Operation vehicle risk prediction method and platform based on multi-factor analysis

Abstract

The invention discloses an operating vehicle risk prediction method and platform based on multi-factor analysis, and relates to the technical field of automobile information processing, and the method comprises the steps: collecting and obtaining a vehicle multi-dimensional operation data stream; extracting and obtaining risk factor information; crawling based on the risk factor information to obtain a risk data set, and carrying out weight distribution on the risk factor information by adopting the risk data set to obtain weight distribution information; training and optimizing the risk data set based on the weight distribution information, and constructing a vehicle risk prediction model; and performing risk prediction on the vehicle multi-dimensional operation data flow by using the vehicle risk prediction model, outputting a vehicle risk prediction score, and performing vehicle risk strategy management based on the vehicle risk prediction score. The technical problems of dependence of a single factor and incomplete data in an existing vehicle risk scoring system are solved, and the technical effects of providing more accurate and comprehensive vehicle risk prediction through multi-dimensional factor analysis and realizing personalized risk management are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of automobile information processing, and in particular to a method and platform for predicting operating vehicle risks based on multi-factor analysis. Background Art

[0002] With the rapid development of social economy, the number of cars has increased sharply, and the frequent occurrence of traffic accidents has become a social problem worldwide. Road traffic safety not only affects public safety and people's lives and property, but also has a profound impact on the stability and development of social economy. In this context, the insurance industry is facing tremendous pressure. Traditional vehicle risk assessment methods usually rely on a single historical accident record or simple driving behavior monitoring, resulting in insufficient accuracy and comprehensiveness of the assessment results, which is difficult to meet the increasingly complex traffic environment and personalized needs.

[0003] In recent years, with the rapid development of vehicle-mounted sensor technology and big data analysis technology, it has become possible to monitor driver behavior and vehicle status in real time. The popularity of sensors such as GPS, cameras, gyroscopes, and radars has enabled the vehicle's operating data to be more accurately collected and transmitted to the cloud for analysis. However, existing vehicle risk scores mostly rely on a single factor scoring method, such as driving behavior, accident history, or customer feedback. Although this evaluation method can reflect the driver's safety to a certain extent, it is too dependent on a specific factor and cannot fully and accurately reflect the driver's overall risk status. For example, a driver's behavior score can only reflect his driving performance in a specific situation, while the historical score ignores the dynamic changes in the driver's behavior.

[0004] Therefore, there is an urgent need for a vehicle risk prediction method that can comprehensively consider multiple influencing factors, effectively integrate information from multiple dimensions such as driving behavior, environmental factors, and vehicle performance, and dynamically adjust the scoring criteria to achieve more accurate risk predictions and personalized management strategies. Summary of the invention

[0005] The present application provides a method and platform for predicting operational vehicle risks based on multi-factor analysis, which is used to solve the technical problems of single factor reliance and incomplete data in existing vehicle risk scoring systems.

[0006] In view of the above problems, the present application provides a method and platform for predicting operating vehicle risks based on multi-factor analysis.

[0007] The first aspect of the present application provides a method for predicting risks of operating vehicles based on multi-factor analysis, the method comprising: acquiring a multi-dimensional operation data stream of a vehicle through a sensor perception network, wherein the sensor perception network comprises a GPS, a camera, a gyroscope and a radar; extracting and obtaining vehicle risk factor information, wherein the vehicle risk factor information comprises a driving behavior factor set, an environmental factor factor set and a vehicle status factor set; crawling and obtaining an operating vehicle risk data set based on the vehicle risk factor information, and using the operating vehicle risk data set to weight the vehicle risk factor information to obtain risk factor weight allocation information; training and optimizing the operating vehicle risk data set based on the risk factor weight allocation information to construct a vehicle risk prediction model; using the vehicle risk prediction model to perform risk prediction on the multi-dimensional operation data stream of the vehicle, outputting a vehicle risk prediction score, and performing vehicle risk strategy management based on the vehicle risk prediction score.

[0008] The second aspect of the present application provides an operational vehicle risk prediction platform based on multi-factor analysis, the platform comprising: a data acquisition module, the data acquisition module is used to collect and acquire a vehicle multi-dimensional operation data stream through a sensor perception network, wherein the sensor perception network includes GPS, a camera, a gyroscope and a radar; a risk factor extraction module, the risk factor extraction module is used to extract and acquire vehicle risk factor information, the vehicle risk factor information includes a driving behavior factor set, an environmental factor factor set and a vehicle status factor set; a weight allocation module, the weight allocation module is used to crawl and acquire an operational vehicle risk data set based on the vehicle risk factor information, and use the operational vehicle risk data set to weight the vehicle risk factor information to obtain risk factor weight allocation information; a training and tuning module, the training and tuning module is used to train and tune the operational vehicle risk data set based on the risk factor weight allocation information to construct a vehicle risk prediction model; a risk management module, the risk management module is used to use the vehicle risk prediction model to perform risk prediction on the vehicle multi-dimensional operation data stream, output a vehicle risk prediction score, and perform vehicle risk strategy management based on the vehicle risk prediction score.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages: The vehicle multi-dimensional operation data stream is acquired through the sensor perception network, wherein the sensor perception network includes GPS, camera, gyroscope and radar; the vehicle risk factor information is extracted and obtained, and the vehicle risk factor information includes a driving behavior factor set, an environmental factor factor set and a vehicle status factor set; based on the vehicle risk factor information, the operational vehicle risk data set is crawled to obtain the vehicle risk factor information, and the vehicle risk factor weight distribution information is obtained by using the operational vehicle risk data set; the operational vehicle risk data set is trained and optimized based on the risk factor weight distribution information to construct a vehicle risk prediction model; the vehicle risk prediction model is used to predict the risk of the vehicle multi-dimensional operation data stream, and the vehicle risk prediction score is output, and the vehicle risk strategy management is performed based on the vehicle risk prediction score. The technical effect of providing more accurate and comprehensive vehicle risk prediction through multi-dimensional factor analysis and realizing personalized risk management is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 A schematic flow chart of a method for predicting operating vehicle risks based on multi-factor analysis provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of an operating vehicle risk prediction platform based on multi-factor analysis provided in an embodiment of the present application.

