Intelligent screening and management and control method for high-risk vehicles

Abstract

The invention provides a high-risk vehicle intelligent screening and management and control method, and relates to the technical field of intelligent traffic management, and the method comprises the steps: obtaining the multi-source data of a vehicle in real time, the multi-source data comprising vehicle basic information, a driving track, a violation record and an accident record; performing data cleaning and fusion processing on the multi-source data, and constructing a vehicle dynamic portrait; based on the vehicle dynamic portrait, performing multi-dimensional feature extraction and risk assessment by using a deep learning model, and generating a vehicle risk score and a risk level; based on the risk level and the user interaction operation, intelligent screening is carried out, and a targeted management and control suggestion is generated; and based on user feedback data, optimizing the deep learning model and the screening strategy through an online learning mechanism.

Description

Technical Field

[0001] This application belongs to the field of intelligent traffic management technology, specifically relating to a method for intelligent screening and control of high-risk vehicles. Background Technology

[0002] In the field of public security traffic management, the core challenge of ensuring road traffic safety and public safety is how to accurately and efficiently identify high-risk vehicles (such as vehicles suspected of violations, accidents, or suspicious behavior) from massive amounts of vehicle information and effectively control them.

[0003] Currently, traditional vehicle screening systems suffer from the following limitations: First, these systems typically rely on manual operation, requiring users to manually combine multiple conditions (such as license plate, vehicle type, and violation type) for filtering. Due to the heterogeneous data sources (including basic information, driving trajectories, and violation records), this manual screening process is extremely cumbersome, time-consuming, and labor-intensive, failing to meet the real-time response requirements of dynamic risk scenarios. Second, existing systems lack the ability to quantitatively assess vehicle risks and cannot automatically learn and identify risk patterns from multi-dimensional characteristics such as driving behavior, violation frequency, and special markings. This leads to a heavy reliance on human experience in control decisions, making accuracy and consistency difficult to guarantee. Finally, the screening results are severely disconnected from subsequent control measures. The system cannot automatically generate precise control recommendations (such as monitoring periods and checkpoint locations) based on specific vehicle risk characteristics, resulting in unreasonable allocation of police resources and limited early warning and interception effectiveness. Summary of the Invention

[0004] This application provides a method for intelligent screening and control of high-risk vehicles to solve one of the aforementioned technical problems.

[0005] The technical solution adopted in this application is as follows: This application provides a method for intelligent screening and control of high-risk vehicles, including: Real-time acquisition of multi-source vehicle data, including basic vehicle information, driving trajectory, traffic violation records, and accident records; The multi-source data is cleaned and fused to construct a dynamic vehicle profile; Based on the vehicle dynamic profile, a deep learning model is used to extract multi-dimensional features and assess risks, generating vehicle risk scores and risk levels. Based on the risk level and user interaction, intelligent screening is performed, and targeted control suggestions are generated. Based on user feedback data, the deep learning model and selection strategy are optimized through an online learning mechanism.

[0006] According to one embodiment of this application, the real-time acquisition of multi-source vehicle data includes: By connecting with the public security big data platform, basic vehicle information, driving trajectory, traffic violation records, and accident records are collected in real time.

[0007] According to one embodiment of this application, the step of performing data cleaning and fusion processing on the multi-source data to construct a vehicle dynamic profile includes: Data cleaning algorithms are used to remove outlier data and noise, and missing data is filled in. Based on spatiotemporal correlation rules, multi-source data are fused into a unified vehicle dynamic profile, which includes static attributes and dynamic behavioral features.

[0008] According to one embodiment of this application, the multi-dimensional feature extraction and risk assessment using a deep learning model includes: Temporal features, including movement patterns and abnormal dwell points, are extracted using an attention-based Transformer model or a temporal convolutional network. Statistical analysis methods were used to calculate the frequency of violations and the correlation between accidents as statistical characteristics. Convolutional neural networks were used to analyze special markings in vehicle images as visual features. A reinforcement learning model is used to identify high-risk driving behaviors as behavioral features. Based on the extracted multi-dimensional features, a vehicle risk score is generated through a risk assessment model, and the risk level is dynamically classified based on an adaptive threshold.

[0009] According to one embodiment of this application, the step of intelligently screening based on the risk level and user interaction operations, and generating targeted control suggestions, includes: It provides a visual interactive interface that allows users to combine filter criteria by dragging or checking. Based on the importance analysis of the multi-dimensional features, high-frequency or high-weight screening conditions are dynamically recommended. Based on the screening results and risk characteristics, the system combines a rule engine with a machine learning model to generate control recommendations, including key monitoring periods, mandatory checkpoint locations, and warning frequencies.

[0010] According to one embodiment of this application, optimizing the deep learning model and screening strategy based on user feedback data through an online learning mechanism includes: Collect user feedback data on screening results and control recommendations, including false alarms and missed alarms; The parameters of the deep learning model are updated using incremental learning techniques, and the weights and control rules of the screening criteria are adjusted.

[0011] According to one embodiment of this application, the method further includes: Based on the aforementioned risk levels and control recommendations, differentiated vehicle control strategies should be developed.

