature dimension. This mechanism enables the evaluation and dynamic weighting of the importance of different features.

[0052] This embodiment deeply optimizes the design of the multi-head attention mechanism. Independent attention processing branches are constructed to address the different characteristics of user profile features, behavioral feature vectors, and scene feature vectors. The system first performs a linear transformation on each feature, generating a query matrix, key matrix, and value matrix. Parallel attention computation heads capture multi-dimensional correlations between features. For example, there may be a potential correlation between a user's consumption habits and driving behavior; the multi-head mechanism can simultaneously focus on feature patterns from different perspectives. The system employs a residual connection structure to enhance feature representation and layer normalization to ensure numerical stability.

[0053] This embodiment innovatively implements a feature fusion strategy. Based on the attention mechanism, a deep feature fusion network is constructed. The system uses a feedforward neural network to perform a nonlinear transformation on the attention output, and extracts high-order representations of features through a multilayer perceptron. Special attention is paid to the interaction effects of features; for example, certain scene features may strengthen or suppress the influence of specific behavioral features. A dropout mechanism is used to prevent overfitting and ensure the generalization ability of the fused features. This multi-layered fusion mechanism can fully explore the complex relationships between features, generating richer feature representations.

[0054] This embodiment deeply optimizes the interest vector generation mechanism. A three-layer fully connected network is constructed based on fused features to achieve the mapping from the feature space to the interest space. The first layer is responsible for feature dimensionality reduction, introducing a non-linear transformation through the ReLU activation function; the second layer learns the user's preference distribution across different service types; and the third layer generates fixed-dimensional interest vectors. The system employs batch normalization to ensure the stability of the numerical distribution and uses regularization methods to control model complexity. This hierarchical design enables the model to progressively extract the user's abstract interest features.

[0055] This embodiment innovatively designs an online learning framework. User interaction feedback data is collected in real time through an in-vehicle terminal, including click behavior on recommended content, page dwell time, and actual visit records. The system employs a sliding window mechanism to maintain a dynamic training sample set, ensuring that the learning process reflects the latest changes in user interests. Feature importance is analyzed using a gradient boosting tree model, and the parameter update strategy is dynamically adjusted: Learning_Rate = Base_Rate × Importance_Score, where Base_Rate is the base learning rate and Importance_Score is the feature importance score.

[0056] This embodiment achieves continuous optimization of the interest model through deep learning technology. Network parameters are updated using stochastic gradient descent, and the system continuously adjusts the model's prediction direction based on real-time feedback. Particular attention is paid to the temporal evolution characteristics of user interests, such as seasonal changes and periodic patterns. A dynamic weight adjustment mechanism ensures that the model can quickly respond to sudden changes in user interests. This adaptive learning mechanism significantly improves the timeliness and accuracy of the recommendation system.

[0057] This embodiment's innovative design not only solves the feature fusion and interest modeling problems in traditional methods but also establishes a continuously optimizing learning framework. By combining attention mechanisms and online learning, the system can accurately capture the dynamic changes in user interests, providing a reliable basis for intelligent recommendations. This deep learning-based modeling mechanism ensures that the system maintains high analytical capabilities and reliable predictive performance even when faced with complex user behaviors.

[0058] This embodiment achieves a precise understanding of user interests by establishing a complete feature fusion, interest modeling, and online optimization chain. The system can dynamically update the interest model based on real-time observed multi-dimensional features, avoiding the limitations of traditional fixed models. Through deep learning and online optimization, the accuracy and adaptability of interest prediction are significantly improved, providing reliable decision support for subsequent location recommendations. This intelligent modeling mechanism demonstrates strong generalization ability and prediction accuracy in practical applications.

[0059] Step S103: Construct an interest point search range based on the destination coordinates, retrieve candidate location information within the search range from the interest point database, calculate the distance score between the candidate location and the destination coordinates, calculate the interest matching score of the candidate location based on the user interest vector, weight and fuse the distance score and the interest matching score to obtain a recommendation priority score, select the location with the highest recommendation priority score to generate a recommendation result, determine the push timing based on the arrival time, and send the recommendation result to the vehicle terminal display interface.

[0060] Optionally, this embodiment addresses the problems of unreasonable search range, inaccurate interest matching, and improper timing of push notifications in traditional in-vehicle recommendation systems by innovatively designing an intelligent recommendation scheme based on multi-objective optimization. This embodiment first constructs a dynamic search range based on the predicted destination coordinates, employing an adaptive radius design: Search_Radius = Base_Radius × Speed_Factor × Time_Factor, where Base_Radius is the basic search radius, Speed_Factor is the speed adjustment factor, and Time_Factor is the time adjustment factor. This design ensures that the search range can be dynamically adjusted according to vehicle speed and remaining time, improving the real-time performance and rationality of the recommendations.

[0061] This embodiment deeply optimizes the candidate location retrieval mechanism. It employs geospatial indexing technology to efficiently retrieve points of interest within the target area. The system uses an R-tree structure to organize spatial data, enabling rapid range queries. Special attention is paid to location operating hours information, with time filters used to pre-filter options that do not meet time constraints. For example, locations whose expected arrival time exceeds operating hours are automatically filtered out. The system extracts multi-dimensional information from the retrieved locations, including geographic coordinates, service type, ratings, and other attributes, providing a data foundation for subsequent score calculations.

[0062] This embodiment innovatively implements a distance score calculation strategy. Combining actual road network data, the system employs a multi-source distance calculation method: Distance_Score = w1Manhattan_Distance + w2Road_Distance, where Manhattan_Distance is the straight-line distance, Road_Distance is the actual path distance, and w1 and w2 are weighting coefficients. By considering actual road conditions, the system can more accurately assess the accessibility of locations. Especially in complex road network environments, considering only straight-line distance may lead to recommendation bias, while combining actual path distance provides a more reasonable distance assessment.

[0063] This embodiment deeply optimizes the interest matching mechanism. Based on user interest vectors and location feature vectors, the system calculates the matching degree using an improved cosine similarity method: Interest_Score = cos(User_Vector, POI_Vector) × Category_Weight, where Category_Weight is the weight coefficient for the service type. This design not only considers the similarity of the vector space but also introduces a weighting mechanism based on category preferences. For example, for users who frequently visit a certain type of location, the weight of that category will be appropriately increased.

[0064] This embodiment innovatively designs a multi-objective fusion scheme. A priority scoring model is constructed, fusing distance and interest scores through an adaptive weighting mechanism: Priority_Score = αDistance_Score + βInterest_Score, where α and β are dynamically adjusted fusion weights. The system dynamically adjusts the weight ratios according to user preferences in different scenarios; for example, increasing distance weight when time is limited and increasing interest weight in leisure scenarios. This flexible weight adjustment achieves scenario adaptability of the recommendation results.

[0065] This embodiment achieves precise control over the timing of push notifications through intelligent decision-making technology. The system constructs a multi-level push strategy based on the estimated arrival time and designs a time-window-based triggering mechanism. Different push priorities are assigned based on the remaining travel time; for example, reminder pushes are sent when the destination is approaching, while planning pushes are sent when the destination is further away. Special attention is paid to driving safety factors; the system avoids sending disruptive pushes in complex road conditions or at high speeds.

[0066] This innovative design not only solves the problems of recommendation accuracy and timing control in traditional methods, but also establishes a continuously optimizeable recommendation framework. Through multi-objective optimization and dynamic weight adjustment, the system can accurately grasp user needs and scenario characteristics, providing drivers with the most suitable recommended content. This recommendation mechanism based on multi-dimensional analysis ensures that the system maintains efficient recommendation capabilities and reliable service quality when facing complex driving scenarios.