[0012] Explanation of the accompanying drawings: data acquisition module 11, risk factor extraction module 12, weight allocation module 13, training and tuning module 14, risk management module 15. DETAILED DESCRIPTION

[0013] The present application provides an operational vehicle risk prediction method and platform based on multi-factor analysis, which is used to solve the technical problems of single factor reliance and incomplete data in the existing vehicle risk scoring system.

[0014] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0015] Embodiment 1, as Figure 1 As shown, the present application provides a method for predicting operating vehicle risks based on multi-factor analysis, the method comprising: The multi-dimensional operation data stream of the vehicle is collected and acquired through a sensor perception network, wherein the sensor perception network includes GPS, camera, gyroscope and radar.

[0016] In one embodiment, a sensor perception network can collect multi-dimensional operation data of operating vehicles in real time. This sensor perception network includes different types of sensors such as GPS (Global Positioning System), cameras, gyroscopes and radars. Each sensor is responsible for obtaining a specific type of data. Among them, GPS is used to obtain the precise location, driving path, speed, and geographical area (such as cities, road sections, highways, etc.) of the vehicle. Through GPS, the vehicle's driving trajectory can be tracked in real time to understand the vehicle's mileage, speed, and whether there is speeding or other behaviors. The camera is responsible for photographing the environment around the vehicle, monitoring possible traffic conditions, obstacles, road conditions, etc., and can also obtain driver behavior data, such as whether the driver is distracted (such as making phone calls, smoking, etc.). These image data can assist in judging the driver's driving environment and driving behavior, and further affect risk assessment; the gyroscope is used to measure the dynamic motion state of the vehicle, including the vehicle's angle change, inclination, steering, etc., to help determine whether the vehicle has sharp turns, rapid lane changes, etc., and then evaluate the driving stability and safety; the radar can detect the position and motion state of objects in the surrounding environment, and is especially suitable for monitoring at night or in low visibility conditions, and can understand obstacles ahead, the approach of other vehicles, and possible collision risks; these sensors work together to collect and transmit large amounts of data in real time, forming a complete multi-dimensional vehicle operation data stream, providing an accurate basis for subsequent risk assessment, behavior analysis and decision-making.

[0017] Vehicle risk factor information is extracted and obtained, wherein the vehicle risk factor information includes a driving behavior factor set, an environmental factor factor set, and a vehicle state factor set.

[0018] In one embodiment, vehicle risk factor information is extracted, and this vehicle risk factor information includes a driving behavior factor set, an environmental factor factor set, and a vehicle status factor set; the driving behavior factor set reflects the driver's specific driving habits and operating behaviors, including the driver's behavioral patterns during driving, such as sudden acceleration, sudden braking, frequent lane changes, speeding, total online time, night operation time, etc.; the environmental factor factor set reflects the impact of the external environment on vehicle driving safety, including factors such as weather, road conditions, and road section characteristics, such as rain, snow, haze, road flatness, number of curves, urban road driving time, highway driving time, etc.; the vehicle status factor set involves the state and performance of the vehicle itself, which may affect its safety, such as the brake system, tire wear, vehicle load, number of vehicle rollovers, etc.; by extracting and obtaining these factors, the safety of operating vehicles can be comprehensively evaluated, and an accurate calculation basis can be provided for vehicle risk scoring, providing drivers, insurance companies, and fleet managers with more scientific risk prediction and management strategies.

[0019] An operating vehicle risk data set is crawled based on the vehicle risk factor information, and the operating vehicle risk data set is used to perform weight allocation on the vehicle risk factor information to obtain risk factor weight allocation information.

[0020] In one embodiment, after obtaining vehicle risk factor information, various factors in the vehicle risk factor information are used as a benchmark to obtain historical vehicle data related to these factors from multiple data sources (such as fleet management systems, insurance company databases, traffic platforms, etc.) to form an operating vehicle risk data set, including the driver's driving record, accident data, violations, weather conditions, road conditions and other information. These data are not limited to data collected by sensors, but also include data from external sources such as accident history and traffic violation records. After obtaining the operational vehicle risk data set, the risk factors of each vehicle are weighted using this operational vehicle risk data set. The allocation process includes critical analysis and risk correlation assessment. The purpose of the allocation is to quantify the impact of different factors on the overall risk of the vehicle. For example, speeding may be assigned a higher weight because it is directly related to the severity of the accident. For example, the factor of speeding accounts for 3%. If a vehicle has 10 speeding incidents, each time the speed exceeds 20km / h, the risk weight of this factor will increase. If an operational vehicle operates for more than 4 hours at night, a 2% weight will be assigned to this factor through analysis, because the risk of driving at night is relatively high, and drivers are prone to fatigue, which affects safety. Through the above process, the final risk factor weight allocation information is obtained, which describes the relative importance of each factor in the overall risk score. In this way, the risk of each vehicle can be dynamically calculated based on these weights, thereby realizing personalized risk assessment and prediction.

[0021] In a possible implementation, obtaining risk factor weight allocation information includes: Acquire a data preprocessing program, use the data preprocessing program to perform standardized preprocessing on the operating vehicle risk data set to obtain a standard operating vehicle risk data set; perform a criticality analysis on the vehicle risk factor information based on the standard operating vehicle risk data set to screen and obtain key risk factor information; use the standard operating vehicle risk data set to perform a risk correlation assessment on the key risk factor information to obtain a key risk factor correlation coefficient; perform weight allocation on the key risk factor information based on the key risk factor correlation coefficient to obtain the risk factor weight allocation information.