[0012] A second aspect of this application provides a high-risk vehicle intelligent screening and control device, the device comprising: The data acquisition module is used to acquire multi-source vehicle data in real time. The data processing module is used to clean and fuse the multi-source data to construct a dynamic vehicle profile. The risk assessment module is used to perform multi-dimensional feature extraction and risk assessment based on the vehicle dynamic profile, and generate a vehicle risk score and risk level. The screening and control module is used to perform intelligent screening based on the risk level and user interaction operations, and generate targeted control suggestions. The optimization module is used to optimize the deep learning model and screening strategy based on user feedback data through an online learning mechanism.

[0013] A third aspect of this application provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps described in the method.

[0014] A fourth aspect of this application provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method as described.

[0015] Due to the adoption of the above technical solution, the beneficial effects achieved by this application are as follows: This application employs technologies such as real-time acquisition of multi-source data to construct dynamic vehicle profiles and intelligent screening based on risk levels. The system transforms the original process of manually querying, comparing, and combining conditions across different heterogeneous data sources into automated querying and sorting of pre-constructed, fused, and semantically rich dynamic profiles. This significantly reduces user operation time from an average of 10 minutes to less than 2 minutes, improving screening efficiency by over 450%, and truly enabling real-time response to dynamic risk scenarios.

[0016] By introducing the core approach of using deep learning models for multi-dimensional feature extraction and risk assessment, the system can automatically uncover complex risk patterns that are difficult for the human eye to detect from vehicle dynamic profiles (such as abnormal temporal driving patterns, behavioral patterns behind frequent violations, and subtle risk markers in images). This overcomes the excessive reliance on human experience in traditional methods, transforming risk assessment into a quantifiable and reproducible computational process. As a result, the accuracy rate of high-risk vehicle identification is increased to over 95%, while the false alarm rate is reduced to below 3%.

[0017] Based on risk levels and user interaction, the system generates targeted control recommendations, ensuring a direct link between control measures and risk assessment results. The system can automatically derive specific and actionable recommendations, such as "strengthening nighttime monitoring of specific checkpoints" and "deploying controls during peak traffic periods," based on risk characteristics output by the model (e.g., vehicles frequently appearing at night or in specific areas). This improves the targeting of control recommendations by 350%, optimizes police resource allocation by 40%, and significantly enhances the effectiveness of control operations.

[0018] By employing a closed-loop design—based on user feedback data and an online learning mechanism to optimize deep learning models and screening strategies—the system can continuously learn from false positives and false negatives in real-world applications, dynamically adjusting model parameters and screening rule weights. This technology enables the system to quickly adapt to new criminal methods or risk patterns, improving risk identification accuracy by over 15% in long-term operation, demonstrating a level of long-term viability and intelligence unavailable in traditional static systems. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a high-risk vehicle intelligent screening and control method provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0020] Figure label: 810, Processor; 820, Communication interface; 830, Memory; 840, Communication bus. Detailed Implementation

[0021] To more clearly illustrate the overall concept of this application, a detailed explanation is provided below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of this application is not limited to the specific embodiments disclosed below. It should be noted that, unless otherwise specified, the embodiments of this application and the features thereof can be combined with each other.

[0023] In this application, unless otherwise expressly specified and limited, the "above" or "below" of the second feature can mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. In the description of this specification, references to terms such as "an embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples.

[0024] Example 1 like Figure 1 As shown, a method for intelligent screening and control of high-risk vehicles includes: Real-time acquisition of multi-source vehicle data, including basic vehicle information, driving trajectory, traffic violation records, and accident records.

[0025] As described above, in the high-risk vehicle intelligent screening and control system, "real-time acquisition of multi-source vehicle data" is a fundamental step in constructing a dynamic vehicle profile. Its core lies in dynamically and continuously collecting various vehicle-related information from heterogeneous data sources through efficient data interfaces and transmission mechanisms, ensuring the system can respond promptly to changes in risk. Specifically, the multi-source data includes basic vehicle information (such as license plate number, vehicle model, color, registration date, and other static attributes), driving trajectory (spatiotemporal movement paths recorded by GPS, base station positioning, or roadside sensing devices), violation records (historical speeding, running red lights, and other traffic violations), and accident records (historical events involving collisions, injuries, etc.). This data not only covers the inherent characteristics of vehicles but also dynamically reflects their behavioral patterns, providing comprehensive input for subsequent risk assessment. For example, in practical applications, the system can access basic vehicle information and violation records in real time through a data interface with the public security traffic management platform, while continuously capturing driving trajectories using IoT devices (such as smart cameras or in-vehicle terminals) and extracting updated accident records from the accident report database, thus forming a real-time updated data stream.

[0026] It should be noted that, in specific implementation scenarios, the data acquisition can be expanded to include environmental data (such as real-time weather and road condition information) or social data (such as abnormal behavior reports associated with vehicles) based on the above solutions. These expanded data sources can more precisely characterize risk scenarios. At the same time, the acquisition method can also adopt a collaborative architecture of edge computing and cloud platform, and improve real-time performance and reduce transmission load by performing preliminary data filtering and compression on roadside devices or vehicle units.