[0067] This embodiment achieves an intelligent upgrade of the recommendation system by establishing a complete link for location retrieval, score calculation, and push control. The system can dynamically adjust recommendation strategies based on real-time status, avoiding the limitations of traditional fixed rules. Through multi-objective optimization and scene awareness, the accuracy and timeliness of recommendations are significantly improved, providing drivers with more personalized and convenient location services. This intelligent recommendation mechanism demonstrates strong scene adaptability and user satisfaction in practical applications.

[0068] As described above, the intelligent recommendation method based on driving behavior analysis provided in this application can achieve accurate analysis of user interests by innovatively constructing a multi-dimensional feature fusion mechanism and integrating registration information, video data, driving data, and scene perception data. A trajectory prediction model based on a recurrent neural network is designed, combined with a user interest modeling network based on an attention mechanism, and a feature-weighted fusion strategy is established for intelligent matching. An online learning mechanism is introduced to continuously optimize model parameters through click, dwell time, and visit record data, enabling dynamic adjustment of personalized recommendation content. This method effectively solves the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, significantly improving the recommendation service level and user experience in driving scenarios.

[0069] In one embodiment of the intelligent recommendation method based on driving behavior analysis in this application, it may further include the following:

[0070] Step S201: Obtain driver registration information data and scene perception data from the vehicle terminal data interface; perform field parsing and structured processing on the driver registration information data; extract driver identity information, vehicle information, and historical consumption data to construct user profile features; extract the current time period identifier from the scene perception data; extract weather condition information based on precipitation data and visibility data collected by the vehicle rain sensor; extract road condition information based on road condition data and traffic flow data obtained by the vehicle communication module; and combine the time period identifier, the weather condition information, and the road condition information to construct a scene feature vector.

[0071] Step S202: Collect driver action video data through the vehicle-mounted camera, extract frame sequences and detect human posture from the video data, extract the driver's head rotation angle and body posture change parameters to construct a behavior feature vector, obtain vehicle driving data from the vehicle positioning module and driving recorder, perform spatiotemporal dimension analysis on the vehicle driving data, extract vehicle position coordinate sequence, driving time data and driving speed data, construct a driving trajectory curve based on the position coordinate sequence, and combine the driving trajectory curve, the driving time data and the driving speed data to construct a trajectory feature vector.

[0072] Optionally, this embodiment addresses the problems of incomplete data collection, insufficient feature extraction, and incomplete information integration in traditional vehicle systems by innovatively designing a feature extraction scheme based on multi-source perception. This embodiment first constructs a multi-dimensional data collection network, managing various types of sensor data uniformly through the vehicle terminal. The system designs a data synchronization mechanism: Sync_Score = w1User_Data + w2Scene_Data + w3Behavior_Data + w4Track_Data, where each item represents the temporal score of user data, scene data, behavior data, and trajectory data, respectively, and w1 to w4 are the corresponding weight coefficients, achieving temporal alignment and quality assessment of multi-source data.

[0073] This embodiment deeply optimizes the user information processing mechanism. A multi-layered parsing framework is designed to address the unstructured nature of driver registration information. The system employs entity recognition technology to extract key fields, including identity attributes, vehicle configuration, and historical behavior information. Particular attention is paid to users' consumption habits, using time-series analysis to uncover consumption patterns, such as commonly used payment methods, consumption time distribution, and location type preferences. These features are standardized to construct a unified user profile vector, providing foundational data support for subsequent interest modeling.

[0074] This embodiment innovatively implements a scene feature analysis strategy. It collects multi-dimensional scene data based on an onboard environmental perception system, including time information, weather conditions, and traffic patterns. The system monitors precipitation intensity using a high-precision rain sensor and assesses visibility conditions using a visibility sensor. For road condition information, real-time traffic data, including road congestion index, average vehicle speed, and accident information, is obtained through a V2X communication network. A scene feature fusion algorithm, Scene_Vector = Concat(Time_ID, Weather_State, Traffic_Flow), is used, where each component represents the time period code, weather state vector, and traffic flow characteristics, respectively, to achieve a unified representation of scene information.

[0075] This embodiment features a deeply optimized behavior analysis mechanism. Real-time video of the driver's actions is captured by an in-vehicle high-definition camera, and the system employs an adaptive frame rate control strategy for video sampling. For human posture detection, an improved keypoint detection algorithm is used to extract motion parameters of the head and torso. Particular attention is paid to abnormal behavioral patterns during driving; for example, frequent head turns may indicate searching for a target location, and leaning forward may indicate focused attention. These behavioral characteristics directly reflect the driver's intentions and state, and are of significant value for subsequent destination prediction.

[0076] This embodiment innovatively designs a trajectory feature extraction scheme. It integrates data from the vehicle positioning module and the dashcam to construct a precise trajectory tracking system. The system employs sliding window technology to process continuous position data and optimizes trajectory accuracy using a Kalman filter algorithm. Special attention is paid to key trajectory feature points, including turning points, stopping points, and acceleration / deceleration points, as these feature points often contain important behavioral semantics. A smooth trajectory representation is constructed using curve fitting methods, and combined with time-dimension information to form a complete motion description.

[0077] This embodiment utilizes multimodal sensing technology to construct feature vectors. The system employs a hierarchical feature organization structure to ensure effective fusion of features from different sources. For behavioral features, posture parameters and action sequences are encoded into fixed-dimensional vectors; for trajectory features, spatial location, temporal information, and velocity data are combined into a unified feature representation. Feature standardization ensures the consistency of numerical distribution, providing standardized input data for subsequent model training.

[0078] This embodiment's innovative design not only solves the data acquisition and feature extraction problems of traditional methods but also establishes a scalable feature engineering framework. Through the fusion analysis of multi-source data, the system can comprehensively characterize driving scenarios and user behavior, providing rich feature support for intelligent recommendations. This depth-perception-based feature extraction mechanism ensures that the system maintains efficient feature acquisition capabilities and reliable data quality even in complex driving environments.

[0079] This embodiment achieves accurate perception of driving scenarios by establishing a complete data acquisition, feature extraction, and information fusion chain. The system can dynamically update feature representations based on real-time observation data, avoiding the limitations of traditional fixed features. Through multimodal sensing and feature engineering, the richness and accuracy of feature representation are significantly improved, providing a reliable data foundation for subsequent model training and recommendation decisions. This intelligent feature extraction mechanism demonstrates strong environmental adaptability and feature representation capabilities in practical applications.

[0080] In one embodiment of the intelligent recommendation method based on driving behavior analysis in this application, it may further include the following:

[0081] Step S301: Construct a bidirectional long short-term memory neural network as the encoder of the trajectory prediction model. The input dimension of the encoder corresponds to the dimension of the trajectory feature vector. The hidden layer contains multiple memory units. Each memory unit consists of an input gate, a forget gate, and an output gate. Input historical labeled data into the encoder to extract trajectory temporal features. Input the trajectory temporal features into the decoder to generate a prediction sequence. Use a teacher-forced training strategy to train the trajectory prediction model. Save the trained model parameters to the prediction model library.