[0022] Optionally, after obtaining the operational vehicle risk data set, obtain and activate a data preprocessing program. After activating the data preprocessing program, use this data preprocessing program to perform standardized preprocessing on the operational vehicle risk data set so that the data in the operational vehicle risk data set are on the same scale. In this process, by subtracting the mean of the feature from each data point in the operational vehicle risk data set and dividing it by the standard deviation of the feature, the data is converted into zero mean and unit variance to obtain a standard operational vehicle risk data set, so that the operational vehicle risk data set has the same scale for subsequent analysis. Subsequently, using the standard operational vehicle risk data set, each risk factor is evaluated to determine which factors have a greater impact on risk prediction. The evaluation process includes contribution evaluation, criticality screening, etc. After analyzing these steps, a data set containing key risk factor information is generated. For the screened key risk factor information, a risk correlation assessment is performed to evaluate the relationship between each factor and the overall risk, as well as the relationship between the factors, so as to obtain the key risk factor correlation coefficient. For example, the number of sudden acceleration alarms may be strongly correlated with the accident risk, so the correlation coefficient is high, while a certain Some factors, such as mileage, may have weak correlation. After obtaining the key risk factor correlation coefficients, each key risk factor will be weighted based on these coefficients. The higher the correlation coefficient, the closer the relationship between the factor and the risk, so its weight is greater. Factors with lower correlation coefficients have relatively smaller weights. The specific calculation of the weight of each key risk factor is the key risk factor correlation coefficient corresponding to the key risk factor divided by the sum of the correlation coefficients of all key risk factors. For example, the correlation coefficient of the number of sudden acceleration alarms is 0.8, and the correlation coefficient of the number of sudden braking alarms is 0. .7, the correlation coefficient of speeding times is 0.6, the correlation coefficient of night operation time is 0.5, the correlation coefficient of fatigue driving times is 0.4, the correlation coefficient of urban road driving time is 0.3, the correlation coefficient of highway driving time is 0.3, and the total correlation coefficient is 3.6. Then, the weight of sudden acceleration alarm number is 0.8 / 3.6, the weight of sudden braking alarm number is 0.7 / 3.6, and so on; finally, according to the calculated weights of all factors, a clear risk factor weight distribution information is formed, which provides a basis for the subsequent risk prediction model training.

[0023] In a possible implementation, obtaining key risk factor information includes: According to the vehicle risk factor information, an initial driving behavior factor cluster, an initial environmental factor cluster and an initial vehicle state factor cluster are determined; based on the standard operating vehicle risk data set, the contribution of each risk factor in the initial driving behavior factor cluster, the initial environmental factor cluster and the initial vehicle state factor cluster is evaluated to obtain a driving behavior factor contribution set, an environmental factor contribution set and a vehicle state factor contribution set; based on the driving behavior factor contribution set, the environmental factor contribution set and the vehicle state factor contribution set, the initial driving behavior factor cluster, the initial environmental factor cluster and the initial vehicle state factor cluster are critically screened to determine key risk factor information.

[0024] Optionally, based on the driving behavior factor set, environmental factor factor set and vehicle state factor set in the obtained vehicle risk factor information, an initial driving behavior factor cluster, an initial environmental factor factor cluster and an initial vehicle state factor cluster are defined, wherein the initial driving behavior factor cluster includes all factors related to driver behavior, such as sudden acceleration, sudden braking, frequent lane changes, speeding, total online time, night operation time, etc.; the initial environmental factor factor cluster includes factors related to the external environment, such as rain, snow, haze, road flatness, number of curves, urban road driving time, and highway driving time. etc.; the initial vehicle status factor cluster includes factors related to the status of the vehicle itself, such as the brake system, tire wear, vehicle load, number of vehicle rollovers, etc.; the initial set of these factor clusters covers all dimensions of vehicle risk assessment, providing a preliminary factor set for subsequent analysis and screening; after obtaining the initial driving behavior factor cluster, the initial environmental factor factor cluster and the initial vehicle status factor cluster, each factor cluster is traversed, and according to the traversed factors, the data corresponding to the factor is extracted from the standard operating vehicle risk data set, and based on these data, the information gain of the factor is calculated.

[0025] Taking the number of sudden accelerations as an example, data related to the number of sudden accelerations are extracted from the standard operating vehicle risk data set. These data reflect the number of sudden acceleration behaviors of operating vehicles in a certain period of time and whether each sudden acceleration is related to an accident. Assuming that there are 100 vehicles in the data set, 30 of which have accidents and 70 have not. By calculating the entropy of the target variable, the entropy of the accident is 0.88 bits. Subsequently, the conditional entropy of the target variable is calculated under different sudden acceleration times. When the number of sudden accelerations is 0, there are 60 vehicles, 10 accidents occur, and 50 no accidents occur. Under this number, the conditional entropy is 0.61 bits, and the probability is 0.6. When the number of sudden accelerations is 1, there are 20 vehicles, 5 accidents occur, and 15 no accidents occur. Under this number, the conditional entropy is 0.81 bits, and the probability is 0.2. When the number of sudden accelerations is 2 or more, there are 20 vehicles, and accidents occur. 15 times, 5 times without accident, then under this number, the conditional entropy is 0.81 bits, the probability is 0.2, these conditional entropies represent the uncertainty of accident occurrence when the number of sudden acceleration is known; then, the probability under each number is multiplied by the conditional entropy, and all the products are added together to obtain the total conditional entropy of the number of sudden accelerations, and then the entropy of the accident is subtracted from the total conditional entropy of the number of sudden accelerations to obtain the information gain; the larger the information gain, the more helpful the factor is for predicting the target variable, which means that the number of sudden accelerations has a greater impact on the occurrence of accidents. Therefore, the calculated information gain is used as the contribution of the factor; repeat the above process to calculate the contribution of each factor, and according to the corresponding relationships in the initial driving behavior factor cluster, the initial environmental factor factor cluster and the initial vehicle state factor cluster, these contributions are sorted into a driving behavior factor contribution set, an environmental factor factor contribution set and a vehicle state factor contribution set.

[0026] After evaluating the contribution of each factor to the overall risk, a critical screening is performed. The screening criteria are based on the contribution assessment results, that is, those factors that have a significant impact on risk prediction are retained, and factors with less impact on risk are eliminated. For example, the contribution threshold is set at 10%, so that only factors with a contribution greater than the threshold will be retained; through the above screening, the key risk factor information is finally determined. These factors play a decisive role in vehicle risk assessment. The final set of key factors may include the number of sudden acceleration alarms, weather conditions, brake system conditions, etc. These factors will become the basis for building a risk prediction model.

