Multi-dimensional data fusion system for operational truck risk rating

Abstract

The application provides a multi-dimensional data fusion system for truck operation risk rating, and relates to the technical field of data processing.The system comprises a data acquisition module for acquiring dynamic operation data and historical static information of a target truck; a data preprocessing module for field alignment, abnormal value elimination and format standardization; a feature analysis module for extracting dynamic behavior features of vehicle operation and performing statistical calculation within a preset time window; an event identification and risk assignment module for identifying single events or combined events, assigning event risk weights to the events, and determining corresponding road condition risk levels in combination with vehicle location information; a multi-dimensional fusion module for fusing features and generating a time-sequenced multi-dimensional risk feature vector; and a risk rating module for performing scoring processing on the multi-dimensional risk feature vector and outputting a corresponding risk rating result.The application improves the accuracy of the multi-dimensional data fusion system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a multi-dimensional data fusion system for risk rating of operating trucks. BACKGROUND

[0002] Existing risk rating technologies for operating trucks mostly rely on single-source data, such as vehicle historical violation records, annual inspection results, or single GPS trajectory information. The system usually matches these single-dimensional data with a pre-established risk model by setting weights to obtain a risk score. For example, some truck fleet management platforms generate a risk level based on the number of past traffic violations of the driver and the vehicle annual inspection results, combined with simple scoring rules. This method has low implementation cost and simple structure, but the data dimension is limited and it is difficult to fully reflect the risk status of the vehicle in different operating states.

[0003] In the cross-provincial long-distance transportation scenario, the existing rating method relying on single-dimensional data may have missed risk. For example, a truck has no major violation records in the local area and passes the annual inspection, but it has multiple sudden accelerations and sudden braking phenomena when driving on the highway for a long time. Such behaviors can only be captured by dynamic sensor data during vehicle operation. Since the existing system does not fuse these real-time dynamic data, the risk rating result of this truck may still be low, which may mislead the decision-making in insurance premium rate assessment or dispatch allocation, resulting in potential safety hazards being ignored. SUMMARY

[0004] The purpose of the present application is to provide a multi-dimensional data fusion system for risk rating of operating trucks, aiming to solve the problems mentioned in the background.

[0005] To solve the above technical problems, the technical solutions of the present application are as follows:

[0006] The multi-dimensional data fusion system for risk rating of operating trucks comprises:

[0007] A data acquisition module for acquiring dynamic operation data and historical static information of the target operating truck to form a preliminary multi-dimensional data set;

[0008] A data preprocessing module for field alignment, outlier removal, and format standardization of the preliminary multi-dimensional data set to generate structured multi-dimensional data;

[0009] A feature analysis module for extracting dynamic behavior features of vehicle operation from the structured multi-dimensional data and performing statistical calculation within a preset time window to form a feature statistical value data set;

[0010] An event identification and risk assignment module is configured to identify a single event or a combined event according to the feature statistical value dataset and a preset event trigger condition, and assign an event risk weight to the single event or the combined event, and determine a corresponding road condition risk level in combination with vehicle location information, and aggregate to form an event risk feature set;

[0011] A multi-dimensional fusion module is configured to fuse the event risk feature set with historical static information in structured multi-dimensional data to generate a time-sequenced multi-dimensional risk feature vector;

[0012] A risk rating module is configured to score the multi-dimensional risk feature vector to output a corresponding risk rating result.

[0013] Preferably, the feature analysis module comprises:

[0014] A dynamic feature segmentation unit is configured to divide the dynamic operation data into a plurality of time periods according to a preset time window to obtain time segmentation data;

[0015] A feature value calculation unit is configured to process the time segmentation data to calculate instantaneous values and cumulative values of the dynamic behavior features to form a feature value set;

[0016] A statistics generation unit is configured to calculate a change amplitude, a peak value and a cumulative impact of each dynamic behavior feature in a time period according to the feature value set, and generate a feature statistical value dataset.

[0017] Preferably, the event identification and risk assignment module comprises:

[0018] An event discrimination unit is configured to identify a single event or a combined event of the target truck in a preset time period according to the feature statistical value dataset and a preset trigger condition to obtain an event set;

[0019] A weight distribution unit is configured to distribute a preset risk weight value to the single event according to an event type of the event set, and perform a weighted operation on the combined event based on statistical values of each dynamic behavior feature and corresponding risk weights to obtain an event risk weight value;

[0020] A road condition evaluation unit is configured to look up a preset road condition risk level table in combination with location information of the event set to determine a corresponding road condition risk level, and aggregate the road condition risk level and the event risk weight value to form an event risk feature set.

[0021] Preferably, the multi-dimensional fusion module comprises:

[0022] A feature matching unit is configured to associate and match the event risk feature set with historical static information of the target truck according to a unique identifier of the target truck and a timestamp of the event occurrence to obtain fusion input data corresponding to the event;

[0023] a weighted fusion unit configured to multiply and sum the event risk feature set and the historical static information parameter respectively by a preset weight coefficient to generate a multi-dimensional risk feature vector containing static factors and dynamic factors according to the fused input data;

[0024] a sequence generation unit configured to arrange the multi-dimensional risk feature vector in a time sequence to form a multi-dimensional risk feature vector sequence.

[0025] Preferably, the statistical generation unit comprises:

[0026] a magnitude calculation unit configured to calculate the difference between the maximum value and the minimum value of each dynamic behavior feature in the time period according to the feature value set to generate a magnitude calculation result;

[0027] a peak extraction unit configured to analyze the time sequence of the dynamic behavior feature according to the magnitude calculation result, identify local maximum values and record their sizes and occurrence times to generate a peak extraction result;

[0028] a cumulative statistical unit configured to cumulatively add all dynamic behavior feature values in the time period according to the peak extraction result, and combine preset weights to weight the values of different time segments to form a cumulative impact quantity statistical result;

[0029] a result generation unit configured to combine the magnitude calculation result, the peak extraction result and the cumulative impact quantity statistical result to generate a feature statistical value data set.

[0030] Preferably, the event discrimination unit comprises:

[0031] a single event detection unit configured to compare the statistical value of each dynamic behavior feature with a corresponding preset trigger condition according to the feature statistical value data set, and generate a single event record when the statistical value of a certain dynamic behavior feature exceeds a preset first number threshold;

[0032] a combined event recognition unit configured to analyze a plurality of single events in the same time window according to the single event record, and combine these single events into a combined event to generate a combined event record when a preset combination condition is met, wherein the preset combination condition is that different types of single events are triggered simultaneously in the same time window, or the number of triggers of the same type of single event in a preset time interval exceeds a preset second number threshold;

[0033] an event set generation unit configured to aggregate the single event record and the combined event record to generate an event set.