[0082] Step S302: Input the trajectory feature vector into the trained trajectory prediction model, extract the temporal feature representation of the current trajectory based on the encoder, generate a future trajectory sequence based on the temporal feature representation, perform endpoint clustering analysis on the future trajectory sequence to obtain a set of candidate destinations, calculate the probability score of each candidate destination, select the coordinates with the highest probability score as the predicted destination coordinates, calculate the path distance based on the vehicle's current position and the predicted destination coordinates, and calculate the estimated arrival time based on the path distance and the current driving speed.

[0083] Optionally, this embodiment addresses the problems of insufficient temporal feature extraction, low prediction accuracy, and inaccurate destination identification in traditional trajectory prediction methods by innovatively designing a deep learning-based trajectory prediction scheme. This embodiment first constructs a bidirectional LSTM network as an encoder, capturing the bidirectional dependencies of the trajectory sequence through forward and backward hidden layer states. The system designs a dedicated memory unit structure: Cell_State = ft × Ct-1 + it × Ct, where ft is the output of the forget gate, it is the output of the input gate, Ct-1 is the previous state, and Ct is the current candidate state. This gating mechanism enables the network to selectively retain and update trajectory information.

[0084] This embodiment features a deeply optimized memory unit design. Each LSTM unit comprises three control components: an input gate, a forget gate, and an output gate, which control the information flow via a sigmoid function. The input gate determines the degree of reception of new information at the current moment, the forget gate controls the proportion of historical information retained, and the output gate manages the output intensity of information. The system pays special attention to key turning points in the trajectory, such as turning at intersections or sudden speed changes, assigning higher weights to the input gates to ensure that important trajectory features are effectively captured. This fine-grained information control mechanism significantly enhances the model's ability to understand trajectory patterns.

[0085] This embodiment innovatively implements a bidirectional encoding strategy. By using a bidirectional LSTM network to simultaneously process trajectory sequences from both forward and backward directions, it captures complete contextual information about trajectory features. The system employs a feature fusion mechanism to merge bidirectional hidden state: Hidden_State = concat(forward_h, backward_h), where forward_h and backward_h represent the forward and backward hidden states, respectively. This bidirectional processing mechanism enables a better understanding of the global semantics of the trajectory, especially when dealing with complex path selection, by fully utilizing historical and future trajectory information.

[0086] This embodiment deeply optimizes the decoder design. Based on the temporal features extracted by the encoder, an autoregressive decoder is constructed to generate future trajectory sequences. The system employs an attention mechanism to dynamically focus on different parts of the encoder output, improving prediction accuracy. During the training phase, a teacher-forced strategy is used to accelerate model convergence, gradually reducing the proportion of teacher signals used to avoid the model over-reliance on real labels. This gradual training approach enhances the model's generalization ability.

[0087] This embodiment innovatively designs a destination clustering scheme. It performs destination analysis on the predicted trajectory sequence and uses a density clustering algorithm to identify possible destination sets. The system considers factors such as historical access frequency and temporal correlation to calculate the probability score for each candidate destination. Particular attention is paid to temporal patterns of driving habits; for example, some destinations may be strongly correlated with specific time periods. The most probable destination coordinates are determined through probability ranking, providing precise location constraints for subsequent recommendation services.

[0088] This embodiment achieves accurate arrival time prediction through intelligent algorithms. Based on the predicted destination and current location, the system calls upon road network data to calculate the optimal route distance. Combining real-time vehicle speed and historical driving data, an arrival time prediction model is established. The impact of road conditions is specifically considered, with the estimated time appropriately extended in congested sections. This dynamic time prediction mechanism provides users with more accurate trip planning support.

[0089] This embodiment's innovative design not only solves the trajectory prediction and destination identification problems of traditional methods but also establishes a continuously optimizing prediction framework. Through the combination of deep learning and intelligent clustering, the system can accurately understand the driver's driving intentions, providing reliable spatiotemporal constraints for intelligent recommendations. This deep learning-based prediction mechanism ensures that the system maintains efficient analytical capabilities and reliable prediction results even in complex driving scenarios.

[0090] This embodiment achieves accurate prediction of driving trajectories by establishing a complete link of feature encoding, sequence prediction, and destination analysis. The system can dynamically predict destinations and arrival times based on historical data and real-time observations, avoiding the limitations of traditional fixed rules. Through deep learning and dynamic prediction, the accuracy and timeliness of trajectory prediction are significantly improved, providing reliable decision support for subsequent location recommendations. This intelligent prediction mechanism demonstrates strong scene adaptability and prediction accuracy in practical applications.

[0091] In one embodiment of the intelligent recommendation method based on driving behavior analysis in this application, it may further include the following:

[0092] Step S401: Construct a multi-head self-attention module as a feature extraction layer. Perform linear transformations on the user profile features, the behavior feature vector, and the scene feature vector to obtain a query matrix, a key matrix, and a value matrix, respectively. Perform matrix multiplication on the query matrix and the key matrix to obtain an attention score matrix. Perform softmax normalization on the attention score matrix to obtain an attention weight matrix. Perform matrix multiplication on the attention weight matrix and the value matrix to obtain a weighted feature matrix. Concatenate the outputs of multiple attention heads to obtain multi-head attention features.

[0093] Step S402: Construct a feature fusion layer, use a residual connection structure to enhance the multi-head attention features, standardize the enhanced features through a layer normalization module, input the standardized features into a feedforward neural network for nonlinear transformation, perform a second layer normalization on the transformed features to obtain a fused feature vector, use a dropout mechanism to randomly deactivate the fused feature vector, and input the processed features into the next layer of the user interest modeling network.

[0094] Optionally, this embodiment addresses the problems of insufficient feature interaction, unreasonable dimension adaptation, and weak expressive power in traditional feature fusion methods by innovatively designing a feature extraction and fusion scheme based on an attention mechanism. This embodiment first constructs a multi-head self-attention module, generating three matrix spaces—query, key, and value—through linear transformation. The system implements the attention calculation mechanism: Attention(Q,K,V)=softmax(QK^T / sqrt(d))V, where Q, K, and V represent the query matrix, key matrix, and value matrix, respectively, d is the feature dimension, and sqrt(d) is a scaling factor used to prevent the softmax gradient from vanishing due to excessively large inner product values.

[0095] This embodiment deeply optimizes the multi-head attention mechanism. Independent attention processing branches are constructed for different types of features. The system uses a parameter matrix to project input features into different subspaces, with each attention head focusing on different aspects of the feature. For example, when processing user profile features, different attention heads may focus on different dimensions such as consumption habits, time preferences, and location preferences. Through the parallel multi-head mechanism, the system can simultaneously capture multi-dimensional relationships between features, enhancing the richness of feature representation.

[0096] This embodiment innovatively implements an attention weight calculation strategy. After obtaining the initial score through matrix multiplication, the system uses a softmax function for normalization, transforming the attention score into a probability distribution. This design ensures that the weights of different feature dimensions sum to 1, enabling the model to dynamically adjust the degree of attention given to different features. Especially when processing behavioral features, the system can adaptively adjust the attention allocation according to the current scenario; for example, when the driving state is abnormal, the relevant behavioral features will receive higher attention weights.

[0097] This embodiment deeply optimizes the feature fusion layer design. A residual connection structure is employed to enhance feature representation: Residual = LayerNorm(X + Attention(X)), where X is the input feature, Attention(X) is the attention output, and LayerNorm is the layer normalization operation. This residual structure not only preserves the original feature information but also introduces higher-order feature interactions, effectively alleviating the training difficulties of deep networks. The system ensures the stability of feature distribution through layer normalization, avoiding the impact of inconsistent feature scales on model training.