[0027] The multi-source data is cleaned and fused to construct a dynamic vehicle profile.

[0028] As described above, in the high-risk vehicle intelligent screening and control system, "data cleaning and fusion processing of multi-source data to construct a dynamic vehicle profile" is a key transformation step from raw data to understandable risk characteristics. Specifically, data cleaning includes identifying and correcting outliers, missing values, and noisy data from multiple sources. For example, setting reasonable numerical thresholds can filter obviously erroneous GPS coordinates, or time-series-based interpolation methods can be used to complete missing trajectory points. Data fusion involves associating and integrating data from different sources and formats within a unified spatiotemporal framework. For instance, using unique vehicle identifiers (such as license plate numbers) to link heterogeneous data such as basic information, trajectory points, and traffic violations, and aligning them based on timestamps to form a coherent sequence of vehicle behavior. Constructing a dynamic vehicle profile involves extracting and organizing a set of features that comprehensively reflect the vehicle's state and behavior based on the cleaned and fused data. These features include not only static attributes (such as vehicle type and color) but, more importantly, dynamically updated behavioral characteristics (such as recent average driving speed, high-frequency activity areas, and violation patterns). For example, after cleaning and removing drift points from trajectory data, the system spatiotemporally correlates continuous trajectory points with traffic violation records. When it is found that a vehicle frequently exhibits rapid acceleration behavior in a specific area and is accompanied by multiple speeding records, the system can mark its dynamic profile with the feature tag "having a high-risk driving behavior pattern".

[0029] It should be noted that, in specific implementation scenarios, more complex association rules can be introduced in the data fusion stage based on the above solutions. For example, correlation analysis based on vehicle model and driving behavior, or mining spatiotemporal association patterns between vehicles through graph computing methods. In terms of profile construction, behavior prediction models can be introduced to predict the future activity patterns and risk trends of vehicles based on historical data, or external environmental factors (such as weather and road conditions) can be incorporated into the profile system as contextual features.

[0030] Based on the vehicle dynamic profile, a deep learning model is used to extract multi-dimensional features and assess risks, generating vehicle risk scores and risk levels.

[0031] As described above, in the high-risk vehicle intelligent screening and control system, "multi-dimensional feature extraction and risk assessment based on vehicle dynamic profiles using deep learning models" is the core analytical step in realizing intelligent decision-making from basic data. Specifically, this process uses a specially designed deep learning architecture to deeply mine the constructed vehicle dynamic profiles, extracting discriminative risk features from multiple dimensions such as temporal behavior, statistical regularities, visual features, and driving patterns. For example, a temporal model based on an attention mechanism is used to analyze the movement patterns and abnormal stopping patterns in the vehicle trajectory; convolutional neural networks are used to identify abnormal visual features such as occlusion and dirt in vehicle images; statistical learning methods are combined to calculate quantitative indicators such as violation frequency and accident correlation; finally, these multi-dimensional features are weighted and integrated through a feature fusion layer and input into the risk assessment module. Based on these fused features, the risk assessment module outputs a quantified vehicle risk score through a preset deep learning classifier or regressor, and maps this score to specific risk levels such as high, medium, and low according to a dynamically adjusted threshold range. For example, by analyzing the dynamic profile of a vehicle, the system discovers that it has the temporal characteristics of "frequently entering and exiting high-crime areas at night", the statistical characteristics of "8 speeding violations within a year", the visual characteristics of "continuous occlusion of the face of the front-seat occupants", and the behavioral characteristics of "frequent sudden braking and sharp turns". The deep learning model will integrate and calculate these features, and finally generate a high-risk score of 0.92 and classify it as a high-risk level.

[0032] It should be noted that, in specific implementation scenarios, based on the above solutions, a multi-task learning mechanism can be introduced in the feature extraction stage, enabling the model to simultaneously learn risk prediction and related auxiliary tasks (such as vehicle type recognition and behavior pattern classification), thereby improving feature representation capabilities through the correlation between tasks; in the risk assessment stage, an adaptive threshold adjustment strategy can be adopted to dynamically adjust the risk level classification criteria based on contextual factors such as region, time period, and vehicle type; in addition, an uncertainty estimation module can be introduced to provide a confidence index for each risk score, helping users understand the reliability of the model's decisions.

[0033] Based on the risk level and user interaction, intelligent screening is performed, and targeted control suggestions are generated.