[0034] Preferably, the weight allocation unit comprises:

[0035] An initial weight generating unit is configured to determine an initial risk weight value by comparing an event type of the event set with a preset risk weight table;

[0036] A dynamic adjustment unit is configured to correct the initial risk weight value according to historical accident data and operation environment data to generate a dynamically adjusted event risk weight value, wherein the correction process is: multiplying the risk level corresponding to different types of events and the occurrence frequency of the events in historical accident samples, and normalizing all product results;

[0037] A nonlinear interaction unit is configured to process the combined event according to the dynamically adjusted event risk weight value, multiply the statistical value of each feature and the corresponding dynamically adjusted weight value when two or more dynamic behavior characteristics are triggered at the same time window to obtain a single risk contribution, and enhance and superimpose multiple risk contributions according to a preset amplification factor to generate a combined event weight value.

[0038] Preferably, the road condition evaluation unit comprises:

[0039] A basic level retrieval unit is configured to determine an initial road condition risk level by comparing the location information of the target operating truck with a preset road condition risk level table;

[0040] An external data acquisition unit is configured to acquire external environment data of a road section where the vehicle is located when an event occurs, the external environment data including road congestion degree, real-time weather condition and historical accident occurrence frequency, and generate external road condition data;

[0041] A level correction unit is configured to correct the initial road condition risk level according to the external road condition data, increase the risk level when the external environment data indicates that the road risk increases, and decrease the risk level when the external environment data indicates that the road risk decreases, to obtain a dynamically updated road condition risk level.

[0042] Preferably, the weighted fusion unit comprises:

[0043] A coefficient optimization unit is configured to iteratively optimize a preset weighting coefficient by training multi-dimensional data samples of historical operating trucks to generate an optimized weighting coefficient, wherein the iterative optimization process is: inputting the historical samples into a machine learning model, comparing the difference between the risk feature vector output by the model and the corresponding accident record, and gradually adjusting the weighting coefficient until convergence;

[0044] A time sequence weighting unit is configured to fuse the event risk feature set and the historical static information parameters in the fusion input data according to the optimized weighting coefficient, and introduce a time decay factor in the processing process to reduce the weight of data at an earlier time and increase the weight of recent data, to obtain a time sequence weighting result;

[0045] a nonlinear fusion unit configured to perform nonlinear mapping and weighted summation on the interaction between the set of event risk features and the historical static information according to the time-series weighting result, to generate a multi-dimensional risk feature vector containing static factors and dynamic factors.

[0046] Preferably, the nonlinear fusion unit comprises:

[0047] a feature interaction modeling unit configured to model the feature interaction relationship between the set of event risk features and the historical static information according to the time-series weighting result, to convert the interaction strength of the two types of features into an interaction feature value, and form an interaction modeling result;

[0048] a nonlinear mapping unit configured to perform nonlinear function transformation on the dynamic behavior features, the event risk weights and the road risk levels respectively according to the interaction modeling result, and combine the mapping values through layer-by-layer iteration to generate a nonlinear mapping result, wherein the nonlinear function is a segmented function capable of amplifying high-risk feature contribution and suppressing low-risk feature noise;

[0049] a fusion generation unit configured to perform weighted summation on the dynamic behavior features and the static information parameters according to the nonlinear mapping result, to obtain a multi-dimensional risk feature vector containing feature interaction.

[0050] The above scheme of the present application at least includes the following beneficial effects:

[0051] By setting the data acquisition module, the data preprocessing module, the feature analysis module, the event recognition and risk assignment module, the multi-dimensional fusion module and the risk rating module, the multi-dimensional comprehensive evaluation of the operating truck risk rating can be realized. Compared with the existing scheme relying on a single data source, the present application can simultaneously obtain the dynamic running data and the historical static information of the vehicle, and process them under a unified structured data framework, thereby ensuring the data comprehensiveness and reliability of the risk analysis process.

[0052] At the data processing level, the feature analysis module can extract the dynamic behavior features such as acceleration, braking intensity and steering amplitude of the vehicle within a preset time window, and calculate the amplitude, peak value and cumulative impact, thereby truly reflecting the driving behavior of the vehicle in the actual running process. This processing method avoids the problem of inaccurate risk judgment caused by relying only on static information such as violation records and annual inspection results.

[0053] At the risk event level, the event identification and risk assignment module can convert abstract numerical features into identifiable driving events, and introduce road risk levels combined with location information, and finally aggregate to form an event risk feature set. For example, in the cross-provincial highway transportation scenario, if the vehicle frequently exhibits combined behaviors such as sudden acceleration and sharp turning, the present application can identify it as a high-risk event, thereby avoiding the situation that the traditional method misjudges it as a low-risk vehicle.

[0054] At the fusion analysis level, the present application generates a time-sequenced multi-dimensional risk feature vector by weighting and fusing the event risk feature set and the vehicle's historical static information through the multi-dimensional fusion module. This method not only retains the basic risk information of the vehicle's static condition, but also superimposes the dynamic risk features generated during operation, achieving a comprehensive description of the vehicle's risk state.

[0055] The risk rating result output by the risk rating module can intuitively reflect the vehicle risk level in the form of a score or a level, facilitating direct application in insurance premium rate assessment, vehicle fleet dispatch optimization, and transportation safety supervision scenarios. Thus, the present application can effectively solve the risk missed judgment and misjudgment problems caused by limited data dimensions in the prior art, improve the accuracy and practicality of the risk rating result, and reduce potential safety hazards. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 is the architecture diagram of the multi-dimensional data fusion system for operating truck risk rating provided by the embodiment of the present application. DETAILED DESCRIPTION

[0057] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0058] As Figure 1 shown, the embodiment of the present application proposes a multi-dimensional data fusion system for operating truck risk rating, which comprises:

[0059] A data acquisition module for acquiring dynamic running data and historical static information of a target operating truck to form a preliminary multi-dimensional data set;

[0060] A data preprocessing module for field alignment, outlier removal and format standardization of the preliminary multi-dimensional data set to generate structured multi-dimensional data;

[0061] a feature analysis module configured to extract dynamic behavior features of vehicle operation from the structured multi-dimensional data and perform statistical calculation within a preset time window to form a feature statistical value dataset;

[0062] an event identification and risk assignment module configured to identify single events or combined events according to the feature statistical value dataset and preset event triggering conditions, assign event risk weights to the events, and determine corresponding road condition risk levels in combination with vehicle location information to form an event risk feature set;

[0063] a multi-dimensional fusion module configured to fuse the event risk feature set with historical static information in the structured multi-dimensional data to generate a time-sequenced multi-dimensional risk feature vector;

[0064] a risk rating module configured to perform scoring processing on the multi-dimensional risk feature vector and output corresponding risk rating results.

[0065] In the embodiments of the present application, the data acquisition module, the data preprocessing module, the feature analysis module, the event identification and risk assignment module, the multi-dimensional fusion module, and the risk rating module are provided to realize multi-dimensional comprehensive evaluation of the operation risk of the operating truck. The system can simultaneously acquire dynamic operation data and historical static information of the vehicle, and form structured multi-dimensional data after data preprocessing, thereby ensuring the reliability and accuracy of subsequent analysis links. Through feature analysis, dynamic behavior features such as vehicle acceleration, braking intensity, and steering amplitude can be extracted within a preset time window, and the amplitude, peak value, and cumulative impact are calculated, thereby avoiding the problem of inaccurate risk judgment due to reliance on a single data source.