[0098] This embodiment innovatively designs a nonlinear transformation module. Standardized features are input into a feedforward neural network, and deep feature representations are extracted using a multilayer perceptron. The system employs the ReLU activation function to introduce a nonlinear transformation, enhancing the model's expressive power. Special attention is paid to the semantic consistency of features; a second layer of normalization ensures a reasonable numerical distribution of the fused features. A dropout mechanism is used to randomly deactivate some neurons, improving the model's generalization ability and robustness.

[0099] This embodiment achieves adaptive feature fusion through deep learning technology. The system dynamically adjusts the fusion weights according to the importance of different features to ensure that key information is fully represented. Especially when processing scene features, the system can adjust the feature fusion strategy according to the current driving environment, such as increasing the weight of road condition features in complex road conditions and strengthening the influence of weather features under special weather conditions.

[0100] This embodiment's innovative design not only solves the feature fusion problem in traditional methods but also establishes a scalable feature processing framework. By combining multi-head attention and deep fusion, the system can accurately capture complex relationships between features, providing rich feature support for user interest modeling. This deep learning-based feature processing mechanism ensures that the system maintains efficient feature extraction capabilities and reliable fusion results even when faced with complex user behaviors.

[0101] This embodiment achieves intelligent processing of multi-source features by establishing a complete feature extraction, attention calculation, and deep fusion chain. The system can dynamically adjust the feature fusion strategy based on real-time scenarios, avoiding the limitations of traditional fixed weights. Through multi-head attention and nonlinear transformation, the accuracy and adaptability of feature representation are significantly improved, providing a reliable feature foundation for subsequent interest modeling. This intelligent feature processing mechanism demonstrates strong scene adaptability and feature representation capabilities in practical applications.

[0102] In one embodiment of the intelligent recommendation method based on driving behavior analysis in this application, it may further include the following:

[0103] Step S501: Construct a three-layer fully connected network as the interest modeling layer. The first fully connected layer performs dimensionality reduction mapping on the fused feature vector to obtain an intermediate feature representation. The ReLU activation function is used to introduce a nonlinear transformation. The second fully connected layer learns the user's preference distribution on different service types based on the intermediate feature representation. The third fully connected layer transforms the preference distribution into a fixed-dimensional user interest vector. The batch normalization method is used to perform numerical stabilization processing on the user interest vector.

[0104] Step S502: Construct an online learning module based on gradient boosting tree. Collect user click operation data, page dwell time data, and actual visit record data from the vehicle terminal. Construct sample labels from the click operation data, the dwell time data, and the visit record data. Maintain the latest training sample set using a sliding time window method. Calculate the importance score of the model parameters based on the training sample set. Adjust the parameter update step size according to the importance score. Update the weights of the user interest modeling network online using stochastic gradient descent.

[0105] Optionally, this embodiment addresses the problems of unreasonable feature mapping, inaccurate preference learning, and untimely model updates in traditional interest modeling methods by innovatively designing a dynamic interest modeling scheme based on deep learning. This embodiment first constructs a three-layer fully connected network as the core architecture for interest modeling, achieving the mapping from fused features to interest representations through hierarchical feature transformation. The system designs a feature dimensionality reduction mechanism: Hidden = ReLU(W1×X+b1), where X is the fused feature vector, W1 is the weight matrix, b1 is the bias term, and ReLU is the activation function, realizing non-linear transformation and dimensionality reduction of features.

[0106] This embodiment deeply optimizes the intermediate feature representation mechanism. The first-layer fully connected network employs a dimensionality reduction mapping strategy to compress high-dimensional fused features into a suitable dimensional space. The system pays special attention to feature information preservation, introducing a nonlinear transformation through the ReLU activation function while maintaining feature sparsity. This design not only reduces computational complexity but also highlights the expression of key features. For example, when processing driving scenarios, the system can effectively extract important features relevant to the current context and filter out irrelevant noise information.

[0107] This embodiment innovatively implements a preference distribution learning strategy. The second-layer fully connected network focuses on learning the user's preference distribution across different service types. The system uses the Softmax activation function to convert the neuron output into a probability distribution: Preference = Softmax(W2 × Hidden + b2), where W2 is the weight matrix, b2 is the bias term, and Hidden is the intermediate feature of the first-layer output. This probabilistic representation intuitively reflects the strength of users' preferences for different types of services, providing a reliable basis for subsequent recommendation decisions.

[0108] This embodiment deeply optimizes the interest vector generation mechanism. The third-layer fully connected network transforms the preference distribution into fixed-dimensional interest vectors, and batch normalization is used to ensure the stability of the numerical distribution. The system eliminates distribution differences between different batches of data through normalization, improving the stability of model training. Special attention is paid to the expressive power of interest vectors to ensure accurate characterization of user interest changes in different scenarios.

[0109] This embodiment innovatively designs an online learning framework. A dynamic optimization module is built based on a gradient boosting tree model to collect user interaction feedback data in real time. The system employs multi-dimensional evaluation metrics, including click behavior, dwell time, and actual visit records, to construct a complete feedback evaluation system. A sliding time window mechanism is used to maintain the latest training samples, ensuring that the learning process can promptly reflect changes in user interests.

[0110] This embodiment achieves parameter optimization through feature importance analysis. The system calculates the importance score for each feature based on the training sample set, employing an adaptive learning rate strategy: Learning_Rate = Base_Rate × Importance_Score, where Base_Rate is the base learning rate and Importance_Score is the feature importance score. This dynamic adjustment mechanism ensures that important features receive larger update steps, accelerating the model's learning speed of key patterns.

[0111] This embodiment's innovative design not only solves the problems of interest modeling and parameter optimization in traditional methods, but also establishes a continuously evolving learning framework. Through the combination of deep learning and online optimization, the system can accurately capture the dynamic changes in user interests, providing a reliable basis for intelligent recommendations. This deep learning-based modeling mechanism ensures that the system maintains high learning efficiency and reliable predictive performance even when faced with complex user behaviors.

[0112] This embodiment achieves accurate modeling of user interests by establishing a complete feature mapping, preference learning, and online optimization chain. The system can dynamically adjust model parameters based on real-time feedback, avoiding the limitations of traditional fixed models. Through deep learning and dynamic optimization, the accuracy and adaptability of interest modeling are significantly improved, providing reliable decision support for subsequent location recommendations. This intelligent modeling mechanism demonstrates strong scene adaptability and predictive accuracy in practical applications.

[0113] In one embodiment of the intelligent recommendation method based on driving behavior analysis in this application, it may further include the following:

[0114] Step S601: Construct a circular search range based on the destination coordinates, take the destination coordinates as the center, determine the search radius according to the vehicle speed, use the geospatial indexing method to retrieve the location information within the search range from the point of interest database, classify and filter the retrieved location information to obtain a candidate location set, and extract the geographical coordinate information, business hours information, and service type information of each location in the candidate location set;

[0115] Step S602: Construct a distance scoring model and an interest matching model. Use Manhattan distance to calculate the straight-line distance between the candidate location and the destination coordinates. Combine road network data to calculate the actual driving path length. Weight the straight-line distance and the path length to obtain a distance score. Convert the service type information of the candidate location into a location feature vector. Calculate the cosine similarity between the location feature vector and the user interest vector to obtain an interest matching score.

