Artificial intelligence motor vehicle safety performance detection method and system

Abstract

The present invention discloses an artificial intelligence motor vehicle safety performance detection method and system, including constructing a motor vehicle perception network to obtain motor vehicle perception data, determining a driving factor, performing a first edge calculation and a second edge calculation to obtain an environmental collision index and a structural operation index, performing direct warning and marking operations according to driving state data, the environmental collision index and the structural operation index, performing deep collision analysis and cloud out-of-control prediction, performing collision warning and out-of-control warning, constructing a causal graph model according to the cloud out-of-control prediction result, inputting the motor vehicle state data to be detected into the causal graph model to determine the root cause of hidden dangers and intervention indicators, and performing safety performance detection according to the root cause of hidden dangers and intervention indicators. This method can not only improve the efficiency and accuracy of motor vehicle safety performance detection, but also has good interpretability, and can be directly applied to motor vehicle safety performance detection systems.

Description

Technical Field

[0001] The present invention relates to the field of safety detection, and in particular to an artificial intelligence method and system for detecting the safety performance of a motor vehicle. Background Art

[0002] The safety performance of motor vehicles is directly related to the safety of life and property of drivers and passengers, as well as the smoothness of road traffic. With the rapid development of the automobile industry and the increasing attention people pay to travel safety, motor vehicle safety performance testing technology plays an increasingly important role in ensuring road traffic safety. Driven by the wave of intelligence, the use of artificial intelligence technology to improve the accuracy, timeliness and comprehensiveness of motor vehicle safety performance testing has far-reaching significance for promoting the sustainable development of the transportation industry.

[0003] There are many deficiencies in the traditional motor vehicle safety performance detection technology. For example, the traditional detection method often relies on a single data collection method and simple analysis means, which makes it difficult to fully and accurately obtain the operating status information of the motor vehicle. In addition, when faced with a large amount of detection data, the traditional technology lacks efficient data fusion, analysis and processing capabilities, and mainly relies on local computing equipment, and there is an obvious problem of insufficient computing power, resulting in low detection efficiency and inability to make accurate judgments and early warnings on the safety status of the vehicle in a timely manner. The present invention is based on artificial intelligence technology, and multi-source data is integrated and intelligently analyzed to improve the comprehensiveness of data and detection accuracy. It uses edge computing and cloud computing technology to improve data processing and early warning speed, and ensures the efficiency of deep data mining while reducing cloud computing pressure, effectively improving detection efficiency. This not only helps to ensure the driving safety of motor vehicles and reduce the incidence of traffic accidents, but also can provide a scientific basis for vehicle maintenance and extend the service life of vehicles, which has important practical significance and application value. Summary of the invention

[0004] The purpose of the present invention is to provide an artificial intelligence motor vehicle safety performance detection method and system.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions: The present invention comprises the following steps: Constructing a motor vehicle perception network to obtain motor vehicle perception data; the motor vehicle perception data includes environmental data and state data; the state data includes driving state data and structural state data; Directly warning and marking the driving state data, dividing the driving state data, and determining a driving factor according to the division result; the driving factor includes a first driving factor and a second driving factor; Performing a first edge calculation on the environmental data to obtain an environmental collision index, and performing direct warning and marking operations according to the environmental collision index; Performing a deep collision analysis according to the environmental collision index and the driving factor, and performing a collision warning and a safety performance check according to the deep collision analysis result; Performing a second edge calculation on the structure state data to obtain a structure operation index, performing a direct warning and marking operation according to the structure operation index, and performing a cloud out-of-control prediction and out-of-control warning according to the structure operation index and the driving factor; A causal graph model is constructed based on the cloud out-of-control prediction results, the state data of the motor vehicle to be tested is input into the causal graph model to determine the root causes of hidden dangers and intervention indicators, and safety performance testing is performed based on the root causes of hidden dangers and intervention indicators.

[0006] Furthermore, the method for obtaining the motor vehicle perception data includes: The motor vehicle perception network includes OBD terminals, V2X communication modules and vehicle-mounted sensor systems; Directly collect real-time vehicle status data through the OBD terminal; Exchange real-time traffic and weather information with cloud servers through the V2X communication module; The vehicle-mounted sensor system includes a laser radar and a high-definition camera. The high-definition camera collects vehicle environment images, and the laser radar collects vehicle environment distance information; Traffic weather information, vehicle environment image and vehicle environment distance information are combined into environmental data.