[0066] In the specific implementation process, the event identification and risk assignment module can identify single events or combined events, assign risk weights to the events, determine road condition risk levels in combination with vehicle location information, and finally form an event risk feature set. For example, in the long-distance transportation process on a highway, if a truck frequently exhibits combined behaviors of sudden acceleration and sudden turning during operation, although the truck has no major violations in historical records, the system can accurately reflect its potential risk through the event risk feature set, rather than relying only on historical static data for judgment.

[0067] Through the multi-dimensional fusion module, the event risk feature set and the vehicle historical static information are fused in a weighted manner to generate a time-sequenced multi-dimensional risk feature vector, so that the risk representation not only includes dynamic behavior, but also includes static information such as vehicle annual inspection results and past violations, thereby realizing deep fusion of multi-source data. The risk rating module outputs quantitative risk rating results, which can be in the form of scores or levels, for direct application in insurance pricing, vehicle fleet scheduling, or safety supervision.

[0068] Therefore, the system can comprehensively, dynamically and in combination with specific operating environment evaluate the risk level of the truck, effectively solves the problem of risk misjudgment or misjudgment caused by relying on single dimension data, and improves the accuracy and guidance of the risk rating result in actual application.

[0069] In an embodiment of the present application, the dynamic operation data can include but is not limited to vehicle speed, longitudinal acceleration, lateral acceleration, braking intensity, fuel consumption rate, engine speed, steering wheel angle, driving time and continuous driving time, etc.; the historical static information can include but is not limited to vehicle annual inspection record, maintenance record, historical insurance claim record, traffic violation record, vehicle manufacturing date and service life, etc. The above data can be obtained through a vehicle terminal device, a vehicle sensor, a telematics platform and a data interface of a traffic management department.

[0070] In a specific application scenario, the risk rating result can adopt a percentage numerical form, for example, the risk level is divided into 0-100 points, and the higher the score is, the greater the risk is; or an ordinal form, for example, 1-5 levels, wherein 1 represents low risk and 5 represents high risk. The rating result can be directly used as a basis for insurance pricing, for example, a vehicle with high risk level corresponds to a higher insurance rate, and can also be used in vehicle fleet scheduling and operation management, for example, a high-risk vehicle is preferentially arranged for short-distance transportation or a low-risk road. Through the above design, the system realizes the fusion analysis of dynamic and static information, and ensures the scientificity and application value of the risk rating result.

[0071] In an embodiment of the present application, the dynamic operation data collected by the data collection module can be obtained at a sampling frequency of 10Hz to 100Hz to ensure real-time tracking of the vehicle operating state. Among them, the high-frequency change characteristics such as acceleration and braking intensity are preferably sampled at 50Hz, and the low-frequency characteristics such as steering wheel angle can be sampled at 10Hz to reduce data redundancy. In order to ensure data quality, the data preprocessing module can remove outliers during collection. When the vehicle acceleration exceeds ±20 meters per second squared or the fuel consumption rate exceeds the reasonable interval, it is automatically marked as an anomaly and removed. In addition, the data from different sensors are aligned by time stamp and interpolated to the same sampling frequency, thereby forming a structured multi-dimensional data with consistent format, ensuring the reliability of the subsequent feature analysis link.

[0072] In a preferred embodiment of the present application, the feature analysis module comprises:

[0073] a dynamic feature segmentation unit, configured to divide the dynamic operation data into a plurality of time periods according to a preset time window to obtain time segmentation data;

[0074] The feature value calculation unit is configured to process the time segmentation data, calculate instantaneous values and cumulative values of the dynamic behavior features, and form a feature value set.

[0075] The statistical generation unit is configured to calculate variation amplitudes, peak values and cumulative influence amounts of each dynamic behavior feature in the time period according to the feature value set, and generate a feature statistical value data set.

[0076] In the embodiment of the present application, the dynamic feature segmentation unit, the feature value calculation unit and the statistical generation unit are arranged in the feature analysis module, so that the segmented processing and feature quantization of the vehicle operation data are realized. The dynamic feature segmentation unit can divide the dynamic operation data of the vehicle into multiple time periods according to the preset time window, so that the system can observe the vehicle operation features in the time dimension of fine granularity. The feature value calculation unit extracts the instantaneous values and the cumulative values in each time period on this basis, so as to ensure that both the sudden behavior and the long-term cumulative effect can be captured. The statistical generation unit further calculates the amplitudes, the peak values and the cumulative influence amounts of these feature values, and obtains the structured feature statistical value data set.

[0077] Through the above processing, the system can establish a connection between the short-time driving behavior and the long-time driving habit. For example, when the acceleration peak value of a truck frequently exceeds the standard in one hour of driving, and the cumulative braking intensity is much higher than the average level, even if the vehicle performs well in the historical violation records, the potential high-risk driving behavior of the vehicle can also be identified through the analysis of the module, so as to improve the accuracy of subsequent event identification.

[0078] In an embodiment of the present application, the preset time window can be set as a fixed time length (for example, 5 minutes, 10 minutes or 30 minutes) according to actual needs, or can be set as a fixed mileage section (for example, 5 kilometers or 10 kilometers) according to the cumulative driving mileage of the vehicle, or the time window length can be dynamically adjusted by an algorithm to adapt to different operating environments.

[0079] In the time window, the feature value calculation unit can calculate the instantaneous values and the cumulative values of the dynamic behavior features. The instantaneous value can be defined as the instantaneous physical quantity at the sampling time, for example, the longitudinal acceleration at a certain time. The cumulative value can be obtained by adding all the instantaneous values in the time window or calculating the average value, so as to reflect the overall level of the driving behavior. For example, in a 10-minute time window, if 60 longitudinal acceleration values are collected, the instantaneous value is the value at each sampling time, and the cumulative value can be the sum of the 60 values, or the average value obtained by dividing the sum by the sampling number.

[0080] In a specific application scenario, if a truck has multiple longitudinal acceleration instantaneous values exceeding 3 meters per second squared within a 10-minute time window, and the cumulative braking intensity of the time window is much higher than the average level of the vehicle fleet, the feature statistical value dataset generated by the statistical generation unit will highlight the risk characteristics of the vehicle, providing high-quality input data for subsequent event identification.

[0081] In an embodiment of the present application, the feature analysis module adopts different calculation methods for different types of dynamic behavior features when performing feature calculation. The cumulative value of the vehicle longitudinal acceleration can be obtained by accumulating the instantaneous values in the time window, the cumulative value of the braking intensity can be obtained by weighted accumulation of the absolute value of the deceleration, and the cumulative value of the steering amplitude can be obtained by accumulating the absolute change of the steering wheel angle to reflect the steering frequency.