[0027] In a possible implementation, determining key risk factor information includes: Based on the driving behavior factor contribution set, the environmental factor factor contribution set and the vehicle state factor contribution set, the initial driving behavior factor cluster, the initial environmental factor factor cluster and the initial vehicle state factor cluster are clustered and updated respectively to obtain a driving behavior factor update cluster, an environmental factor factor update cluster and a vehicle state factor update cluster; a preset contribution selection threshold is obtained, and a risk factor cluster distance screening threshold is determined according to the preset contribution selection threshold; based on the risk factor cluster distance screening threshold, the driving behavior factor update cluster, the environmental factor factor update cluster and the vehicle state factor update cluster are screened for key factors in turn to determine key risk factor information.

[0028] Optionally, after obtaining the driving behavior factor contribution set, the environmental factor factor contribution set, and the vehicle status factor contribution set, each factor cluster will be clustered and updated. The main purpose of this process is to merge factors with similar contributions, optimize the structure of the factor cluster, and avoid redundant factors. For each factor cluster (driving behavior, environmental factors, vehicle status), all factors in it are treated as a separate cluster and sorted from high to low according to their contribution, and then the distance between each two factors is measured using the absolute difference. Based on the calculated distance, the factors are clustered, so that factors with similar contributions are combined together (i.e., the distance meets the distance threshold) to form a new factor sub-cluster. Through this method, factors that have a greater impact on risk prediction can be identified and classified into the same sub-cluster for subsequent analysis and model training. If the distance between two factors is small, i.e., it meets the distance threshold, they will be divided into the same sub-cluster; if the difference is large, they will be assigned to different sub-clusters. After clustering, each factor cluster will be updated to a factor update cluster, which contains multiple factor sub-clusters with similar contributions, and the average distance between all factors in the factor sub-cluster is used as the cluster within the sub-cluster. Distance; After completing the cluster update of the factors, in order to ensure that the screened factors have the greatest impact on risk assessment, a preset contribution selection threshold will be obtained. This threshold is determined based on historical data and practical experience. Then, by calculating the mean of the difference between all contributions and this preset contribution selection threshold, the risk factor cluster distance screening threshold is obtained, which is used to measure the redundancy between factors within the cluster. When the distance within the cluster is less than the threshold, it means that the data within the cluster is too redundant; then, by updating the cluster of each sub-cluster in the driving behavior factor update cluster, the environmental factor factor update cluster, and the vehicle status factor update cluster, the cluster The inter-cluster distance is compared with the risk factor cluster distance screening threshold, and the sub-clusters greater than or equal to the risk factor cluster distance screening threshold are screened out to avoid excessive redundancy of data within the sub-cluster, and then the mean of the cluster contribution of the screened sub-clusters is compared with the preset contribution selection threshold, and the sub-clusters greater than or equal to the preset contribution selection threshold are stored in the key risk factor information. These factors in the key risk factor information are factors that contribute more to the overall risk assessment and have less redundancy; through the above process, the representativeness and accuracy of the screened factors in the risk prediction model are ensured, thereby improving the accuracy and reliability of the overall risk assessment.

[0029] In a possible implementation, obtaining the key risk factor correlation coefficient includes: The standard operational vehicle risk data set is used to perform risk correlation assessment on the driving behavior factor update cluster, environmental factor factor update cluster and vehicle state factor update cluster in the key risk factor information, respectively, to obtain the driving behavior cluster risk correlation coefficient, the environmental factor cluster risk correlation coefficient and the vehicle state cluster risk correlation coefficient; based on the standard operational vehicle risk data set, intra-cluster correlation assessment is performed on each risk factor in the driving behavior factor update cluster, the environmental factor factor update cluster and the vehicle state factor update cluster, respectively, to obtain the driving behavior cluster factor correlation coefficient, the environmental factor cluster factor correlation coefficient and the vehicle state cluster factor correlation coefficient; based on the driving behavior cluster risk correlation coefficient, the environmental factor cluster risk correlation coefficient and the vehicle state cluster risk correlation coefficient, the driving behavior cluster factor correlation coefficient, the environmental factor cluster factor correlation coefficient and the vehicle state cluster factor correlation coefficient are weightedly calculated to obtain the key risk factor correlation coefficient.

[0030] Optionally, after obtaining the key risk factor information, the factor clusters (driving behavior factor cluster, environmental factor factor cluster, vehicle status factor cluster) in the key risk factor information are evaluated for risk correlation using the standard operational vehicle risk data set; for each factor cluster, the data corresponding to each factor cluster is extracted from the standard operational vehicle risk data set, and based on these data, the risk correlation coefficient of each cluster is calculated using the Pearson correlation coefficient, thereby obtaining the driving behavior cluster risk correlation coefficient, the environmental factor cluster risk correlation coefficient, and the vehicle status cluster risk correlation coefficient; in addition, an intra-cluster correlation evaluation is performed on each risk factor in the driving behavior factor update cluster, the environmental factor factor update cluster, and the vehicle status factor update cluster, that is, the data corresponding to each factor is extracted from the standard operational vehicle risk data set, and the factor correlation coefficient of each factor is calculated using the Pearson correlation coefficient. , and then sort out the factor correlation coefficients belonging to the same factor cluster to obtain the driving behavior cluster factor correlation coefficient, the environmental factor cluster factor correlation coefficient and the vehicle state cluster factor correlation coefficient; then, use the driving behavior cluster risk correlation coefficient, the environmental factor cluster risk correlation coefficient and the vehicle state cluster risk correlation coefficient as the weights of the driving behavior cluster factor correlation coefficient, the environmental factor cluster factor correlation coefficient and the vehicle state cluster factor correlation coefficient respectively, and perform weighted calculation on the driving behavior cluster factor correlation coefficient, the environmental factor cluster factor correlation coefficient and the vehicle state cluster factor correlation coefficient, and then normalize the weighted calculation results by the maximum and minimum values ​​to obtain the key risk factor correlation coefficients, which reflect the importance of each key factor to the overall risk assessment, thereby improving the accuracy and effectiveness of the model.

[0031] The operational vehicle risk data set is trained and optimized based on the risk factor weight allocation information to construct a vehicle risk prediction model.