[0007] Furthermore, the method for determining the driving factor includes: Determine a first warning threshold, when any driving state data is greater than the first warning threshold, directly warn according to the corresponding driving state data; determine a first marking threshold according to the first warning threshold, when all driving state data are less than the first warning threshold and any driving state data is greater than the first marking threshold, mark the driving state data to obtain a driving hazard marking value; when all driving state data are less than the first marking threshold, determine that the motor vehicle is safe to drive; When any driving state data is greater than the first marking threshold, the driving state data is divided into first driving data and second driving data by using the clustering method. The first driving data is input into the first driving function to obtain the first driving factor, and the second driving data is input into the second driving function to obtain the second driving factor. The expression is: in is the first travel factor, is the second travel factor, is the speed weight, To start the weight, is the current driving speed of the vehicle, is the current acceleration of the vehicle, is the maximum vehicle acceleration, is the current fuel consumption of the vehicle, is the standard fuel consumption of motor vehicles, is the current number of passengers in the motor vehicle, is the rated number of passengers of a motor vehicle, is the motor vehicle gear index, is the current gear position of the motor vehicle, The maximum gear position of the motor vehicle, is the braking weight, To stabilize the weight, To control the weight, is the fatigue weight, is the current braking strength of the motor vehicle, is the friction parameter of the motor vehicle, is the maximum braking strength of the motor vehicle, For system update level, is the system error rate, is the system correlation, is the system response time, for Random numbers, is the motor vehicle control intensity, is the current accumulated running time of the motor vehicle, The accumulated running time of the motor vehicle is stipulated. The accumulated running time of the motor vehicle from the last maintenance; When all driving status data are less than the first marking threshold, the first driving factor and the second driving factor are set to 1.

[0008] Furthermore, the method for performing direct warning and marking operations according to the environmental collision index includes: The first edge computing includes image processing and environmental data correction, and the specific steps are: The vehicle environment image is processed by a CNN lightweight model to obtain vehicle environment features; the vehicle environment features include a first vehicle environment feature, a second vehicle environment feature, a third vehicle environment feature and a fourth vehicle environment feature; the first vehicle environment feature is related to the road meteorological environment; the second vehicle environment feature is related to the distance between the road and obstacles; the third vehicle environment feature is related to traffic signs and danger sources; the fourth vehicle environment feature is related to external defects of the vehicle; The CNN lightweight model runs in real time on the edge computing node, and specifically includes an input layer, a convolution layer, a pooling layer, and an output layer. The model parameters and the amount of calculation are reduced through model pruning and quantization. The convolution layer includes depth convolution and point-by-point convolution. The input layer and the convolution layer are connected; the output layer includes 4 output nodes, uses the Softmax function to convert the output into a probability distribution, uses the cross entropy loss function to evaluate the difference between the predicted probability and the true label, and outputs the classification probability vector of the vehicle environment feature; Taking the intersection of the first vehicle environment feature and the traffic weather information as the road meteorological feature, and fusing the road meteorological feature with the first vehicle environment feature to obtain the first road meteorological environment feature; Performing data comparison between the second vehicle environment feature and the vehicle environment distance information to obtain a distance difference, performing feature crossover between the distance difference and the second vehicle environment feature to obtain a first road obstacle distance feature, and performing feature fusion between the first road obstacle distance feature and the vehicle environment distance information to obtain a second road obstacle distance feature; Comparing the fourth vehicle environment feature with the standard vehicle appearance feature to obtain a vehicle exterior defect feature, and obtaining a vehicle appearance evaluation index according to the vehicle exterior defect feature; The environmental collision index is composed of a first road meteorological environment characteristic, a second road obstacle distance characteristic and a third vehicle environment characteristic; A second warning threshold is determined. When any environmental collision indicator is greater than the second warning threshold, a direct warning is issued according to the corresponding environmental collision indicator. A second marking threshold is determined according to the second warning threshold. When all environmental collision indicators are less than the second warning threshold and any environmental collision indicator is greater than the second marking threshold, the environmental collision indicator is marked to obtain a collision hazard marking value. When all environmental collision indicators are less than the second marking threshold, the motor vehicle is determined to be safe to drive.