[0082] In a preferred embodiment of the present application, the event identification and risk assignment module comprises:

[0083] An event discrimination unit is configured to identify single events or combined events of the target operating truck in a preset time period according to the feature statistical value dataset and a preset trigger condition, and obtain an event set;

[0084] A weight distribution unit is configured to assign a preset risk weight value to the single event according to the event type of the event set, and perform weighted operation on the combined event based on the statistical value of each dynamic behavior feature and the corresponding risk weight, to obtain an event risk weight value;

[0085] A road condition evaluation unit is configured to find a preset road condition risk level table in combination with the location information of the event set, determine the corresponding road condition risk level, and combine the road condition risk level and the event risk weight value to form an event risk feature set.

[0086] In the embodiment of the present application, by setting the event discrimination unit, the weight distribution unit and the road condition evaluation unit in the event identification and risk assignment module, the risk eventization representation of the dynamic behavior features is realized. The event discrimination unit can identify single events or combined events according to the feature statistical value dataset, thereby converting the abstract numerical features into specific driving event records. The weight distribution unit assigns a preset risk weight value to different types of events on this basis, and obtains the event risk weight through weighted calculation, so that the contribution of different events to the overall risk can be quantified. The road condition evaluation unit combines the vehicle location information to find the road condition risk level table, corrects the event risk weight value and outputs the event risk feature set.

[0087] In this way, the system can not only identify the risk of driving behavior itself, but also adjust the risk level in combination with the external environment. For example, the same sudden braking event may only be assigned a lower risk level if it occurs on an urban road, but the risk level will be significantly increased if it occurs on a highway in the rain. This module can dynamically adjust the event risk representation in different situations, thereby avoiding missed or false judgments caused by a single risk weight.

[0088] In an embodiment of the present application, the preset event trigger condition can be set according to traffic regulations, industry safety standards, and historical accident statistics. For example, an instantaneous longitudinal acceleration value greater than 3 meters per second squared is determined as a sudden acceleration event, a braking deceleration less than -4 meters per second squared is determined as a sudden braking event, and a steering wheel angular velocity greater than 300 degrees per second is determined as a sudden turning event.

[0089] In this embodiment, the event identification unit can analyze the feature statistical value dataset to identify single events or combined events of the target operating truck within a preset time period. For example, if sudden acceleration and sudden turning are detected within the same time window, it can be identified as a "dangerous overtaking event"; if three sudden braking events occur consecutively within a 5-minute time period, it can be identified as a "fatigue driving high-risk event".

[0090] The weight allocation unit can assign a preset risk weight value to a single event according to the event type of the event set, such as 0.3 for a sudden braking event, 0.2 for a sudden acceleration event, and 0.25 for a sudden turning event. For a combined event, the risk weight value of the combined event can be obtained by multiplying the statistical value of each feature by the corresponding weight and summing them up. For example, for the "sudden acceleration + sudden turning" event, the weight value is 0.2x acceleration statistical value + 0.25x turning statistical value.

[0091] The road condition assessment unit can combine the location information of the event set to find the corresponding level in the preset road condition risk level table. For example, the risk level of a highway is 1, the risk level of a national road is 2, and the risk level of a mountain road is 3. In a specific scenario, when a sudden braking event occurs on a mountain road in the rain, the road condition assessment unit will significantly increase its risk level, and combine the corrected risk level with the event risk weight to form the event risk feature set. In this way, it can avoid the missed and false judgments caused by the traditional method of judging risk based on a single data.

[0092] In a preferred embodiment of the present application, the multi-dimensional fusion module comprises:

[0093] The feature matching unit is configured to associate and match the event risk feature set with the historical static information of the target operating truck according to the unique identifier of the target operating truck and the timestamp of the event occurrence, to obtain fusion input data corresponding to the event.

[0094] a weighted fusion unit configured to multiply and sum the event risk feature set and the historical static information parameters respectively with preset weighting coefficients to generate a multi-dimensional risk feature vector containing static and dynamic factors according to the fusion input data;

[0095] a sequence generation unit configured to arrange the multi-dimensional risk feature vectors in time sequence to form a sequence of multi-dimensional risk feature vectors.

[0096] In the embodiment of the present application, by setting the feature matching unit, the weighted fusion unit and the sequence generation unit in the multi-dimensional fusion module, the deep fusion of the event risk feature set and the historical static information is realized. The feature matching unit can associate the real-time event risk feature with the historical record of the vehicle according to the unique identification of the vehicle and the event timestamp, so that the risk analysis is no longer limited to a single dimension. The weighted fusion unit then weights and sums the event risk feature set and the static parameters such as the vehicle annual inspection result and the violation record to obtain a multi-dimensional risk feature vector containing dynamic and static factors. The sequence generation unit further arranges these vectors in time sequence to form a time series input.

[0097] The technical effect of this fusion method is that it can realize the complementation of dynamic and static risk information. For example, if a vehicle has a minor unqualified item in the annual inspection and frequently appears high-risk events such as sharp turns in recent operation, the multi-dimensional risk feature vector after fusion will significantly improve the overall risk level of the vehicle, so that the rating is more in line with the real operating state. In contrast, the traditional rating method based on single static information may underestimate the risk level of the vehicle, while the fusion mechanism of the present application effectively improves the accuracy and sensitivity of risk assessment.

[0098] In an embodiment of the present application, the feature matching unit can match the event risk feature set with the historical static information of the target truck according to the unique identification of the target truck (such as the license plate number or the vehicle identification number VIN) and the timestamp of the event occurrence. For example, a vehicle has an emergency braking event at 10:00 on January 10, 2025, and the system can associate the event with the historical violation record, annual inspection result and insurance claim record of the vehicle to obtain the fusion input data.

[0099] The weighted fusion unit can multiply and weight sum the event risk feature set and the historical static information parameters respectively with preset weighting coefficients to form a risk feature vector containing multi-dimensional factors. For example, in one calculation scenario, the event risk feature set accounts for 0.5, the historical violation record accounts for 0.3, and the vehicle annual inspection information accounts for 0.2, then the final risk feature vector can be represented as [0.5 x event risk feature value + 0.3 x violation feature value + 0.2 x annual inspection feature value].

[0100] The sequence generation unit can arrange the generated multi-dimensional risk feature vectors in time sequence to form a sequence of multi-dimensional risk feature vectors. In this way, the risk states at different times are retained and can be provided as input to the risk rating module. For example, during a day of transportation, the vehicle can experience various situations such as low risk (normal driving on urban roads), medium risk (sudden braking on a highway), and high risk (sudden turning on a mountain road), and the sequence generation unit can concatenate the risk features at different time periods into a complete risk change trajectory for more accurate risk scoring.