[0034] As described above, in the high-risk vehicle intelligent screening and control system, "intelligent screening based on risk level and user interaction, and generation of targeted control suggestions" is a key decision-making step in realizing the transition from risk analysis to practical control application. Specifically, the intelligent screening process combines the risk level output by the deep learning model with the user's specific operational intentions through a visual interactive interface. Users can dynamically combine screening conditions through interactive methods such as dragging, checking, or customizing thresholds. The system also proactively recommends high-frequency or high-weight screening conditions based on the results of feature importance analysis in the risk assessment module to improve operational efficiency. When generating targeted control suggestions, the system uses a combination of a rule engine and a machine learning model to automatically derive specific control measures based on the screened vehicle set and its risk characteristics, including but not limited to key monitoring periods, mandatory checkpoint locations, warning frequencies, and police deployment suggestions. For example, when a user selects the filter criteria of "high risk level" and "recent nighttime driving record" on the interface, the system not only presents a list of vehicles that meet the criteria in real time, but also automatically generates control suggestions such as "strengthen checkpoint inspections at XX intersection from 9 pm to 2 am" based on the dynamic profile characteristics of these vehicles - such as their frequent appearance in urban-rural fringe areas during specific time periods. At the same time, it recommends collaborative deployment plans in adjacent areas.

[0035] It should be noted that, in specific implementation scenarios, based on the above solutions, a personalized recommendation mechanism based on user operation history can be introduced in the intelligent screening stage. By learning the operation preferences and success cases of different users, the accuracy of the recommendation criteria can be optimized. In terms of generating control suggestions, a resource optimization algorithm can be introduced to comprehensively consider multiple constraints such as police force allocation, checkpoint pass rate, and early warning response time to generate the optimal control resource allocation plan. In addition, a control effect evaluation and feedback mechanism can be established to re-input the execution effect data of historical control measures into the system, forming a complete closed loop from decision-making to evaluation to optimization.

[0036] Based on user feedback data, the deep learning model and selection strategy are optimized through an online learning mechanism.

[0037] As described above, in the high-risk vehicle intelligent screening and control system, "optimizing the deep learning model and screening strategy based on user feedback data through an online learning mechanism" is the core adaptive link for achieving continuous system evolution and self-improvement. Specifically, this process collects user feedback on the system's output results in practical applications—including confirmation and correction of screening results, evaluation of the effectiveness of control recommendations, and annotation of false alarms and missed alarms—and uses this feedback data as a monitoring signal to dynamically adjust the parameters of the deep learning model and the rule weights of the screening strategy through online learning technology without interrupting the normal operation of the system. For example, when the system misclassifies a normally operating ride-hailing vehicle as a high-risk vehicle, the user can mark the case as a "false alarm" on the interface and add the explanation that "the vehicle has a ride-hailing compliance mark." The system will then adjust the parameters of the visual feature recognition module and the risk assessment module through incremental learning algorithms, reducing the weight of the single feature "nighttime activity" while enhancing the recognition ability of the "compliance mark" feature. Similarly, if the user frequently uses the combination of "first time entering the city + driving without a license plate" as the filtering condition and the practical results are good, the system will automatically increase the priority of this combination of conditions in the recommendation list.

[0038] It should be noted that, in specific implementation scenarios, reinforcement learning mechanisms can be introduced on the basis of the above solutions. User feedback behaviors (such as adopting suggestions and ignoring warnings) can be regarded as environmental reward signals, automatically optimizing the entire decision-making chain from risk assessment to the generation of control suggestions. A feedback weight allocation mechanism based on time decay can be established, enabling the system to better balance the relationship between historical experience and emerging risk patterns. Furthermore, a multi-user feedback fusion algorithm can be developed to integrate practical feedback from users in different jurisdictions and with different experience levels, while protecting the privacy of each operator, to form a more generalizable optimization strategy.

[0039] According to one embodiment of this application, the real-time acquisition of multi-source vehicle data includes: By connecting with the public security big data platform, basic vehicle information, driving trajectory, traffic violation records, and accident records are collected in real time.

[0040] As mentioned above, the real-time acquisition of multi-source vehicle data specifically refers to the system establishing a secure and stable data connection with the public security department's big data platform through a pre-established data communication interface. This interface is typically based on a standardized data exchange protocol and can continuously pull or receive the latest data proactively pushed by the platform in a streaming or timed polling manner.

[0041] The collected basic vehicle information mainly includes static attribute data registered and issued by the vehicle management department, such as vehicle license plate, vehicle identification number, vehicle type, brand and model, body color, registration date, and basic information of the vehicle owner.

[0042] The driving trajectory data refers to vehicle movement information recorded by road checkpoint monitoring systems, GPS positioning devices or other location sensing devices, which usually includes elements such as timestamps, latitude and longitude coordinates, driving speed, direction and road segments or checkpoint numbers passed through.

[0043] The traffic violation record data comes from the traffic violation processing system and records the vehicle's unprocessed traffic violations, such as speeding, disobeying traffic signals, and illegal parking. Each record usually includes the time, location, specific behavior code, and penalty information of the violation.

[0044] The accident record data is obtained from the traffic accident handling system, including information on road traffic accidents involving the vehicle, such as the time and location of the accident, the type of accident, the division of responsibility, and the resulting damage.

[0045] According to one embodiment of this application, the step of performing data cleaning and fusion processing on the multi-source data to construct a vehicle dynamic profile includes: Data cleaning algorithms are used to remove outlier data and noise, and missing data is filled in. Based on spatiotemporal correlation rules, multi-source data are fused into a unified vehicle dynamic profile, which includes static attributes and dynamic behavioral features.