[0116] Optionally, this embodiment addresses the problems of unreasonable search range, inaccurate distance calculation, and imprecise interest matching in traditional point-of-interest recommendation systems by innovatively designing a location retrieval and scoring scheme based on dynamic radius. This embodiment first constructs an adaptive search range, dynamically adjusting the search radius through a vehicle speed perception mechanism: Search_Radius = Base_Radius × (1 + α × Speed_Factor), where Base_Radius is the base search radius, Speed_Factor is the speed adjustment factor, and α is the adaptive coefficient. This design ensures that the search range can be dynamically adjusted with vehicle speed, expanding the search range at high speeds and achieving precise positioning at low speeds.

[0117] This embodiment deeply optimizes the geospatial indexing mechanism. To address the need for efficient retrieval of large-scale point-of-interest (POI) data, the system employs an improved R-tree structure to organize spatial data. A multi-level indexing mechanism organizes POIs hierarchically according to geographical location, with each node maintaining the minimum bounding rectangle covered by its subtree. Special attention is paid to the dynamic update requirements of the index, and efficient insertion and deletion strategies are designed. For example, when a change in POI information is detected, the system can quickly locate the affected index nodes and update them. This efficient indexing structure significantly improves the performance of spatial queries.

[0118] This embodiment innovatively implements a venue information filtering strategy. It extracts and categorizes retrieved venues from multiple dimensions. The system focuses on venue operating hours, using a time window filter to eliminate options that do not meet time constraints. It also considers service type matching, initially ranking candidate venues based on user historical preferences. This multi-layered filtering mechanism ensures the practicality and timeliness of the recommendation results.

[0119] This embodiment deeply optimizes the distance scoring model. The system adopts an improved Manhattan distance calculation method: Distance_Score = w1Manhattan_Distance + w2Path_Distance, where Manhattan_Distance is the grid distance, Path_Distance is the actual path distance, and w1 and w2 are dynamic weight coefficients. Special attention is paid to the impact of road network structure, and path distance evaluation is adjusted based on real-time traffic data. For example, the weight of path distance is appropriately increased in congested road sections, while straight-line distance is given more emphasis on unobstructed roads.

[0120] This embodiment innovatively designs an interest matching mechanism. It transforms venue service type information into standardized feature vectors and extracts semantic features of the venues using a deep learning model. The system employs an improved cosine similarity calculation method to evaluate the degree of interest matching: Interest_Score = cos(POI_Vector, User_Vector) * Category_Weight, where POI_Vector is the venue feature vector, User_Vector is the user interest vector, and Category_Weight is the category weight coefficient. This semantic-level matching mechanism can more accurately assess the fit between venues and user interests.

[0121] This embodiment utilizes deep learning technology to generate and match feature vectors. The system employs a pre-trained semantic encoder to transform location description information into a dense vector representation, capturing the deep semantic features of the location. Particular attention is paid to the diverse characteristics of locations; for example, restaurants may possess features of multiple cuisines, and shopping venues may contain multiple product category attributes. Through multi-dimensional feature encoding, the complete expression of location features is ensured.

[0122] This embodiment's innovative design not only solves the retrieval efficiency and matching accuracy problems of traditional methods but also establishes a scalable scoring framework. Through the combination of dynamic search and multi-dimensional matching, the system can accurately discover locations that meet user needs, providing a reliable candidate set for intelligent recommendations. This deep learning-based matching mechanism ensures that the system maintains high retrieval capabilities and reliable scoring performance even when faced with complex user needs.

[0123] This embodiment achieves precise filtering of points of interest by establishing a complete spatial retrieval, distance assessment, and interest matching chain. The system can dynamically adjust the search strategy based on real-time status, avoiding the limitations of traditional fixed-range methods. Through multi-dimensional scoring and intelligent matching, the accuracy and personalization of recommendations are significantly improved, providing users with more personalized location services. This intelligent recommendation mechanism demonstrates strong scene adaptability and user satisfaction in practical applications.

[0124] In one embodiment of the intelligent recommendation method based on driving behavior analysis in this application, it may further include the following:

[0125] Step S701: Construct a multi-objective optimization model to calculate the recommendation priority. Based on the min-max normalization method, normalize the distance score and the interest matching score numerically. Use an adaptive weight allocation strategy to calculate the fusion weight coefficient. Sum the distance score and the interest matching score according to the weight coefficient to obtain the recommendation priority score. Sort the candidate venue set in descending order according to the recommendation priority score. Select the venue identifier, venue name, venue location, and service type with the highest ranking to generate the recommendation result.

[0126] Step S702: Construct a push rule model based on the arrival time, calculate the remaining travel time from the vehicle's current location to the recommended location, divide the push priority interval based on the remaining travel time, determine the optimal push time by combining the location's business hours information, encapsulate the recommendation result into a message packet, and send the message packet to the display control module through the vehicle terminal's message queue service. The display control module displays the recommended content in a designated area of ​​the terminal interface at the optimal push time.

[0127] Optionally, this embodiment addresses the problems of single ranking strategies, unreasonable push timing, and poor user experience in traditional recommendation systems by innovatively designing an intelligent recommendation ranking and push scheme based on multi-objective optimization. This embodiment first constructs a multi-objective optimization model, employing min-max normalization to process the rating data: Score_Norm = (Score - Min_Score) / (Max_Score - Min_Score), where Score is the original score, and Min_Score and Max_Score are the minimum and maximum scores, respectively, ensuring that ratings from different dimensions are compared within the same numerical range.

[0128] This embodiment deeply optimizes the weight allocation mechanism. An adaptive weight calculation strategy is designed to address the different characteristics of distance scores and interest matching scores. The system dynamically evaluates the features of the current scenario and adjusts the weight allocation: Priority_Score = αDistance_Score + βInterest_Score, where α and β are adaptive weight coefficients, and α + β = 1. For example, in scenarios where users are short on time, the weight of the distance factor will be increased accordingly; in leisure situations, more emphasis will be placed on the degree of interest matching. This flexible weight adjustment mechanism ensures that the recommendation results can better adapt to different usage scenarios.

[0129] This embodiment innovatively implements a recommendation priority calculation strategy. Based on the weighted fusion scoring results, the system accurately ranks candidate locations. It pays particular attention to the comprehensive evaluation of locations, considering not only distance and interest dimensions but also real-time status information. For example, for locations nearing closure, even with high comprehensive scores, their ranking will be appropriately lowered. This multi-dimensional ranking mechanism ensures the practicality and timeliness of the recommendation results.

[0130] This embodiment features a deeply optimized push notification rule design. Based on the estimated arrival time, the system constructs a refined push notification control model. Remaining travel time is calculated using real-time traffic data, and a push notification time window is constructed by combining this with the operating hours of the locations. The system divides the remaining time into multiple priority intervals, such as an "emergency reminder interval," a "pre-planning interval," and a "reference information interval," employing differentiated push notification strategies for each interval. This time-based hierarchical push notification mechanism effectively avoids information interference and improves the accuracy of push notifications.

[0131] This embodiment innovatively designs a message encapsulation scheme. Recommendation results are organized into structured message packets containing key location information and display parameters. The system employs a hierarchical display strategy, prioritizing important information such as location name, distance, and estimated arrival time, while supplementing with detailed service information. This rational information organization ensures drivers can quickly access key information, reducing distraction.

[0132] This embodiment achieves reliable information push through message queue technology. An asynchronous communication mechanism is used to process push tasks, ensuring the reliability and real-time performance of message transmission. The system implements message priority management, giving higher priority to push content with higher urgency. Especially when network conditions are unstable, the system will initiate a message retransmission mechanism to ensure that important push information is delivered in a timely manner.