[0101] In a preferred embodiment of the present application, the statistical generation unit comprises:

[0102] The amplitude calculation unit is configured to calculate the difference between the maximum value and the minimum value of each dynamic behavior feature in the time period according to the feature value set, and generate an amplitude calculation result.

[0103] The peak extraction unit is configured to analyze the time sequence of the dynamic behavior features according to the amplitude calculation result, identify local maximum values and record their magnitudes and occurrence times, and generate a peak extraction result.

[0104] The cumulative statistical unit is configured to cumulatively add all dynamic behavior feature values in the time period according to the peak extraction result, and combine the preset weights to weight the values of different time segments, and form a cumulative influence amount statistical result.

[0105] The result generation unit is configured to combine the amplitude calculation result, the peak extraction result, and the cumulative influence amount statistical result to generate a feature statistical value data set.

[0106] In the embodiment of the present application, by setting the amplitude calculation unit, the peak extraction unit, the cumulative statistical unit, and the result generation unit in the statistical generation unit, multi-angle quantitative processing of the dynamic behavior features is realized. The amplitude calculation unit can reflect the fluctuation range of the features in the time window through the difference between the maximum value and the minimum value, the peak extraction unit can capture extreme operations in the driving process, the cumulative statistical unit can reflect the cumulative influence of the driving behavior in the entire window, and the result generation unit can integrate these results of different dimensions to form a feature statistical value data set.

[0107] Through this multi-level feature extraction method, the system can comprehensively reflect the risk in the vehicle operation process. For example, in the same time period, two trucks both have sudden braking behavior, but one of them has a larger amplitude change range, a higher peak intensity, and a significantly larger cumulative influence amount. Through the processing of the statistical generation unit, the risk features of this vehicle are highlighted, providing more accurate data support for subsequent event identification, thereby improving the discrimination of risk rating.

[0108] In an embodiment of the present application, the amplitude calculation unit is capable of comparing the maximum value and the minimum value of each dynamic behavior feature in the feature value set within a time window and calculating the difference, thereby obtaining the amplitude of the feature. For example, in a 10-minute time window, if the maximum value of the longitudinal acceleration of the vehicle is 4 meters per second squared and the minimum value is -2 meters per second squared, the acceleration amplitude calculation result is 6 meters per second squared.

[0109] The peak extraction unit is capable of identifying local maximum or minimum values in the time series and recording their values and occurrence time points. For example, if the vehicle has three peak values of braking intensity in a 10-minute window, which are -5, -6, and -7 meters per second squared, respectively, the system will record these peak values and their corresponding time points.

[0110] The cumulative statistics unit is capable of adding all the dynamic behavior feature values within a time window item by item, and weighting different time segments according to a preset weight to highlight the risk impact of recent behavior. For example, if the vehicle acceleration is sampled 100 times in a window, the cumulative statistics unit can add the 100 values item by item, and give higher weight to the last 20 data in the window to reflect the importance of recent driving behavior.

[0111] The result generation unit combines the amplitude calculation result, the peak extraction result, and the cumulative impact statistics result to finally generate a feature statistics value data set. The data set can comprehensively reflect the instantaneous risk, extreme operation risk, and long-term cumulative risk of the vehicle, thereby providing high-quality input data for subsequent event identification.

[0112] In a preferred embodiment of the present application, the event discrimination unit comprises:

[0113] A single event detection unit is configured to compare the statistics value of each dynamic behavior feature with a corresponding preset trigger condition according to the feature statistics value data set, and generate a single event record when the statistics value of a certain dynamic behavior feature exceeds a preset first number threshold.

[0114] A combined event identification unit is configured to analyze multiple single events within the same time window according to the single event record, and combine these single events into a combined event to generate a combined event record when a preset combination condition is met, wherein the preset combination condition is that different types of single events are triggered at the same time within the same time window, or the number of triggers of the same type of single event within a preset time interval exceeds a preset second number threshold.

[0115] An event set generation unit is configured to aggregate the single event record and the combined event record to generate an event set.

[0116] In the embodiment of the present application, by setting a single event detection unit, a combined event identification unit and an event set generation unit in the event identification unit, hierarchical identification and set management of driving events are realized. The single event detection unit can identify single risk behaviors such as sudden braking and sudden acceleration according to characteristic statistical values and preset threshold values, the combined event identification unit further judges whether concurrent triggering of multiple single events exists within the same time window, or whether high-frequency occurrence of the same type of event exists within a certain time interval, and generates a combined event record when the preset combined condition is met. The event set generation unit can integrate single events and combined events to form an event set.

[0117] Through the design, the system can more accurately represent complex driving behaviors. For example, a single sudden braking may only represent an emergency situation, but if it occurs multiple times in a short period of time, it can be determined as a fatigue driving risk; if sudden acceleration and sudden turning occur concurrently within the same window, it may indicate that the vehicle has a dangerous overtaking behavior. Through identification and set management of combined events, the system can more truly reflect the potential risks of driving behaviors and avoid missed judgment and underestimation caused by single event judgment.

[0118] In an embodiment of the present application, the single event detection unit can compare the characteristic statistical values with the preset threshold values, and generate a single event record when a certain dynamic behavior characteristic exceeds the threshold value. For example, if the longitudinal acceleration of the vehicle within 5 minutes exceeds 3 meters per square second once, it is determined as a “sudden acceleration event”; if the braking deceleration is less than -4 meters per square second once, it is determined as a “sudden braking event”.

[0119] The combined event identification unit analyzes the single event records, and generates a combined event record when multiple single events of different types are triggered simultaneously within the same time window, or when high-frequency occurrence of the same type of event occurs within a certain time interval. For example, if sudden acceleration and sudden turning are detected simultaneously within the same time window, it can be determined as a “dangerous overtaking event”; if 3 sudden braking events occur consecutively within 10 minutes, it can be determined as a “fatigue driving high-risk event”.

[0120] The event set generation unit can integrate single events and combined events to form an event set. The event set can not only represent single risk events within a certain time window, but also reflect complex driving risk situations with multiple events superimposed. For example, if a truck simultaneously triggers a sudden braking event and a sudden turning event while driving on a mountain road, the event set will retain the combined event, so that subsequent risk weight allocation and road condition correction are more accurate.

[0121] In a preferred embodiment of the present application, the weight allocation unit comprises:

[0122] An initial weight generating unit is configured to determine an initial risk weight value by comparing an event type of the event set with a preset risk weight table;

[0123] A dynamic adjustment unit is configured to correct the initial risk weight value according to historical accident data and operation environment data to generate a dynamically adjusted event risk weight value, wherein the correction process is: multiplying the risk level corresponding to different types of events with the occurrence frequency of the events in historical accident samples, and normalizing all the product results;

[0124] A nonlinear interaction unit is configured to process the combined event according to the dynamically adjusted event risk weight value, multiply the statistical value of each feature with the corresponding dynamically adjusted weight value to obtain a single risk contribution when two or more dynamic behavior characteristics are triggered at the same time window, and enhance and superimpose multiple risk contributions according to a preset amplification factor to generate a combined event weight value.