[0032] In one embodiment, a risk prediction algorithm model is built based on the vehicle risk prediction target and the data characteristic information of the operating vehicle risk data set; then, the risk factor weight distribution information is synchronized to the risk prediction algorithm model, and the operating vehicle risk data set is divided into a training set and a validation set; thereafter, the training set is input into the risk prediction algorithm model for training, and the training results are verified, evaluated, adjusted and optimized through the validation set, thereby constructing a vehicle risk prediction model. This vehicle risk prediction model can perform real-time risk prediction based on new vehicle data (such as driving behavior, environmental factors, and vehicle status, etc.), and output the vehicle risk score, thereby providing reliable support for subsequent risk assessment, insurance pricing, and fleet management.

[0033] In a possible implementation, obtaining the key risk factor correlation coefficient includes: The operational vehicle risk data set is classified and identified according to the key risk factor information to obtain a vehicle key risk factor data set; a risk prediction algorithm model is selected according to the vehicle risk prediction target and the data characteristic information of the vehicle key risk factor data set; the risk prediction algorithm model is used to perform prediction training on the operational vehicle risk data set based on the risk factor weight distribution information to obtain an initial risk prediction model; the initial risk prediction model is verified, evaluated, adjusted and optimized to construct the vehicle risk prediction model.

[0034] Optionally, based on the determined key risk factors, data related to these key factors are extracted from the operational vehicle risk data set, and the extracted data are identified and classified to ensure that each factor data corresponds to the corresponding vehicle risk-related information one-to-one, forming a vehicle key risk factor data set, which contains all factor information closely related to the risk prediction target; then, based on the vehicle risk prediction target, the target variable type is determined. If the target variable is continuous (such as risk score), a regression neural network can be selected; if the target is a classification type (such as whether an accident occurs or not), a classification neural network is used to analyze the vehicle key risk factor data set to determine whether the feature is continuous data or categorical data. If it is categorical data, unique hot encoding or other preprocessing methods are required to enable it to be processed by the neural network. Here, a neural network is used as an example, and random forests, support vector machines, etc. can also be used; for regression tasks such as risk scoring, the output layer can use a linear activation function to predict the risk score. For classification tasks such as whether an accident occurs or not, the Softmax activation function is selected as the output layer for multi-class classification, or the Sigmoid function is used for binary classification; in vehicle risk prediction After the data characteristic information of the target and vehicle key risk factor data sets are analyzed, the corresponding algorithm is selected according to the analysis results to construct a risk prediction algorithm model. For example, a deep neural network is used to construct a risk prediction algorithm model. Subsequently, the risk factor weight allocation information is used to initialize the weight of the risk prediction algorithm model, and the training set divided based on the operating vehicle risk data set is input into the risk prediction algorithm model. The predicted score is obtained through forward propagation, and the loss value between the predicted score and the actual score is calculated through the mean square error. The back propagation algorithm (Backpropagation) and gradient descent (such as Adam optimizer) are used to optimize the parameters of the risk prediction algorithm model. This training process includes multiple iterations. Each iteration adjusts the weight of the risk prediction algorithm model according to the value of the loss function to reduce the prediction error. When the maximum number of training times is reached or the accuracy meets the corresponding expectations, the training process ends and the current risk prediction algorithm model is used as the initial risk prediction model. After that, the initial risk prediction model is verified, evaluated and adjusted using the validation set. When the verification result meets the model performance evaluation indicator set, a vehicle risk prediction model is constructed to provide strong support for subsequent risk management.

[0035] In a possible implementation, the constructing the vehicle risk prediction model includes: According to the vehicle risk prediction target, a model performance evaluation index set is obtained; the initial risk prediction model is verified and evaluated according to the model performance evaluation index set to obtain model prediction performance evaluation parameters; an adjustment strategy analysis is performed on the model prediction performance evaluation parameters to determine the model optimization strategy, and the initial risk prediction model is adjusted and optimized based on the model optimization strategy to obtain the vehicle risk prediction model.

[0036] Optionally, according to the vehicle risk prediction goal, select appropriate performance evaluation indicators to build a model performance evaluation indicator set, which includes mean square error (MSE), mean absolute error (MAE), determination coefficient (R 2 ); Subsequently, the divided validation set is input into the initial risk prediction model to obtain the prediction score, and then the prediction score of the model is compared with the actual score in the validation set, and the performance evaluation index in the model performance evaluation index set is calculated to obtain the model prediction performance evaluation parameter; then, the model prediction performance evaluation parameter is compared with the corresponding threshold in the model performance evaluation index set. If the MSE or MAE is large, it means that the prediction error of the model is large and the model needs to be optimized; if R 2 If the value is low, it means that the model fails to fit the data well and may have an underfitting problem. Based on the comparison of these parameters, the model optimization strategy is determined. For example, if the MSE or MAE is large, the amount of training data can be increased to help the model learn more patterns and features. The number of neural network layers and the number of neurons in each layer can also be increased to improve the model's learning ability. If R 2 If the risk factor is too low, you can reduce the number of layers to prevent the network from being too deep, or use L2 regularization or Dropout to prevent overfitting. Then, use these optimization strategies to adjust the initial risk prediction model and retrain the model to obtain the final vehicle risk prediction model.

[0037] In a possible implementation, the constructing the vehicle risk prediction model includes: Based on the model optimization strategy, the initial risk prediction model is subjected to parameter optimization analysis to obtain a model optimization parameter combination space; a global optimization search is performed within the model optimization parameter combination space to determine the target model optimization parameters, and the target model optimization parameters are used to optimize the initial risk prediction model to obtain the vehicle risk prediction model.