[0046] As mentioned above, data cleaning algorithms are used to preprocess the raw multi-source data. Removing outlier data involves identifying and filtering data values ​​that are clearly illogical or beyond reasonable limits, such as deleting trajectory points with speed values ​​exceeding physical limits or correcting data records with obviously incorrect timestamps. Removing noisy data involves smoothing data fluctuations caused by equipment errors or transient interference, such as filtering GPS trajectory points to eliminate positioning drift. Completing missing data involves using interpolation or prediction methods to fill in incomplete data sequences, such as completing missing location information based on the spatiotemporal relationship of preceding and following trajectory points, or inferring missing attribute values ​​based on historical behavior patterns.

[0047] Based on spatiotemporal association rules, cleaned multi-source data is fused into a unified vehicle dynamic profile. The fusion process uses the vehicle's unique identifier (such as license plate number) as the primary key to link and integrate information scattered across different data sources. Specifically, driving trajectory points are spatiotemporally aligned and matched with traffic violation records and accident records in chronological order, ensuring accurate location of each behavioral event in the spatiotemporal dimension. Static attributes refer to relatively stable vehicle characteristics, including basic information such as vehicle model, color, and registration date; dynamic behavioral characteristics refer to continuously updated behavioral pattern data over time, including real-time location, speed changes, high-frequency activity areas, history of traffic violations, accident correlation, and driving habit characteristics. The resulting vehicle dynamic profile constitutes a digital vehicle archive containing a complete spatiotemporal behavioral chain.

[0048] According to one embodiment of this application, the multi-dimensional feature extraction and risk assessment using a deep learning model includes: Temporal features, including movement patterns and abnormal dwell points, are extracted using an attention-based Transformer model or a temporal convolutional network. Statistical analysis methods were used to calculate the frequency of violations and the correlation between accidents as statistical characteristics. Convolutional neural networks were used to analyze special markings in vehicle images as visual features. A reinforcement learning model is used to identify high-risk driving behaviors as behavioral features. Based on the extracted multi-dimensional features, a vehicle risk score is generated through a risk assessment model, and the risk level is dynamically classified based on an adaptive threshold.

[0049] As mentioned above, in terms of temporal feature extraction, attention-based Transformer models or temporal convolutional networks are used to perform in-depth analysis of vehicle trajectory data. These models can capture the long-term dependencies and periodic patterns of vehicle movement, and identify regular movement patterns and abnormal behavioral characteristics, including but not limited to abnormal stopping points, deviation from the regular route, and frequent travel to and from specific areas.

[0050] In terms of statistical feature calculation, statistical analysis methods are used to quantify vehicle violation and accident records. Specifically, this includes calculating the frequency of violations per unit time, the distribution characteristics of violation types, the number of accidents involved, and their severity. Furthermore, correlation algorithms are used to analyze the correlation between different violations and accident occurrences, forming quantifiable statistical feature indicators.

[0051] In terms of visual feature analysis, a convolutional neural network is used to extract deep features from vehicle capture images. This network can automatically identify special identifying features in the images, including visual risk features such as license plate obstruction, abnormal window tinting, obvious vehicle modifications, and occupant facial occupancy.

[0052] In terms of behavioral feature recognition, a reinforcement learning model is used to analyze continuous driving behavior sequences. This model identifies high-risk driving behavior patterns by establishing a correlation between driving behavior and risk outcomes, including dangerous operation characteristics such as frequent rapid acceleration and deceleration, continuous lane changes, and speeding.

[0053] The aforementioned multi-dimensional features are input into the risk assessment model for fusion calculation. This model outputs a quantified vehicle risk score by weighted integration of the influencing factors of each feature dimension. Based on real-time updated risk distribution, the system uses adaptive thresholds to dynamically classify risk levels, ensuring that the risk assessment results accurately reflect the actual needs of the current road safety situation.

[0054] According to one embodiment of this application, the step of intelligently screening based on the risk level and user interaction operations, and generating targeted control suggestions, includes: It provides a visual interactive interface that allows users to combine filter criteria by dragging or checking. Based on the importance analysis of the multi-dimensional features, high-frequency or high-weight screening conditions are dynamically recommended. Based on the screening results and risk characteristics, the system combines a rule engine with a machine learning model to generate control recommendations, including key monitoring periods, mandatory checkpoint locations, and warning frequencies.

[0055] As mentioned above, the system provides a visual interactive interface, allowing users to flexibly combine filter criteria by dragging and dropping filter components or selecting preset conditions to form filtering logic. The interface displays real-time statistics on the number of filtering results and risk distribution, and users can dynamically adjust the combination of filtering conditions and parameter thresholds through interactive operations.

[0056] Based on the importance analysis results of each feature dimension in the risk assessment model, the system dynamically recommends screening conditions that are frequently used or have high weighting. The recommendation logic comprehensively considers the contribution of features to risk assessment, the user's historical usage frequency, and the needs of the current management scenario, presenting the recommended conditions to the user in the interactive interface through highlighting or intelligent sorting.