[0125] In the embodiment of the present application, by setting the initial weight generating unit, the dynamic adjustment unit and the nonlinear interaction unit in the weight allocation unit, the hierarchical weighting and dynamic correction of the risk event are realized. The initial weight generating unit can determine the basic risk weight of the event according to the event type and the preset risk weight table, and the dynamic adjustment unit corrects the weight in combination with the historical accident data and the operation environment, so that the weight can change with the change of the actual driving environment. The nonlinear interaction unit further introduces the coupling effect between events in the combined event processing. When multiple risk behaviors occur at the same time, the risk contribution is enhanced and superimposed by introducing the amplification factor, so as to generate a combined event weight value which is more consistent with the real risk level.

[0126] The weight allocation unit can realize the fine and intelligent allocation of risk weight. For example, a single sudden braking event may have a low risk weight in urban roads, but it will be dynamically adjusted to a higher value in mountainous roads; when sudden acceleration and sudden turning are triggered at the same time, the system not only considers their respective risk weights, but also amplifies the combined effect through the nonlinear interaction unit, so that the risk result is more close to the actual accident occurrence rule. Through this processing mode, the system can output reasonable event risk feature set in different scenarios, thereby improving the accuracy and adaptability of the overall rating.

[0127] In an embodiment of the present application, the initial weight generating unit can find the corresponding initial risk weight value from the preset risk weight table according to the event type of the event set. For example, the initial weight of sudden acceleration event is set to 0.2, the initial weight of sudden braking event is set to 0.3, and the initial weight of sudden turning event is set to 0.25.

[0128] The dynamic adjustment unit can correct the initial weight in combination with historical accident data and operating environment. The correction process is: multiplying the risk level of a certain type of event by its occurrence frequency in historical accident samples, and normalizing the results of all event types, so that the sum of each risk weight value remains 1. For example, if the accident probability of sudden braking events is significantly higher than that of sudden acceleration events in historical data, the system will increase the weight value of sudden braking and decrease the weight value of sudden acceleration.

[0129] The nonlinear interaction unit can introduce the coupling effect between events in the combined event processing. When two or more dynamic behavior characteristics trigger at the same time window, the system multiplies the statistical value of each characteristic with the corresponding dynamically adjusted weight value to obtain a single risk contribution, and then enhances and superimposes these risk contributions through a preset amplification factor. For example, when sudden acceleration and sudden turning occur at the same time, the system will multiply the original risk contribution by an amplification factor of 1.5, thereby obtaining a higher combined event risk weight value, to more realistically reflect the increased accident risk brought by the coupling of driving behaviors.

[0130] Through the above settings, the weight allocation unit can achieve fine weighting of risk events, so that the system can not only reflect the risk of single driving behavior, but also accurately capture the superimposed risk effect under complex driving behavior, thereby improving the rationality and accuracy of the overall risk assessment result.

[0131] In a preferred embodiment of the present application, the road condition assessment unit comprises:

[0132] A basic level retrieval unit is configured to compare the position information of the target truck with a preset road condition risk level table to determine an initial road condition risk level.

[0133] An external data acquisition unit is configured to acquire external environment data of the road section where the vehicle is located when an event occurs, the external environment data including road congestion degree, real-time weather conditions and historical accident occurrence frequency, and generate external road condition data.

[0134] A level correction unit is configured to correct the initial road condition risk level according to the external road condition data, increase the risk level when the external environment data indicates an increase in road risk, and decrease the risk level when the external environment data indicates a decrease in road risk, to obtain a dynamically updated road condition risk level.

[0135] In the embodiment of the present application, by setting a basic level retrieval unit, an external data acquisition unit and a level correction unit in the road condition evaluation unit, dynamic risk correction of the vehicle operating environment is realized. The basic level retrieval unit can quickly match the initial level in the preset road condition risk level table according to the position information of the vehicle, for example, the expressway is low level, and the mountain road is high level. The external data acquisition unit further introduces real-time road state data, including traffic congestion index, real-time weather information and historical accident frequency of the road section, thereby making up for the limitations brought by relying only on the fixed level table. The level correction unit corrects the initial level based on these real-time external data, so that the road condition risk level can be dynamically updated with the change of environmental conditions.

[0136] Through the road condition evaluation unit, the timeliness and accuracy of the risk evaluation result can be ensured. For example, the risk of the same expressway is significantly different in sunny weather and in foggy weather. By introducing external information such as weather and historical accident rate, the risk level can be adjusted from 1 to 3, so that the event risk feature set is closer to the actual operating environment. In this way, not only the risk is not underestimated, but also more detailed basis is provided for insurance pricing and dispatch management.

[0137] In an embodiment of the present application, the basic level retrieval unit can find the initial road condition risk level from the preset road condition risk level table according to the position information of the target operating truck. For example, the expressway section can be set to level 1, the urban road to level 2, the mountain road to level 3, and the special dangerous section (such as sharp bend, long downhill) to level 4.

[0138] The external data acquisition unit can acquire external environmental data of the road section where the vehicle is located when the event occurs. The data can be collected through the vehicle-mounted camera, the traffic management platform or the meteorological service interface. Specifically, it includes: road congestion index (such as 1-5 levels), weather conditions (such as sunny, rainy, foggy, snowy, etc.), and historical accident frequency of the road section (such as the number of accidents or accident rate in the past year).

[0139] The level correction unit can correct the initial road condition risk level according to the external data. For example, when the system identifies that the vehicle is driving on the expressway in the rain, the initial risk level can be raised from 1 to 2; when it is detected that the expressway section has a high accident frequency in history, it can be further raised to 3; otherwise, when the road is smooth and the weather is good, the system can keep or reduce the risk level. The finally obtained dynamically updated road condition risk level can more accurately reflect the actual risk level of the environment where the vehicle is located.

[0140] In a preferred embodiment of the present application, the weighted fusion unit comprises:

[0141] a coefficient optimization unit configured to iteratively optimize preset weighting coefficients by training multidimensional data samples of historical operating trucks to generate optimized weighting coefficients, the iterative optimization process comprising: inputting the historical samples into a machine learning model, comparing differences between risk feature vectors output by the model and corresponding accident records, and gradually adjusting the weighting coefficients until convergence is achieved;

[0142] a time-series weighting unit configured to fuse event risk feature sets and historical static information parameters in the fusion input data according to the optimized weighting coefficients, and introduce a time decay factor in the fusion process to reduce weights of data at earlier time points and increase weights of recent data, to obtain time-series weighting results;

[0143] a nonlinear fusion unit configured to nonlinearly map and weightedly sum interactions between the event risk feature sets and the historical static information according to the time-series weighting results, to generate a multidimensional risk feature vector containing static and dynamic factors.