[0133] This embodiment's innovative design not only solves the sorting optimization and push control problems in traditional methods but also establishes a continuously optimizing recommendation display framework. Through the combination of multi-objective optimization and intelligent push, the system can accurately grasp user needs and scenario characteristics, providing drivers with the most suitable recommended content. This push mechanism based on multi-dimensional analysis ensures that the system maintains efficient recommendation capabilities and reliable service quality even in complex driving scenarios.

[0134] This embodiment achieves an intelligent upgrade of the recommendation system by establishing a complete priority calculation, push rules, and display control chain. The system can dynamically adjust its push strategy based on real-time status, avoiding the limitations of traditional fixed rules. Through multi-objective optimization and intelligent push, the accuracy of recommendations and user experience are significantly improved, providing drivers with more intelligent and user-friendly location services. This intelligent recommendation mechanism demonstrates strong scenario adaptability and user satisfaction in practical applications.

[0135] To effectively address the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, and to significantly improve recommendation service levels and user experience in driving scenarios, this application provides an embodiment of an intelligent recommendation device based on driving behavior analysis for implementing all or part of the aforementioned intelligent recommendation method based on driving behavior analysis. See [link to embodiment]. Figure 2 The intelligent recommendation device based on driving behavior analysis specifically includes the following:

[0136] The trajectory prediction module 10 is used to collect driver registration information data, vehicle camera video data, vehicle driving data, and scene perception data. It extracts user profile features from the registration information data, performs driving behavior recognition on the video data to obtain a behavior feature vector, extracts location information, driving duration, and driving trajectory from the vehicle driving data to construct a trajectory feature vector, extracts time period identifiers, weather conditions, and road conditions from the scene perception data to construct a scene feature vector, constructs a trajectory prediction model based on a recurrent neural network, and inputs the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time.

[0137] The interest modeling module 20 is used to construct a user interest modeling network. The user profile features, the behavioral feature vector, and the scene feature vector are input into the user interest modeling network. The features are weighted and fused based on an attention mechanism to obtain a fused feature vector. The fused feature vector is input into a fully connected layer to generate a user interest vector. The module collects user click operation data, dwell time data, and visit record data for recommended content. Based on the click operation data, dwell time data, and visit record data, an online learning module is constructed. The online learning module dynamically optimizes the parameters of the user interest modeling network.

[0138] The intelligent recommendation module 30 is used to construct a search range of points of interest based on the destination coordinates, retrieve candidate location information within the search range from the point of interest database, calculate the distance score between the candidate location and the destination coordinates, calculate the interest matching score of the candidate location based on the user interest vector, weight and fuse the distance score and the interest matching score to obtain a recommendation priority score, select the location with the highest recommendation priority score to generate a recommendation result, determine the push timing based on the arrival time, and send the recommendation result to the vehicle terminal display interface.

[0139] As described above, the intelligent recommendation device based on driving behavior analysis provided in this application can achieve accurate analysis of user interests by innovatively constructing a multi-dimensional feature fusion mechanism and integrating registration information, video data, driving data, and scene perception data. A trajectory prediction model based on a recurrent neural network is designed, combined with a user interest modeling network based on an attention mechanism, and a feature-weighted fusion strategy is established for intelligent matching. An online learning mechanism is introduced to continuously optimize model parameters through click, dwell time, and visit record data, achieving dynamic adjustment of personalized recommendation content. This method effectively solves the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, significantly improving the recommendation service level and user experience in driving scenarios.

[0140] From a hardware perspective, in order to effectively address the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, and significantly improve the recommendation service level and user experience in driving scenarios, this application provides an embodiment of an electronic device for implementing all or part of the aforementioned intelligent recommendation method based on driving behavior analysis. The electronic device specifically includes the following components:

[0141] The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the intelligent recommendation device based on driving behavior analysis and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the intelligent recommendation method based on driving behavior analysis and the embodiments of the intelligent recommendation device based on driving behavior analysis in the embodiments, the content of which is incorporated herein, and repeated details will not be described again.

[0142] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0143] In practical applications, the intelligent recommendation method based on driving behavior analysis can be partially executed on the electronic device side as described above, or all operations can be completed on the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed on the client device, the client device may further include a processor.

[0144] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0145] Figure 3 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 3 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 3 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0146] In one embodiment, the intelligent recommendation method based on driving behavior analysis can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0147] Step S101: Collect driver registration information data, vehicle camera video data, vehicle driving data, and scene perception data. Extract user profile features from the registration information data. Perform driving behavior recognition on the video data to obtain a behavior feature vector. Extract location information, driving duration, and driving trajectory from the vehicle driving data to construct a trajectory feature vector. Extract time period identifiers, weather conditions, and road conditions from the scene perception data to construct a scene feature vector. Construct a trajectory prediction model based on a recurrent neural network. Input the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time.

[0148] Step S102: Construct a user interest modeling network. Input the user profile features, the behavior feature vector, and the scene feature vector into the user interest modeling network. Based on the attention mechanism, perform weighted fusion of the features to obtain a fused feature vector. Input the fused feature vector into a fully connected layer to generate a user interest vector. Collect user click operation data, dwell time data, and visit record data for recommended content. Based on the click operation data, dwell time data, and visit record data, construct an online learning module. The online learning module dynamically optimizes the parameters of the user interest modeling network.

[0149] Step S103: Construct an interest point search range based on the destination coordinates, retrieve candidate location information within the search range from the interest point database, calculate the distance score between the candidate location and the destination coordinates, calculate the interest matching score of the candidate location based on the user interest vector, weight and fuse the distance score and the interest matching score to obtain a recommendation priority score, select the location with the highest recommendation priority score to generate a recommendation result, determine the push timing based on the arrival time, and send the recommendation result to the vehicle terminal display interface.

[0150] As described above, the electronic device provided in this application innovatively constructs a multi-dimensional feature fusion mechanism, integrating registration information, video data, driving data, and scene perception data to achieve accurate analysis of user interests. A trajectory prediction model based on a recurrent neural network is designed, combined with a user interest modeling network using an attention mechanism, to establish a feature-weighted fusion strategy for intelligent matching. An online learning mechanism is introduced to continuously optimize model parameters using click, dwell time, and visit record data, enabling dynamic adjustment of personalized recommendation content. This method effectively addresses the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, significantly improving the recommendation service level and user experience in driving scenarios.

[0151] In another embodiment, the intelligent recommendation device based on driving behavior analysis can be configured separately from the central processing unit 9100. For example, the intelligent recommendation device based on driving behavior analysis can be configured as a chip connected to the central processing unit 9100, and the intelligent recommendation method function based on driving behavior analysis can be implemented through the control of the central processing unit.

[0152] like Figure 3 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 3 All components shown; in addition, the electronic device 9600 may also include Figure 3 For components not shown, please refer to existing technologies.

[0153] like Figure 3 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0154] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0155] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0156] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0157] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device's communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0158] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0159] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored audio via the speaker 9131.

[0160] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the intelligent recommendation method based on driving behavior analysis, in which the execution subject is a server or client, as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the intelligent recommendation method based on driving behavior analysis, in which the execution subject is a server or client, as described in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0161] Step S101: Collect driver registration information data, vehicle camera video data, vehicle driving data, and scene perception data. Extract user profile features from the registration information data. Perform driving behavior recognition on the video data to obtain a behavior feature vector. Extract location information, driving duration, and driving trajectory from the vehicle driving data to construct a trajectory feature vector. Extract time period identifiers, weather conditions, and road conditions from the scene perception data to construct a scene feature vector. Construct a trajectory prediction model based on a recurrent neural network. Input the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time.