[0038] Optionally, based on the model optimization strategy, the parameters of the initial risk prediction model are first optimized and analyzed to obtain the model optimization parameter combination space. The combination space is a high-dimensional space composed of multiple optimization parameters and their possible values. The value range of each parameter is set according to historical data, experience or prior knowledge. Specifically, key parameters such as learning rate, batch size, number of neural network layers, number of neurons in each layer, etc. are selected as optimization objects. These parameters will affect the training efficiency, prediction accuracy and generalization ability of the model. Therefore, by analyzing the combination of these parameters, the most suitable parameter range for the current task can be determined, laying the foundation for model optimization. After obtaining the model optimization parameter combination space, a global optimization method is used to search for the optimal parameter combination. In order to efficiently find the best combination in the huge parameter space, global optimization techniques such as grid search, random search, Bayesian optimization, etc. are used. Through these methods, all possible parameter combinations can be traversed, or multiple candidate combinations can be randomly selected in the parameter space for evaluation. Each evaluation will train the model and determine the pros and cons of the combination based on the prediction performance on the validation set. After global optimization, It is possible to find an optimal target model optimization parameter combination that provides the best prediction performance on a given data set. After determining the target model optimization parameters, these parameters are applied to the initial risk prediction model for optimization configuration, which includes adjusting the architecture of the neural network, such as modifying the number of network layers, the number of neurons in each layer, selecting a suitable optimizer (such as Adam or SGD), and adjusting hyperparameters such as the learning rate and regularization term. After these optimizations, the model training will be more efficient and accurate, thereby reducing training time and improving prediction accuracy. The optimized model will be retrained on the training set to ensure its generalization ability on unknown data. Finally, the optimized model will be evaluated on the validation set to ensure that it still performs well on unseen data. If the model's prediction performance is improved and there is no sign of overfitting, the current initial risk prediction model will be output as a vehicle risk prediction model and applied to actual risk scoring tasks. Through this series of optimization processes, it is ensured that the model can accurately predict risk scores based on the multi-dimensional characteristics of the vehicle, thereby providing reliable support for fleet management, insurance pricing, and safety decisions.

[0039] The vehicle risk prediction model is used to perform risk prediction on the vehicle multi-dimensional operation data flow, output a vehicle risk prediction score, and perform vehicle risk strategy management based on the vehicle risk prediction score.

[0040] In one embodiment, the constructed vehicle risk prediction model is used to predict the risk of the vehicle's multi-dimensional operation data stream (including driving behavior, environmental factors, vehicle status, etc.), and generate a corresponding vehicle risk prediction score; when predicting risk, the model will automatically calculate a risk score based on the input driving behavior, environmental factors, and vehicle status, combined with the patterns and rules learned during the training process, representing the risk level of the vehicle. The higher the score, the greater the risk, and the higher the probability of potential safety hazards or accidents. Through this risk score, a real-time safety assessment can be provided for each vehicle; then, based on the risk prediction score of the vehicle, risk strategy management is carried out. This stage mainly takes different management and response measures based on the score. For example, for vehicles with high risk scores, safety reminders can be triggered to remind drivers to pay attention to safe driving, and even the fleet management system may be activated to conduct special monitoring or maintenance on the vehicle to reduce potential risks. In addition, based on the risk score, insurance pricing can also be dynamically adjusted. Vehicles with high risk scores may face higher insurance costs, while vehicles with low risk scores enjoy lower premiums. This intelligent risk assessment and management method not only helps fleet managers optimize resource allocation, but also reduces the possibility of accidents and improves overall operational efficiency.

[0041] In a possible implementation, the performing vehicle risk strategy management based on the vehicle risk prediction score includes: A multi-dimensional risk strategy analysis is performed on the vehicle risk prediction score to obtain a vehicle risk management strategy, wherein the vehicle risk management strategy includes safety feedback, fleet management, and insurance pricing, and vehicle risk adaptive management is performed through the vehicle risk management strategy.

[0042] Optionally, after obtaining the risk prediction score of the vehicle, a multi-dimensional risk strategy analysis will be conducted. By comprehensively evaluating the risk data of different dimensions, a specific vehicle risk management strategy will be formulated. The goal of this analysis process is to formulate personalized management measures based on the risk score of the vehicle to ensure the safety of the vehicle and reduce potential risks. Specifically, based on the vehicle risk prediction score, personalized safety feedback will be provided to the driver. For example, if the vehicle risk prediction score is high, the driver will be reminded to improve driving safety and avoid bad driving behaviors such as sudden braking and acceleration. If the score is low, the driver can be reminded to continue to maintain the current safe driving state. By summarizing and analyzing the risk scores of the entire fleet, fleet managers can identify high-risk vehicles and conduct targeted Risk management includes regular inspections of the health of high-risk vehicles, special training to enhance driver safety awareness, and restrictive operation of high-risk vehicles. Low-risk vehicles can maintain normal operations and avoid excessive intervention. Insurance premiums are dynamically adjusted based on vehicle risk prediction scores. Vehicles with high risk scores need to pay higher insurance premiums, while vehicles with low risk scores enjoy discounts or lower premiums. This dynamic insurance pricing can encourage the fleet to adopt safer driving habits, thereby reducing the overall risk of accidents. Based on the formulated risk management strategy, adaptive vehicle risk management is achieved, that is, management measures are automatically adjusted based on the real-time risk score of the vehicle to ensure the optimal allocation of fleet resources, avoid excessive intervention in low-risk vehicles, and improve management efficiency.

[0043] Embodiment 2 is based on the same inventive concept as the method for predicting operating vehicle risk based on multi-factor analysis in the above embodiment. Figure 2As shown, the present application provides an operating vehicle risk prediction platform based on multi-factor analysis, and the platform and method embodiments in the embodiments of the present application are based on the same inventive concept. The platform includes: a data acquisition module 11, which is used to collect and acquire vehicle multi-dimensional operation data streams through a sensor perception network, wherein the sensor perception network includes GPS, cameras, gyroscopes and radars; a risk factor extraction module 12, which is used to extract and acquire vehicle risk factor information, wherein the vehicle risk factor information includes a driving behavior factor set, an environmental factor factor set and a vehicle status factor set; a weight allocation module 13, which is used to crawl and acquire an operating vehicle risk data set based on the vehicle risk factor information, and use the operating vehicle risk data set to weight the vehicle risk factor information to obtain risk factor weight allocation information; a training and tuning module 14, which is used to train and tune the operating vehicle risk data set based on the risk factor weight allocation information to construct a vehicle risk prediction model; a risk management module 15, which is used to use the vehicle risk prediction model to perform risk prediction on the vehicle multi-dimensional operation data stream, output a vehicle risk prediction score, and perform vehicle risk strategy management based on the vehicle risk prediction score.