[0144] In the embodiments of the present application, the coefficient optimization unit, the time-series weighting unit and the nonlinear fusion unit are arranged in the weighting fusion unit, so that intelligent fusion of dynamic and static information is achieved. The coefficient optimization unit can iteratively optimize preset weighting coefficients using historical operating truck data samples, so as to ensure that the fusion process can continuously approach the real accident distribution and improve the prediction ability of the model. The time-series weighting unit introduces a time decay factor in the weighting process, so that the weight of recent data is higher and the weight of long-term data is reduced, thereby highlighting the sensitive influence of short-term driving behavior on the risk level. The nonlinear fusion unit further nonlinearly maps and weights the interactions between the event risk feature sets and the historical static information, to generate a risk feature vector containing multidimensional factors.

[0145] The weighting fusion unit can significantly improve the accuracy and robustness of risk feature expression. For example, when a vehicle frequently brakes suddenly in the past week, even if the vehicle has a good historical annual inspection record, the system will amplify the recent driving risk through time-series weighting and nonlinear fusion, so that the risk rating result is more consistent with the current operating state. Compared with the traditional linear weighting method, the fusion method can more accurately depict the risk level in complex driving scenarios.

[0146] In an embodiment of the present application, the coefficient optimization unit can iteratively optimize the preset weighting coefficients by training the multidimensional data samples of historical operating trucks. For example, 100,000 vehicle operation data samples in the historical accident database are selected as training data, input into a machine learning model (such as logistic regression, random forest or neural network), the difference between the risk feature vector output by the model and the actual accident record is compared, and the weighting coefficients are gradually adjusted by the gradient descent algorithm until the error is less than the set threshold 0.01 or the number of iterations reaches 500, thereby obtaining the optimized weighting coefficients.

[0147] The time-weighting unit can introduce a time decay factor in the fusion process to highlight the risk contribution of recent driving behavior. For example, an exponential decay function can be used as the weight allocation method, the closer the event risk feature set to the current time, the higher the weight; the more distant the event from the current time, the weight exponentially decreases. In this way, the sudden braking event within the past 24 hours will have a more significant impact on risk assessment, while the same event a week ago has less impact.

[0148] The nonlinear fusion unit can further map and weight the interaction between the event risk feature set and the historical static information through a nonlinear function. For example, a piecewise function is used, when the vehicle has a high number of unqualified annual inspections and frequent recent sudden acceleration behavior, the system will amplify the combined feature's contribution value in the risk vector; when both static information and dynamic features are at normal levels, the system will suppress their impact on the risk vector. The final generated multidimensional risk feature vector can more accurately reflect the real risk level of the vehicle at different time periods and in different environments.

[0149] In a preferred embodiment of the present application, the nonlinear fusion unit comprises:

[0150] The feature interaction modeling unit is configured to model the feature interaction relationship between the event risk feature set and the historical static information according to the time-weighting result, convert the interaction strength of the two types of features into an interaction feature value, and form an interaction modeling result.

[0151] The nonlinear mapping unit is configured to perform nonlinear function transformation on the dynamic behavior feature, the event risk weight and the road risk level respectively according to the interaction modeling result, and combine the mapping values through layer-by-layer iteration to generate a nonlinear mapping result, wherein the nonlinear function is a piecewise function that can amplify the contribution of high-risk features and suppress the noise of low-risk features.

[0152] The fusion generation unit is configured to perform weighted summation on the dynamic behavior feature and the static information parameter according to the nonlinear mapping result to obtain a multidimensional risk feature vector containing feature interaction.

[0153] In the embodiment of the present application, by setting the feature interaction modeling unit, the nonlinear mapping unit and the fusion generation unit in the nonlinear fusion unit, deep interaction modeling and enhanced representation of risk features are realized. The feature interaction modeling unit can identify the interaction between dynamic behavior features and static information, such as "vehicle service life x frequent emergency braking", and convert it into a new interaction feature value. The nonlinear mapping unit performs nonlinear transformation on these interaction features through neural network layering calculation or piecewise function mapping, so that high-risk combinations are amplified and low-risk combinations are suppressed, thereby improving the discrimination of feature representation. The fusion generation unit weights and sums the nonlinear mapping results and the original features to obtain a multi-dimensional risk feature vector containing interaction.

[0154] The nonlinear fusion unit can significantly enhance the expression ability of risk features. For example, two vehicles have both experienced emergency braking on the same operating route, but one vehicle is more than 10 years old and the other vehicle is only 2 years old. Through interaction modeling and nonlinear mapping, the system can highlight the higher risk of the former, thereby outputting a more realistic risk rating. This approach not only discovers potential risks that single features cannot reflect, but also improves the accuracy and reliability of risk prediction in complex scenarios.

[0155] In an embodiment of the present application, the feature interaction modeling unit can model the interaction between the event risk feature set and the historical static information according to the time sequence weighting results. For example, the vehicle service life and the number of emergency braking are constructed as a cross feature item to reflect the high risk of old vehicles under frequent emergency braking; and the vehicle maintenance record and the oil consumption anomaly are combined as a cross feature to reflect potential mechanical hidden dangers.

[0156] The nonlinear mapping unit can map these cross features through a nonlinear function, such as using a Sigmoid function to amplify the impact of high-risk combinations, or using a piecewise linear function to suppress the interference of low-risk features. In a preferred embodiment, when the vehicle age exceeds 10 years and the emergency braking frequency is higher than the threshold, the system increases the risk weight to twice the original value through a nonlinear function; and for a vehicle with a vehicle age of only 2 years, even if there is an emergency braking behavior, the weight increase is controlled at a low level.

[0157] The fusion generation unit can weight and sum the nonlinear mapping results with the original dynamic behavior features and static information to generate a multi-dimensional risk feature vector containing feature interactions. For example, the final vector can contain 20 dimensions, including both single event and road condition risks and the "vehicle age x emergency braking frequency" feature value obtained through cross modeling. This vector can more comprehensively represent the risk characteristics of the vehicle in complex operating scenarios, thereby improving the accuracy and robustness of the risk rating results.

[0158] The above describes the preferred embodiments of the present application. It should be noted that, for those skilled in the art, several improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements should also be considered as falling within the scope of the present application.