[0057] Based on vehicle risk characteristics identified in the filtered result set, the system generates targeted control recommendations through a collaborative mechanism between a rule engine and a machine learning model. The rule engine executes predefined business rules, while the machine learning model performs predictive optimization based on historical control effect data, jointly outputting specific control plans including key monitoring periods, mandatory checkpoint locations, and warning frequencies. Specifically, the determination of key monitoring periods comprehensively considers vehicle activity patterns and risk time distribution characteristics; the recommendation of mandatory checkpoint locations is based on vehicle trajectory hotspot analysis and road network structure characteristics; and the warning frequency is dynamically adjusted according to risk levels, forming a tiered and categorized warning mechanism.

[0058] According to one embodiment of this application, optimizing the deep learning model and screening strategy based on user feedback data through an online learning mechanism includes: Collect user feedback data on screening results and control recommendations, including false alarms and missed alarms; The parameters of the deep learning model are updated using incremental learning techniques, and the weights and control rules of the screening criteria are adjusted.

[0059] As mentioned above, the system continuously collects feedback data generated by users during actual use. This data includes user corrections to the screening results, such as marking high-risk vehicles identified by the system as false alarms or adding unidentified risk vehicles as missed reports; it also includes user evaluations of the effectiveness of control recommendations, such as data on the actual control effectiveness after the recommendations were adopted.

[0060] The system employs incremental learning technology to update the deep learning model online. This technology utilizes continuously collected feedback data to dynamically adjust model parameters using a gradient descent algorithm without retraining on the entire dataset. Specifically, the system uses user-confirmed false positives and false negatives as new training samples, gradually updating the weight parameters of each layer of the neural network with a small learning rate. This allows the model to adapt to new risk patterns without forgetting existing knowledge.

[0061] Based on the statistical analysis of feedback data, the system dynamically adjusts the weight allocation of filtering conditions and the parameter settings of control rules. For filtering conditions that are frequently used by users and have good results, the system automatically increases their priority in the recommendation list; for filtering conditions that frequently cause false positives, their weight is reduced accordingly. Regarding control rules, the system continuously optimizes the threshold settings and logical judgment conditions in the rule base based on actual execution results, forming a closed-loop optimization mechanism for continuous improvement.

[0062] According to one embodiment of this application, the method further includes: Based on the aforementioned risk levels and control recommendations, differentiated vehicle control strategies should be developed.

[0063] As described above, the system generates tiered and categorized control implementation plans based on the vehicle risk level output by the risk assessment module and the specific measures provided by the control suggestion generation module. For high-risk vehicles, the system generates key control strategies, including mandatory measures such as implementing full-track tracking, activating real-time early warning mechanisms, and deploying key checkpoints for interception; for medium-risk vehicles, restrictive measures such as regular spot checks and regional traffic restrictions are implemented; and for low-risk vehicles, routine monitoring is maintained.

[0064] Meanwhile, the system dynamically adjusts the allocation of control resources based on the specific content of the control recommendations. It optimizes police deployment and equipment scheduling based on key monitoring periods and mandatory checkpoint locations outlined in the recommendations; and establishes a tiered response mechanism based on warning frequency settings to ensure that control measures are commensurate with the level of risk. The system also establishes a strategy implementation effectiveness evaluation mechanism, continuously optimizing the parameter settings and execution standards of differentiated control strategies by constantly monitoring changes in vehicle behavior and control effectiveness.

[0065] A second aspect of this application provides a high-risk vehicle intelligent screening and control device, the device comprising: The data acquisition module is used to acquire multi-source vehicle data in real time. The data processing module is used to clean and fuse the multi-source data to construct a dynamic vehicle profile. The risk assessment module is used to perform multi-dimensional feature extraction and risk assessment based on the vehicle dynamic profile, and generate a vehicle risk score and risk level. The screening and control module is used to perform intelligent screening based on the risk level and user interaction operations, and generate targeted control suggestions. The optimization module is used to optimize the deep learning model and screening strategy based on user feedback data through an online learning mechanism.

[0066] A third aspect of 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 method described in any of the first aspects above.

[0067] Figure 2 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 2 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call logical instructions in the memory 830 to execute the method in any of the embodiments of the first aspect described above, the method including: Real-time acquisition of multi-source vehicle data, including basic vehicle information, driving trajectory, traffic violation records, and accident records; The multi-source data is cleaned and fused to construct a dynamic vehicle profile; Based on the vehicle dynamic profile, a deep learning model is used to extract multi-dimensional features and assess risks, generating vehicle risk scores and risk levels. Based on the risk level and user interaction, intelligent screening is performed, and targeted control suggestions are generated. Based on user feedback data, the deep learning model and selection strategy are optimized through an online learning mechanism.

[0068] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0069] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program, the computer program being able to be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer being able to perform the methods provided by the above methods, the method comprising: Real-time acquisition of multi-source vehicle data, including basic vehicle information, driving trajectory, traffic violation records, and accident records; The multi-source data is cleaned and fused to construct a dynamic vehicle profile; Based on the vehicle dynamic profile, a deep learning model is used to extract multi-dimensional features and assess risks, generating vehicle risk scores and risk levels. Based on the risk level and user interaction, intelligent screening is performed, and targeted control suggestions are generated. Based on user feedback data, the deep learning model and selection strategy are optimized through an online learning mechanism.