[0009] Furthermore, the method for performing the collision warning and safety performance inspection includes: The deep collision analysis includes cloud collision prediction and edge collision simulation; The specific operation of the edge collision simulation is as follows: constructing an edge collision simulation model according to the environmental collision index, the first driving data and the KITTI map data, and inputting the data to be simulated into the edge collision simulation model to generate a potential collision scene; the edge collision simulation model runs in real time on the edge computing node, including a GAN adversarial network and transfer learning; the GAN adversarial network uses a generator to receive KITTI map data and environmental collision indexes, and outputs a virtual road condition image, uses a discriminator to optimize the generator according to the first driving data discrimination result, and uses a perceptual loss function to improve the generation quality; the transfer learning migrates the virtual road condition image to the real scene, and uses a domain adversarial network to reduce the difference between virtual data and real data; When any environmental collision index is greater than the second marking threshold, a cloud collision prediction is performed: a cloud collision prediction model is constructed according to the environmental collision index, the first driving factor and the second driving factor, and the data to be predicted is input into the cloud collision prediction model to obtain a collision prediction result; the cloud collision prediction model runs in real time on the cloud platform, including a BP neural network and a support vector machine SVM, the BP neural network performs nonlinear regression on the input data and outputs a collision probability, and the vector machine SVM generates a hyperplane according to the input data to divide the collision category; The edge collision simulation and cloud collision prediction results are weightedly fused to output the fused collision category and fused collision probability. When the fused collision probability is greater than the collision warning probability threshold, a collision warning is issued, and the driving hazard mark value, collision hazard mark value and corresponding fused collision category are extracted. The safety performance of the motor vehicle is checked according to the driving hazard mark value and collision hazard mark value.

[0010] Furthermore, the method for performing the cloud out-of-control prediction and out-of-control warning includes: The hierarchical clustering method is used to divide the structural state data to obtain the component operation state data and maintenance log data; the maintenance log data includes the component quality parameters and the corresponding maintenance date; Determine the dynamic safety index of components based on component quality parameters and corresponding maintenance dates, and use the dynamic safety index of components to correct component operation status data to obtain structural operation indicators; Determine a third warning threshold, when any structural operation index is greater than the third warning threshold, issue a direct warning according to the corresponding structural operation index; determine a third marking threshold according to the third warning threshold, when all structural operation indexes are less than the third warning threshold and any structural operation index is greater than the third marking threshold, mark the structural operation index to obtain a risk of loss of control marking value; when all structural operation indexes are less than the third marking threshold, determine that the motor vehicle is safe to drive; When any structural operation index is greater than the third marking threshold, a cloud out-of-control prediction model is constructed according to the structural operation index, the first driving factor, the second driving factor and the vehicle appearance evaluation index, and the data to be predicted is input into the cloud out-of-control prediction model to obtain an out-of-control prediction result; the cloud collision prediction model runs in real time on the cloud platform, including a BP neural network and a support vector machine SVM, the BP neural network performs nonlinear regression on the input data and outputs the out-of-control probability, and the vector machine SVM generates a hyperplane according to the input data to divide the out-of-control category; When the probability of out-of-control is greater than the out-of-control warning probability threshold, an out-of-control warning is issued, and the out-of-control hidden danger mark value and the corresponding out-of-control category are extracted.

[0011] Furthermore, the method for determining the root cause of the hidden danger and the intervention indicator includes: The first hidden danger mark value, the out-of-control category and the out-of-control probability form a causal set, and the causal set is divided into a training set and a test set according to a ratio of 7:3; the first hidden danger mark value includes a driving hidden danger mark value and an out-of-control hidden danger mark value; A causal graph model is constructed, wherein the causal graph model includes a PC algorithm, a causal attention module and a Do-Calculus algorithm; the constructed causal graph model is run in real time on a cloud platform; the PC algorithm is used to infer a causal graph structure according to a causal set; the causal attention module is used to annotate the direction and weight of the causal edge to obtain a causal graph model; the Do-Calculus algorithm is used to calculate the intervention effect to determine the intervention index, which is used to infer the root cause of the hidden danger; the intervention index represents the first hidden danger mark value that intervenes in a certain type of out-of-control effect; the hidden danger root cause represents the relationship between the first hidden danger mark value that causes a certain type of out-of-control situation and the corresponding hidden danger level; The importance of the causal attention module is expressed as: in The first hidden danger mark value Corresponding out-of-control category The original importance of attention, The first hidden danger mark value Corresponding out-of-control category The normalized importance of attention, The first hidden danger mark value Corresponding out-of-control category The probabilistic average causal effect of The first hidden danger mark value Corresponding out-of-control category The expected average causal effect of Out of control category Based on the shared weights of the probabilistic average causal effects, Out of control category Based on the shared weights of the expected average causal effect, is the regularization strength, Out of control category The probability of losing control, is the number of out-of-control categories, is the standard deviation of the causal association strength; The F1 score was used to evaluate the prediction performance of the model, and the test set was used to verify the accuracy of the causal graph model; The state data of the motor vehicle to be tested is input into the cause-effect diagram model to determine the root cause of the hidden danger and the intervention index, and the safety performance test is performed according to the root cause of the hidden danger and the intervention index.