Claims

1. A multi-dimensional data fusion system for risk rating of commercial freight vehicles, characterized in that, The system includes: The data acquisition module is used to acquire dynamic operating data and historical static information of the target operating trucks to form a preliminary multidimensional dataset; The data preprocessing module is used to perform field alignment, outlier removal, and format normalization on the initial multidimensional dataset to generate structured multidimensional data. The feature analysis module is used to extract dynamic behavior features of vehicle operation from structured multidimensional data and perform statistical calculations within a preset time window to form a feature statistical value dataset. The event identification and risk assignment module is used to identify single events or combinations of events based on the feature statistical value dataset and preset event triggering conditions, and assign event risk weights to them. At the same time, it combines vehicle location information to determine the corresponding road condition risk level and summarizes them to form an event risk feature set. The multidimensional fusion module is used to fuse event risk feature sets with historical static information in structured multidimensional data to generate time-series multidimensional risk feature vectors. The risk rating module is used to score multidimensional risk feature vectors and output the corresponding risk rating results. The multi-dimensional fusion module includes: The feature matching unit is used to associate and match the event risk feature set with the historical static information of the target operating truck based on the unique identifier of the target operating truck and the timestamp of the event, so as to obtain the fused input data corresponding to the event. The weighted fusion unit is used to multiply the event risk feature set and historical static information parameters with preset weighting coefficients and sum them according to the fusion input data to generate a multi-dimensional risk feature vector containing static and dynamic factors. The sequence generation unit is used to arrange the multidimensional risk feature vectors in chronological order to form a multidimensional risk feature vector sequence. The weighted fusion unit includes: The coefficient optimization unit is used to iteratively optimize the preset weighting coefficients by training multi-dimensional data samples of historical operating trucks, and generate optimized weighting coefficients. The iterative optimization process is as follows: inputting historical samples into the machine learning model, comparing the difference between the risk feature vector output by the model and the corresponding accident records, and gradually adjusting the weighting coefficients until convergence. The time-series weighting unit is used to fuse the event risk feature set and historical static information parameters in the fused input data according to the optimized weighting coefficients. During the processing, a time decay factor is introduced to reduce the weight of data from earlier times and increase the weight of data from more recent times, so as to obtain the time-series weighting result. The nonlinear fusion unit is used to perform nonlinear mapping and weighted summation on the interaction between the event risk feature set and historical static information based on the time-series weighted results, generating a multidimensional risk feature vector containing static and dynamic factors. The nonlinear fusion unit includes: The feature interaction modeling unit is used to model the feature interaction relationship between the event risk feature set and historical static information based on the time-series weighted results, and to transform the interaction strength of the two types of features into interaction feature values ​​to form the interaction modeling results. The nonlinear mapping unit is used to perform nonlinear function transformations on dynamic behavior features, event risk weights, and road condition risk levels respectively based on the interactive modeling results, and to combine the mapping values ​​through layer-by-layer iteration to generate nonlinear mapping results. The nonlinear function is a piecewise function that can amplify the contribution of high-risk features and suppress the noise of low-risk features. The fusion generation unit is used to perform a weighted summation of dynamic behavioral features and static information parameters based on the nonlinear mapping results, to obtain a multidimensional risk feature vector that includes feature interactions.

2. The multi-dimensional data fusion system for risk rating of commercial freight vehicles according to claim 1, characterized in that, The feature analysis module includes: The dynamic feature segmentation unit is used to divide the dynamic running data into multiple time periods according to a preset time window to obtain time segmented data. The feature value calculation unit is used to process time-segmented data, calculate the instantaneous and cumulative values ​​of dynamic behavior features, and form a feature value set; The statistical generation unit is used to calculate the change range, peak size, and cumulative impact of each dynamic behavioral feature within a time period based on the feature value set, and to generate a feature statistical value dataset.

3. The multi-dimensional data fusion system for risk rating of commercial freight vehicles according to claim 1, characterized in that, The event identification and risk assignment module includes: The event discrimination unit is used to identify single or combined events of the target operating truck within a preset time period based on the feature statistical value dataset and preset triggering conditions, and obtain an event set. The weighting unit is used to assign a preset risk weight value to a single event based on the event type of the event set, and to perform a weighted calculation on the statistical values ​​of the combined events based on each dynamic behavioral characteristic and the corresponding risk weight to obtain the event risk weight value. The road condition assessment unit is used to combine the location information of the event set, look up the preset road condition risk level table, determine the corresponding road condition risk level, and summarize it with the road condition risk level and event risk weight value to form an event risk feature set.

4. The multi-dimensional data fusion system for risk rating of commercial freight vehicles according to claim 2, characterized in that, The statistical generation unit includes: The amplitude calculation unit is used to calculate the difference between the maximum and minimum values ​​of each dynamic behavior feature within a time period based on the feature value set, and generate the amplitude calculation result. The peak extraction unit is used to analyze the time series of dynamic behavior features based on the amplitude calculation results, identify local maxima and record their magnitude and occurrence time, and generate peak extraction results. The cumulative statistics unit is used to accumulate all dynamic behavioral feature values ​​within a time period based on the peak extraction results, and to weight the values ​​of different time segments with preset weights to form the cumulative impact statistical results. The result generation unit is used to combine the amplitude calculation results, peak extraction results, and cumulative impact statistics results to generate a feature statistics dataset.

5. The multi-dimensional data fusion system for risk rating of commercial freight vehicles according to claim 3, characterized in that, The event discrimination unit includes: A single event detection unit is used to compare the statistical value of each dynamic behavior feature with the corresponding preset trigger condition based on the feature statistical value dataset. When the statistical value of a certain dynamic behavior feature exceeds the preset first count threshold, a single event record is generated. The combined event identification unit is used to analyze multiple single events within the same time window based on the single event record. When a preset combination condition is met, the single events are combined into a combined event to generate a combined event record. The preset combination condition is that different types of single events are triggered simultaneously within the same time window, or the number of times the same type of single event is triggered exceeds a preset second threshold within a preset time interval. The event set generation unit is used to summarize the single event record and the combined event record to generate an event set.

6. The multi-dimensional data fusion system for risk rating of commercial freight vehicles according to claim 3, characterized in that, The weight allocation unit includes: The initial weight generation unit is used to determine the initial risk weight value by comparing the event type of the event set with a preset risk weight table. The dynamic adjustment unit is used to correct the initial risk weight value based on historical accident data and operating environment data, and generate dynamically adjusted event risk weight values. The correction process is as follows: multiply the risk level corresponding to different types of events by their occurrence frequency in the historical accident sample, and normalize all product results. The nonlinear interaction unit is used to process combined events based on dynamically adjusted event risk weight values. When two or more dynamic behavioral features are triggered simultaneously within the same time window, the statistical value of each feature is multiplied by the corresponding dynamically adjusted weight value to obtain a single risk contribution. Then, multiple risk contributions are enhanced and superimposed according to a preset amplification factor to generate a combined event weight value.

7. The multi-dimensional data fusion system for risk rating of commercial freight vehicles according to claim 3, characterized in that, The road condition assessment unit includes: The basic level retrieval unit is used to compare the location information of the target operating truck with the preset road condition risk level table to determine the initial road condition risk level. An external data acquisition unit is used to acquire external environmental data of the road segment where the vehicle is located when an event occurs. The external environmental data includes the degree of road congestion, real-time weather conditions, and historical accident frequency, and generates external road condition data. The risk level correction unit is used to correct the initial road risk level based on external road condition data. When the external environmental data indicates that the road risk is increasing, the risk level is increased; when the external environmental data indicates that the road risk is decreasing, the risk level is decreased, resulting in a dynamically updated road risk level.

Images