[0070] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods provided by the above methods, the method comprising: Real-time acquisition of multi-source vehicle data, including basic vehicle information, driving trajectory, traffic violation records, and accident records; The multi-source data is cleaned and fused to construct a dynamic vehicle profile; Based on the vehicle dynamic profile, a deep learning model is used to extract multi-dimensional features and assess risks, generating vehicle risk scores and risk levels. Based on the risk level and user interaction, intelligent screening is performed, and targeted control suggestions are generated. Based on user feedback data, the deep learning model and selection strategy are optimized through an online learning mechanism.

[0071] Example 2 1. System Overview This embodiment achieves accurate assessment and intelligent management of vehicle risks through multi-source data fusion, deep learning models, and real-time feedback mechanisms. The overall system architecture includes a data acquisition layer, a data processing layer, an intelligent analysis layer, an interactive application layer, and an optimization feedback layer. These layers work together to form a closed-loop, self-optimizing technical system.

[0072] 2. Detailed Implementation Method 2.1 Real-time multi-source data acquisition The system connects to the public security big data platform via a secure data interface to collect multi-source vehicle data in real time. Specifically: Basic vehicle information includes static attributes such as license plate number, vehicle model, color, and registration date, which are obtained from the vehicle management department's database.

[0073] Driving trajectory data: Recorded by GPS positioning devices or roadside sensing devices, including timestamps, latitude and longitude coordinates, speed and direction information.

[0074] Traffic violation record data: extracted from the traffic violation processing system, including violation type, time, location, and penalty status.

[0075] Accident record data: obtained from the traffic accident handling system, including the time, location, type, and extent of damage of the accident.

[0076] Data acquisition employs streaming processing to ensure real-time performance. For example, the system updates trajectory data every 5 seconds, and violation and accident records are synchronized to the system within 1 minute of their occurrence.

[0077] 2.2 Data Cleaning and Fusion to Construct Dynamic Vehicle Profiles The collected multi-source data is cleaned and fused: Data cleaning: outliers are identified using statistical anomaly detection algorithms (such as the Z-score algorithm), such as filtering out trajectory points with speeds exceeding 200 km / h; missing data is filled in using linear interpolation or prediction methods based on historical patterns, such as filling in missing coordinates based on the time and positional relationship between previous and subsequent trajectory points.

[0078] Data fusion: Based on spatiotemporal association rules, using the vehicle's unique identifier (such as license plate number) as the key, multi-source data is integrated into a unified dynamic vehicle profile. The fusion process uses a spatiotemporal alignment algorithm to match driving trajectory points with traffic violation records and accident records. For example, trajectory points and traffic violation events are associated as the same event if the time difference is less than 30 seconds and the distance difference is less than 50 meters.

[0079] Vehicle dynamic profile: includes: Static attributes: unchanging features such as vehicle model and color.

[0080] Dynamic behavioral characteristics: real-time location, average speed, high-frequency activity areas (identified by clustering algorithms such as DBSCAN), violation frequency (number of violations per unit time), and accident correlation (the probability of a vehicle being involved in an accident is calculated based on historical accidents).

[0081] 2.3 Multi-dimensional Feature Extraction and Risk Assessment Deep learning models are used to extract features and assess risks in dynamic vehicle profiles. Temporal Feature Extraction: A Transformer model based on an attention mechanism is used to analyze driving trajectory data. The model calculates the dependency weights between trajectory points through a self-attention layer, using the following formula:

[0082] Where Q, K, and V are the query, key, and value matrices, respectively. The model uses as a dimension. It extracts movement patterns (such as periodic commuting routes) and abnormal stop points (locations where the stay time exceeds a threshold).

[0083] Statistical characteristic calculation: Statistical analysis is used to calculate the frequency of violations (e.g., number of violations per month) and the correlation between accidents (using the conditional probability formula:

[0084] Visual feature extraction: Convolutional neural networks (CNNs) are used to analyze vehicle images and identify special markings (such as obscured license plates). CNNs extract feature maps through convolutional and pooling layers, and finally output classification results through fully connected layers.

[0085] Behavioral Feature Recognition: High-risk driving behaviors are identified using reinforcement learning models (such as Q-learning). The model uses the driving action as the state and the risky outcome as the reward, updating the Q-value through the Bellman equation.

[0086] Where α is the learning rate and γ is the discount factor, which identifies behaviors such as frequent lane changes or sudden braking.

[0087] Risk assessment: Multi-dimensional features are input into a fully connected neural network, which outputs a risk score S. The model uses a weighted summation formula:

[0088] in For eigenvalues, These are the feature weights (obtained through training). Risk levels are based on adaptive thresholds: high risk ( ), medium risk ( Low risk ), threshold and Adjust dynamically based on historical data distribution.

[0089] 2.4 Intelligent screening and control recommendations are generated based on risk level and user interaction: Intelligent filtering: Provides a web-based visual interface, allowing users to filter by dragging and dropping conditions (such as risk level and violation type). The system dynamically recommends conditions based on feature importance analysis, for example, prioritizing the combination of "high-risk level + nighttime driving".