[0162] Step S102: Construct a user interest modeling network. Input the user profile features, the behavior feature vector, and the scene feature vector into the user interest modeling network. Based on the attention mechanism, perform weighted fusion of the features to obtain a fused feature vector. Input the fused feature vector into a fully connected layer to generate a user interest vector. Collect user click operation data, dwell time data, and visit record data for recommended content. Based on the click operation data, dwell time data, and visit record data, construct an online learning module. The online learning module dynamically optimizes the parameters of the user interest modeling network.

[0163] Step S103: Construct an interest point search range based on the destination coordinates, retrieve candidate location information within the search range from the interest point database, calculate the distance score between the candidate location and the destination coordinates, calculate the interest matching score of the candidate location based on the user interest vector, weight and fuse the distance score and the interest matching score to obtain a recommendation priority score, select the location with the highest recommendation priority score to generate a recommendation result, determine the push timing based on the arrival time, and send the recommendation result to the vehicle terminal display interface.

[0164] As described above, the computer-readable storage medium provided in this application embodiment innovatively constructs a multi-dimensional feature fusion mechanism, integrating registration information, video data, driving data, and scene perception data to achieve accurate analysis of user interests. A trajectory prediction model based on a recurrent neural network is designed, combined with a user interest modeling network using an attention mechanism, to establish a feature-weighted fusion strategy for intelligent matching. An online learning mechanism is introduced to continuously optimize model parameters using click, dwell time, and visit record data, enabling dynamic adjustment of personalized recommendation content. This method effectively addresses the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, significantly improving the recommendation service level and user experience in driving scenarios.

[0165] Embodiments of this application also provide a computer program product capable of implementing all steps of the intelligent recommendation method based on driving behavior analysis, where the execution subject is a server or client, as described in the above embodiments. When executed by a processor, this computer program / instruction implements the steps of the intelligent recommendation method based on driving behavior analysis. For example, the computer program / instruction implements the following steps:

[0166] Step S101: Collect driver registration information data, vehicle camera video data, vehicle driving data, and scene perception data. Extract user profile features from the registration information data. Perform driving behavior recognition on the video data to obtain a behavior feature vector. Extract location information, driving duration, and driving trajectory from the vehicle driving data to construct a trajectory feature vector. Extract time period identifiers, weather conditions, and road conditions from the scene perception data to construct a scene feature vector. Construct a trajectory prediction model based on a recurrent neural network. Input the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time.

[0167] Step S102: Construct a user interest modeling network. Input the user profile features, the behavior feature vector, and the scene feature vector into the user interest modeling network. Based on the attention mechanism, perform weighted fusion of the features to obtain a fused feature vector. Input the fused feature vector into a fully connected layer to generate a user interest vector. Collect user click operation data, dwell time data, and visit record data for recommended content. Based on the click operation data, dwell time data, and visit record data, construct an online learning module. The online learning module dynamically optimizes the parameters of the user interest modeling network.

[0168] Step S103: Construct an interest point search range based on the destination coordinates, retrieve candidate location information within the search range from the interest point database, calculate the distance score between the candidate location and the destination coordinates, calculate the interest matching score of the candidate location based on the user interest vector, weight and fuse the distance score and the interest matching score to obtain a recommendation priority score, select the location with the highest recommendation priority score to generate a recommendation result, determine the push timing based on the arrival time, and send the recommendation result to the vehicle terminal display interface.

[0169] As described above, the computer program product provided in this application innovatively constructs a multi-dimensional feature fusion mechanism, integrating registration information, video data, driving data, and scene perception data to achieve accurate analysis of user interests. It designs a trajectory prediction model based on a recurrent neural network, combines it with a user interest modeling network based on an attention mechanism, and establishes a feature-weighted fusion strategy for intelligent matching. An online learning mechanism is introduced to continuously optimize model parameters using click, dwell time, and visit record data, enabling dynamic adjustment of personalized recommendation content. This method effectively addresses the shortcomings of traditional technologies in feature extraction, interest modeling, and recommendation strategies, significantly improving the recommendation service level and user experience in driving scenarios.

[0170] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0171] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0172] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0173] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0174] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. An intelligent recommendation method based on driving behavior analysis, characterized in that, The method includes: The system collects driver registration information data, vehicle camera video data, vehicle driving data, and scene perception data. It extracts user profile features from the registration information data, performs driving behavior recognition on the video data to obtain a behavior feature vector, extracts location information, driving duration, and driving trajectory from the vehicle driving data to construct a trajectory feature vector, and extracts time period identifiers, weather conditions, and road conditions from the scene perception data to construct a scene feature vector. It then constructs a trajectory prediction model based on a recurrent neural network and inputs the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time. A user interest modeling network is constructed by inputting the user profile features, the behavioral feature vector, and the scene feature vector into the network. The features are weighted and fused based on an attention mechanism to obtain a fused feature vector. The fused feature vector is then input into a fully connected layer to generate a user interest vector. User click operation data, dwell time data, and visit record data for recommended content are collected. An online learning module is constructed based on the click operation data, dwell time data, and visit record data. The online learning module dynamically optimizes the parameters of the user interest modeling network. Based on the destination coordinates, an interest point search range is constructed. Candidate location information within the search range is retrieved from the interest point database. The distance score between the candidate location and the destination coordinates is calculated. The interest matching score of the candidate location is calculated based on the user interest vector. The distance score and the interest matching score are weighted and fused to obtain a recommendation priority score. The location with the highest recommendation priority score is selected to generate a recommendation result. The push timing is determined based on the arrival time. The recommendation result is sent to the vehicle terminal display interface.

2. The intelligent recommendation method based on driving behavior analysis according to claim 1, characterized in that, The process involves collecting driver registration information data, in-vehicle camera video data, vehicle driving data, and scene perception data. User profile features are extracted from the registration information data. Driving behavior recognition is performed on the video data to obtain a behavior feature vector. Location information, driving duration, and driving trajectory are extracted from the vehicle driving data to construct a trajectory feature vector. Time period identifiers, weather conditions, and road conditions are extracted from the scene perception data to construct a scene feature vector. This includes: Driver registration information and scene perception data are obtained from the vehicle terminal data interface. The driver registration information data is parsed and structured to extract driver identity information, vehicle information, and historical consumption data to construct user profile features. The current time period identifier is extracted from the scene perception data. Weather condition information is extracted based on precipitation data and visibility data collected by the vehicle rain sensor. Road condition information is extracted based on road condition data and traffic flow data obtained by the vehicle communication module. The time period identifier, weather condition information, and road condition information are combined to construct a scene feature vector. The system collects video data of the driver's actions using an in-vehicle camera, extracts frame sequences and detects human postures from the video data, and extracts parameters such as the driver's head rotation angle and body posture changes to construct a behavioral feature vector. It also obtains vehicle driving data from an in-vehicle positioning module and a driving recorder, performs spatiotemporal analysis on the vehicle driving data, and extracts vehicle position coordinate sequences, driving time data, and driving speed data. Based on the position coordinate sequences, it constructs a driving trajectory curve, and combines the driving trajectory curve, the driving time data, and the driving speed data to construct a trajectory feature vector.