[0044] Furthermore, the weight distribution module 13 also includes: Acquire a data preprocessing program, use the data preprocessing program to perform standardized preprocessing on the operating vehicle risk data set to obtain a standard operating vehicle risk data set; perform a criticality analysis on the vehicle risk factor information based on the standard operating vehicle risk data set to screen and obtain key risk factor information; use the standard operating vehicle risk data set to perform a risk correlation assessment on the key risk factor information to obtain a key risk factor correlation coefficient; perform weight allocation on the key risk factor information based on the key risk factor correlation coefficient to obtain the risk factor weight allocation information.

[0045] Furthermore, the weight distribution module 13 also includes: According to the vehicle risk factor information, an initial driving behavior factor cluster, an initial environmental factor cluster and an initial vehicle state factor cluster are determined; based on the standard operating vehicle risk data set, the contribution of each risk factor in the initial driving behavior factor cluster, the initial environmental factor cluster and the initial vehicle state factor cluster is evaluated to obtain a driving behavior factor contribution set, an environmental factor contribution set and a vehicle state factor contribution set; based on the driving behavior factor contribution set, the environmental factor contribution set and the vehicle state factor contribution set, the initial driving behavior factor cluster, the initial environmental factor cluster and the initial vehicle state factor cluster are critically screened to determine key risk factor information.

[0046] Furthermore, the weight distribution module 13 also includes: Based on the driving behavior factor contribution set, the environmental factor factor contribution set and the vehicle state factor contribution set, the initial driving behavior factor cluster, the initial environmental factor factor cluster and the initial vehicle state factor cluster are clustered and updated respectively to obtain a driving behavior factor update cluster, an environmental factor factor update cluster and a vehicle state factor update cluster; a preset contribution selection threshold is obtained, and a risk factor cluster distance screening threshold is determined according to the preset contribution selection threshold; based on the risk factor cluster distance screening threshold, the driving behavior factor update cluster, the environmental factor factor update cluster and the vehicle state factor update cluster are screened for key factors in turn to determine key risk factor information.

[0047] Furthermore, the weight distribution module 13 also includes: The standard operational vehicle risk data set is used to perform risk correlation assessment on the driving behavior factor update cluster, environmental factor factor update cluster and vehicle state factor update cluster in the key risk factor information, respectively, to obtain the driving behavior cluster risk correlation coefficient, the environmental factor cluster risk correlation coefficient and the vehicle state cluster risk correlation coefficient; based on the standard operational vehicle risk data set, intra-cluster correlation assessment is performed on each risk factor in the driving behavior factor update cluster, the environmental factor factor update cluster and the vehicle state factor update cluster, respectively, to obtain the driving behavior cluster factor correlation coefficient, the environmental factor cluster factor correlation coefficient and the vehicle state cluster factor correlation coefficient; based on the driving behavior cluster risk correlation coefficient, the environmental factor cluster risk correlation coefficient and the vehicle state cluster risk correlation coefficient, the driving behavior cluster factor correlation coefficient, the environmental factor cluster factor correlation coefficient and the vehicle state cluster factor correlation coefficient are weightedly calculated to obtain the key risk factor correlation coefficient.

[0048] Furthermore, the training and tuning module 14 also includes: The operational vehicle risk data set is classified and identified according to the key risk factor information to obtain a vehicle key risk factor data set; a risk prediction algorithm model is selected according to the vehicle risk prediction target and the data characteristic information of the vehicle key risk factor data set; the risk prediction algorithm model is used to perform prediction training on the operational vehicle risk data set based on the risk factor weight distribution information to obtain an initial risk prediction model; the initial risk prediction model is verified, evaluated, adjusted and optimized to construct the vehicle risk prediction model.

[0049] Furthermore, the training and tuning module 14 also includes: According to the vehicle risk prediction target, a model performance evaluation index set is obtained; the initial risk prediction model is verified and evaluated according to the model performance evaluation index set to obtain model prediction performance evaluation parameters; an adjustment strategy analysis is performed on the model prediction performance evaluation parameters to determine the model optimization strategy, and the initial risk prediction model is adjusted and optimized based on the model optimization strategy to obtain the vehicle risk prediction model.

[0050] Furthermore, the training and tuning module 14 also includes: Based on the model optimization strategy, the initial risk prediction model is subjected to parameter optimization analysis to obtain a model optimization parameter combination space; a global optimization search is performed within the model optimization parameter combination space to determine the target model optimization parameters, and the target model optimization parameters are used to optimize the initial risk prediction model to obtain the vehicle risk prediction model.

[0051] Furthermore, the risk management module 15 also includes: A multi-dimensional risk strategy analysis is performed based on the vehicle risk prediction score to obtain a vehicle risk management strategy, which includes safety feedback, fleet management and insurance pricing, and the vehicle risk adaptive management is performed through the vehicle risk management strategy.

[0052] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0053] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0054] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for predicting operating vehicle risks based on multi-factor analysis, characterized in that: The method comprises: Acquire multi-dimensional vehicle operation data streams through a sensor perception network, wherein the sensor perception network includes GPS, camera, gyroscope and radar; Extracting and obtaining vehicle risk factor information, wherein the vehicle risk factor information includes a driving behavior factor set, an environmental factor factor set, and a vehicle state factor set; Based on the vehicle risk factor information, crawling obtains an operating vehicle risk data set, and using the operating vehicle risk data set to weight the vehicle risk factor information to obtain risk factor weight distribution information; Based on the risk factor weight allocation information, the operating vehicle risk data set is trained and optimized to construct a vehicle risk prediction model; Using the vehicle risk prediction model to perform risk prediction on the vehicle multi-dimensional operation data stream, outputting a vehicle risk prediction score, and performing vehicle risk strategy management based on the vehicle risk prediction score; The obtaining of risk factor weight allocation information includes: Acquire a data preprocessing program, and use the data preprocessing program to perform standardized preprocessing on the operational vehicle risk data set to obtain a standard operational vehicle risk data set; Performing a critical analysis on the vehicle risk factor information based on the standard operating vehicle risk data set, and screening to obtain critical risk factor information; Using the standard operating vehicle risk data set to perform risk correlation assessment on the key risk factor information to obtain a key risk factor correlation coefficient; The key risk factor information is weighted based on the key risk factor correlation coefficient to obtain the risk factor weight distribution information.