[0090] Control Recommendation Generation: Recommendations are generated through a combination of a rules engine and a machine learning model. The rules engine executes predefined rules (e.g., "IF Risk Level = High AND Activity Area = Business District THEN Recommend Strengthening Monitoring"), while the machine learning model optimizes the recommendations based on historical performance data. Outputs include: Key monitoring periods: Based on the peak periods of vehicle trajectories, Gaussian mixture models are used to identify activity peaks.

[0091] Must-check checkpoints: By combining GIS data and risk heat maps, checkpoint locations are recommended using the shortest path algorithm.

[0092] Warning frequency: Set based on risk level, for example, a warning is issued once per minute for high-risk vehicles.

[0093] 2.5 Feedback-driven model optimization: Optimizing the system based on user feedback data. Feedback collection: Collect user feedback on screening results and control suggestions, such as false positives (marking normal vehicles as high risk) and false negatives (failure to identify high-risk vehicles).

[0094] Online learning: Incremental learning techniques are used to update the deep learning model. For false positives, stochastic gradient descent is used to adjust the model parameters.

[0095] Where θ represents the model parameters, η represents the learning rate, and L represents the loss function. Simultaneously, the weights of the selection criteria are adjusted; for example, the priority of criteria with high false positive rates is reduced.

[0096] Strategy optimization: Update control rules based on feedback, such as optimizing the warning frequency threshold, to ensure that the system adapts to new risk patterns.

[0097] For any parts not mentioned in this application, existing technologies may be used or referenced.

[0098] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0099] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for intelligent screening and control of high-risk vehicles, characterized in that, The method includes: Real-time acquisition of multi-source vehicle data, including basic vehicle information, driving trajectory, traffic violation records, and accident records; The multi-source data is cleaned and fused to construct a dynamic vehicle profile; Based on the vehicle dynamic profile, a deep learning model is used to extract multi-dimensional features and assess risks, generating vehicle risk scores and risk levels. Based on the risk level and user interaction, intelligent screening is performed, and targeted control suggestions are generated. Based on user feedback data, the deep learning model and selection strategy are optimized through an online learning mechanism.

2. The method according to claim 1, characterized in that, The real-time acquisition of multi-source vehicle data includes: By connecting with the public security big data platform, basic vehicle information, driving trajectory, traffic violation records, and accident records are collected in real time.

3. The method according to claim 1, characterized in that, The step of cleaning and fusing the multi-source data to construct a dynamic vehicle profile includes: Data cleaning algorithms are used to remove outlier data and noise, and missing data is filled in. Based on spatiotemporal correlation rules, multi-source data are fused into a unified vehicle dynamic profile, which includes static attributes and dynamic behavioral features.

4. The method according to claim 3, characterized in that, The use of deep learning models for multi-dimensional feature extraction and risk assessment includes: Temporal features, including movement patterns and abnormal dwell points, are extracted using an attention-based Transformer model or a temporal convolutional network. Statistical analysis methods were used to calculate the frequency of violations and the correlation between accidents as statistical characteristics. Convolutional neural networks were used to analyze special markings in vehicle images as visual features. A reinforcement learning model is used to identify high-risk driving behaviors as behavioral features. Based on the extracted multi-dimensional features, a vehicle risk score is generated through a risk assessment model, and the risk level is dynamically classified based on an adaptive threshold.

5. The method according to claim 4, characterized in that, The intelligent screening based on the risk level and user interaction, and the generation of targeted control suggestions, include: It provides a visual interactive interface that allows users to combine filter criteria by dragging or checking. Based on the importance analysis of the multi-dimensional features, high-frequency or high-weight screening conditions are dynamically recommended. Based on the screening results and risk characteristics, the system combines a rule engine with a machine learning model to generate control recommendations, including key monitoring periods, mandatory checkpoint locations, and warning frequencies.

6. The method according to claim 5, characterized in that, The optimization of the deep learning model and selection strategy based on user feedback data through an online learning mechanism includes: Collect user feedback data on screening results and control recommendations, including false alarms and missed alarms; The parameters of the deep learning model are updated using incremental learning techniques, and the weights and control rules of the screening criteria are adjusted.

7. The method according to claim 1, characterized in that, The method further includes: Based on the aforementioned risk levels and control recommendations, differentiated vehicle control strategies should be developed.

8. A high-risk vehicle intelligent screening and control device, characterized in that, The device includes: The data acquisition module is used to acquire multi-source vehicle data in real time. The data processing module is used to clean and fuse the multi-source data to construct a dynamic vehicle profile. The risk assessment module is used to perform multi-dimensional feature extraction and risk assessment based on the vehicle dynamic profile, and generate a vehicle risk score and risk level. The screening and control module is used to perform intelligent screening based on the risk level and user interaction operations, and generate targeted control suggestions. The optimization module is used to optimize the deep learning model and screening strategy based on user feedback data through an online learning mechanism.

9. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-7.

Images