3. The intelligent recommendation method based on driving behavior analysis according to claim 1, characterized in that, The trajectory prediction model constructed based on a recurrent neural network, which inputs the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time, includes: A bidirectional long short-term memory neural network is constructed as the encoder of the trajectory prediction model. The input dimension of the encoder corresponds to the dimension of the trajectory feature vector. The hidden layer contains multiple memory units, each of which consists of an input gate, a forget gate, and an output gate. Historical labeled data is input into the encoder to extract trajectory temporal features. The trajectory temporal features are input into the decoder to generate a prediction sequence. The trajectory prediction model is trained using a teacher-forced training strategy. The trained model parameters are saved to the prediction model library. The trajectory feature vector is input into the trained trajectory prediction model. The encoder extracts the temporal feature representation of the current trajectory. The decoder generates a future trajectory sequence based on the temporal feature representation. The future trajectory sequence is subjected to endpoint clustering analysis to obtain a set of candidate destinations. The probability score of each candidate destination is calculated. The coordinates with the highest probability score are selected as the predicted destination coordinates. The path distance is calculated based on the vehicle's current position and the predicted destination coordinates. The estimated arrival time is calculated based on the path distance and the current driving speed.

4. The intelligent recommendation method based on driving behavior analysis according to claim 1, characterized in that, The process of constructing a user interest modeling network involves inputting the user profile features, the behavioral feature vector, and the scene feature vector into the network, and then weighting and fusing the features based on an attention mechanism to obtain a fused feature vector. This includes: A multi-head self-attention module is constructed as a feature extraction layer. The user profile features, the behavior feature vector, and the scene feature vector are linearly transformed to obtain a query matrix, a key matrix, and a value matrix, respectively. The query matrix and the key matrix are multiplied to obtain an attention score matrix. The attention score matrix is ​​normalized by softmax to obtain an attention weight matrix. The attention weight matrix and the value matrix are multiplied to obtain a weighted feature matrix. The outputs of multiple attention heads are concatenated to obtain multi-head attention features. A feature fusion layer is constructed, and a residual connection structure is used to enhance the multi-head attention features. The enhanced features are standardized through a layer normalization module. The standardized features are input into a feedforward neural network for nonlinear transformation. The transformed features are then subjected to a second layer normalization to obtain a fused feature vector. The fused feature vector is randomly deactivated using a dropout mechanism. The processed features are then input into the next layer of the user interest modeling network.

5. The intelligent recommendation method based on driving behavior analysis according to claim 1, characterized in that, The process involves inputting the fused feature vector into a fully connected layer to generate a user interest vector, collecting user click data, dwell time data, and visit record data for recommended content, and constructing an online learning module based on the click data, dwell time data, and visit record data. This online learning module dynamically optimizes the parameters of the user interest modeling network, including: A three-layer fully connected network is constructed as the interest modeling layer. The first fully connected layer performs dimensionality reduction mapping on the fused feature vector to obtain an intermediate feature representation. The ReLU activation function is used to introduce a nonlinear transformation. The second fully connected layer learns the user's preference distribution on different service types based on the intermediate feature representation. The third fully connected layer transforms the preference distribution into a fixed-dimensional user interest vector. The batch normalization method is used to perform numerical stabilization processing on the user interest vector. An online learning module based on gradient boosting trees is constructed. User click operation data, page dwell time data, and actual visit record data are collected from the vehicle terminal. The click operation data, the dwell time data, and the visit record data are used to construct sample labels. The latest training sample set is maintained using a sliding time window method. The importance score of the model parameters is calculated based on the training sample set. The parameter update step size is adjusted according to the importance score. The weights of the user interest modeling network are updated online using the stochastic gradient descent method.

6. The intelligent recommendation method based on driving behavior analysis according to claim 1, characterized in that, The process of constructing a search range for points of interest based on the destination coordinates, retrieving candidate location information within the search range from the point of interest database, calculating the distance score between the candidate locations and the destination coordinates, and calculating the interest matching score of the candidate locations based on the user interest vector includes: A circular search range is constructed based on the destination coordinates. The destination coordinates are used as the center of the circle. The search radius is determined according to the vehicle speed. The geospatial indexing method is used to retrieve the location information within the search range from the point of interest database. The retrieved location information is classified and filtered to obtain a candidate location set. The geographic coordinate information, business hours information, and service type information of each location in the candidate location set are extracted. A distance scoring model and an interest matching model are constructed. The Manhattan distance is used to calculate the straight-line distance between the candidate location and the destination coordinates. The actual driving path length is calculated by combining road network data. The straight-line distance and the path length are weighted to obtain the distance score. The service type information of the candidate location is converted into a location feature vector. The cosine similarity between the location feature vector and the user interest vector is calculated to obtain the interest matching score.

7. The intelligent recommendation method based on driving behavior analysis according to claim 1, characterized in that, The process of weightedly fusing the distance score and the interest matching score to obtain a recommendation priority score, selecting the location with the highest recommendation priority score to generate a recommendation result, determining the push timing based on the arrival time, and sending the recommendation result to the vehicle terminal display interface includes: A multi-objective optimization model is constructed to calculate the recommendation priority. The distance score and the interest matching score are numerically normalized based on the min-max normalization method. An adaptive weight allocation strategy is used to calculate the fusion weight coefficient. The distance score and the interest matching score are weighted and summed according to the weight coefficient to obtain the recommendation priority score. The candidate venue set is sorted in descending order according to the recommendation priority score. The venue identifier, venue name, venue location, and service type with the highest ranking are selected to generate the recommendation result. Based on the arrival time, a push rule model is constructed, the remaining travel time from the vehicle's current location to the recommended location is calculated, the push priority interval is divided based on the remaining travel time, the optimal push time is determined by combining the location's business hours information, the recommendation result is encapsulated into a message packet, and the message packet is sent to the display control module through the vehicle terminal's message queue service. The display control module displays the recommended content in a designated area of ​​the terminal interface at the optimal push time.

8. An intelligent recommendation device based on driving behavior analysis, characterized in that, The device includes: The trajectory prediction module is used to collect driver registration information data, vehicle camera video data, vehicle driving data, and scene perception data. It extracts user profile features from the registration information data, performs driving behavior recognition on the video data to obtain a behavior feature vector, extracts location information, driving duration, and driving trajectory from the vehicle driving data to construct a trajectory feature vector, extracts time period identifiers, weather conditions, and road conditions from the scene perception data to construct a scene feature vector, constructs a trajectory prediction model based on a recurrent neural network, and inputs the trajectory feature vector into the trajectory prediction model to predict the destination coordinates and arrival time. An interest modeling module is used to construct a user interest modeling network. The user profile features, behavioral feature vectors, and scene feature vectors are input into the user interest modeling network. Based on an attention mechanism, the features are weighted and fused to obtain a fused feature vector. This fused feature vector is then input into a fully connected layer to generate a user interest vector. User click operation data, dwell time data, and visit record data for recommended content are collected. Based on the click operation data, dwell time data, and visit record data, an online learning module is constructed. This online learning module dynamically optimizes the parameters of the user interest modeling network. The intelligent recommendation module is used to construct a search range of points of interest based on the destination coordinates, retrieve candidate location information within the search range from the point of interest database, calculate the distance score between the candidate location and the destination coordinates, calculate the interest matching score of the candidate location based on the user's interest vector, weight and fuse the distance score and the interest matching score to obtain a recommendation priority score, select the location with the highest recommendation priority score to generate a recommendation result, determine the push timing based on the arrival time, and send the recommendation result to the vehicle terminal display interface.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the intelligent recommendation method based on driving behavior analysis as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the intelligent recommendation method based on driving behavior analysis as described in any one of claims 1 to 7.

Images