2. The method for predicting operating vehicle risks based on multi-factor analysis as claimed in claim 1, characterized in that: The obtaining of key risk factor information includes: Determining an initial driving behavior factor cluster, an initial environmental factor factor cluster, and an initial vehicle state factor cluster according to the vehicle risk factor information; Based on the standard operational vehicle risk data set, the contribution of each risk factor in the initial driving behavior factor cluster, the initial environmental factor factor cluster, and the initial vehicle state factor cluster is evaluated to obtain a driving behavior factor contribution set, an environmental factor contribution set, and a vehicle state factor contribution set; Based on the driving behavior factor contribution set, the environmental factor factor contribution set and the vehicle state factor contribution set, the initial driving behavior factor cluster, the initial environmental factor factor cluster and the initial vehicle state factor cluster are critically screened to determine key risk factor information.

3. The method for predicting operating vehicle risks based on multi-factor analysis as claimed in claim 2, characterized in that: The key risk factor information is determined, including: Based on the driving behavior factor contribution set, the environmental factor factor contribution set and the vehicle state factor contribution set, the initial driving behavior factor cluster, the initial environmental factor factor cluster and the initial vehicle state factor cluster are clustered and updated respectively to obtain a driving behavior factor update cluster, an environmental factor factor update cluster and a vehicle state factor update cluster; Obtaining a preset contribution selection threshold, and determining a risk factor cluster distance screening threshold according to the preset contribution selection threshold; Based on the risk factor cluster distance screening threshold, the driving behavior factor update cluster, the environmental factor factor update cluster and the vehicle status factor update cluster are sequentially screened for key factors to determine key risk factor information.

4. The method for predicting operating vehicle risks based on multi-factor analysis as claimed in claim 3, characterized in that: The key risk factor correlation coefficients are obtained, including: Using the standard operating vehicle risk data set, respectively, the driving behavior factor update cluster, the environmental factor factor update cluster, and the vehicle state factor update cluster in the key risk factor information are evaluated for risk correlation, and a driving behavior cluster risk correlation coefficient, an environmental factor cluster risk correlation coefficient, and a vehicle state cluster risk correlation coefficient are obtained; Based on the standard operational vehicle risk data set, intra-cluster correlation evaluation is performed on each risk factor in the driving behavior factor update cluster, the environmental factor factor update cluster, and the vehicle state factor update cluster to obtain a driving behavior cluster factor correlation coefficient, an environmental factor cluster factor correlation coefficient, and a vehicle state cluster factor correlation coefficient; Based on the driving behavior cluster risk correlation coefficient, the environmental factor cluster risk correlation coefficient and the vehicle status cluster risk correlation coefficient, the driving behavior cluster factor correlation coefficient, the environmental factor cluster factor correlation coefficient and the vehicle status cluster factor correlation coefficient are weightedly calculated to obtain the key risk factor correlation coefficient.

5. The method for predicting operating vehicle risks based on multi-factor analysis as claimed in claim 1, characterized in that: The constructing of the vehicle risk prediction model comprises: Classify and identify the operational vehicle risk data set according to the key risk factor information to obtain a vehicle key risk factor data set; Selecting a risk prediction algorithm model according to a vehicle risk prediction target and data characteristic information of the vehicle key risk factor data set; Using the risk prediction algorithm model to perform prediction training on the operating vehicle risk data set based on the risk factor weight allocation information to obtain an initial risk prediction model; The initial risk prediction model is verified, evaluated, adjusted and optimized to construct the vehicle risk prediction model.

6. The method for predicting operating vehicle risks based on multi-factor analysis as claimed in claim 5, characterized in that: The constructing of the vehicle risk prediction model comprises: According to the vehicle risk prediction target, a model performance evaluation indicator set is obtained; Validating and evaluating the initial risk prediction model according to the model performance evaluation indicator set to obtain model prediction performance evaluation parameters; An adjustment strategy analysis is performed on the model prediction performance evaluation parameters to determine a model optimization strategy, and the initial risk prediction model is adjusted and optimized based on the model optimization strategy to obtain the vehicle risk prediction model.

7. The method for predicting operating vehicle risks based on multi-factor analysis as claimed in claim 6, characterized in that: The constructing of the vehicle risk prediction model comprises: Based on the model optimization strategy, the initial risk prediction model is subjected to parameter optimization analysis to obtain a model optimization parameter combination space; A global optimization is performed in the model optimization parameter combination space to determine the target model optimization parameters, and the target model optimization parameters are used to optimize the initial risk prediction model to obtain the vehicle risk prediction model.

8. The method for predicting operating vehicle risks based on multi-factor analysis as claimed in claim 1, characterized in that: The vehicle risk strategy management based on the vehicle risk prediction score includes: A multi-dimensional risk strategy analysis is performed based on the vehicle risk prediction score to obtain a vehicle risk management strategy, which includes safety feedback, fleet management and insurance pricing, and the vehicle risk adaptive management is performed through the vehicle risk management strategy.

9. An operating vehicle risk prediction platform based on multi-factor analysis, characterized in that: The platform is used to execute the operating vehicle risk prediction method based on multi-factor analysis according to any one of claims 1 to 8, comprising: Data acquisition module: collects and acquires multi-dimensional vehicle operation data streams through a sensor perception network, wherein the sensor perception network includes GPS, camera, gyroscope and radar; Risk factor extraction module: extracts and obtains vehicle risk factor information, wherein the vehicle risk factor information includes a driving behavior factor set, an environmental factor factor set, and a vehicle status factor set; Weight allocation module: crawling to obtain an operating vehicle risk data set based on the vehicle risk factor information, using the operating vehicle risk data set to weight the vehicle risk factor information, and obtaining risk factor weight allocation information; Training and tuning module: training and tuning the operational vehicle risk data set based on the risk factor weight allocation information to build a vehicle risk prediction model; Risk management module: Use the vehicle risk prediction model to predict the risk of the vehicle multi-dimensional operation data flow, output the vehicle risk prediction score, and perform vehicle risk strategy management based on the vehicle risk prediction score.

Images