[0012] Second, the AI-based motor vehicle safety performance testing system includes: Perception network module: used to construct a motor vehicle perception network, and use the motor vehicle perception network to obtain motor vehicle perception data; Edge computing module: used to divide the driving state data and determine the driving factor according to the division result; used to perform a first edge computing on the environmental data to obtain an environmental collision index; used to perform a second edge computing on the structural state data to obtain a structural operation index; Cloud computing module: used for performing in-depth collision analysis according to the environmental collision index and the driving factor; and performing cloud out-of-control prediction according to the structural operation index and the driving factor; Causal graph model module: used to construct a causal graph model according to the cloud out-of-control prediction results, and input the state data of the motor vehicle to be detected into the causal graph model to determine the root cause of the hidden danger and the intervention indicator; Intelligent supervision module: used to store, view and manage the driving status data, driving factors, collision indicators and structural operation indicators of motor vehicles, perform direct warning and marking operations based on the driving status data, the environmental collision indicators and the structural operation indicators, perform collision warnings based on the deep collision analysis results, perform out-of-control warnings based on the cloud out-of-control prediction results, and perform safety performance detection based on the root causes of hidden dangers and intervention indicators.

[0013] The beneficial effects of the present invention are: The present invention is an artificial intelligence motor vehicle safety performance detection method and system. Compared with the prior art, the present invention has the following technical effects: The present invention can improve the data preprocessing capabilities and enhance the model adaptability in motor vehicle safety performance testing through image processing, data fusion and correction, data early warning, construction of driving factors, virtual road condition simulation and model construction steps, thereby improving the efficiency of motor vehicle safety performance testing, and optimizing motor vehicle safety performance testing technology, which can greatly save resources, improve work efficiency, and achieve motor vehicle safety performance testing, providing a more reliable guarantee for the safe operation of motor vehicles. It is of great significance to the further development of intelligent transportation systems, and can adapt to different motor vehicle safety performance testing systems, The motor vehicle safety performance testing needs of different users, and have a certain degree of universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The present invention is a flowchart of the steps of the artificial intelligence motor vehicle safety performance detection method. DETAILED DESCRIPTION

[0015] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0016] The artificial intelligence motor vehicle safety performance detection method and system of the present invention comprises the following steps: like Figure 1 As shown, in this embodiment, the following steps are included: Constructing a motor vehicle perception network to obtain motor vehicle perception data; the motor vehicle perception data includes environmental data and state data; the state data includes driving state data and structural state data; Directly warning and marking the driving state data, dividing the driving state data, and determining a driving factor according to the division result; the driving factor includes a first driving factor and a second driving factor; Performing a first edge calculation on the environmental data to obtain an environmental collision index, and performing direct warning and marking operations according to the environmental collision index; Performing a deep collision analysis according to the environmental collision index and the driving factor, and performing a collision warning and a safety performance check according to the deep collision analysis result; Performing a second edge calculation on the structure state data to obtain a structure operation index, performing a direct warning and marking operation according to the structure operation index, and performing a cloud out-of-control prediction and out-of-control warning according to the structure operation index and the driving factor; A causal graph model is constructed based on the cloud out-of-control prediction results, the state data of the motor vehicle to be tested is input into the causal graph model to determine the root causes of hidden dangers and intervention indicators, and safety performance testing is performed based on the root causes of hidden dangers and intervention indicators.

[0017] In this embodiment, the method for obtaining the motor vehicle perception data includes: The motor vehicle perception network includes OBD terminals, V2X communication modules and vehicle-mounted sensor systems; Directly collect real-time vehicle status data through the OBD terminal; Exchange real-time traffic and weather information with cloud servers through the V2X communication module; The vehicle-mounted sensor system includes a laser radar and a high-definition camera. The high-definition camera collects vehicle environment images, and the laser radar collects vehicle environment distance information; Combining traffic weather information, vehicle environment image and vehicle environment distance information into environmental data; In the actual evaluation, a motor vehicle perception network was built to collect driving status data (current vehicle speed 80km / h, current acceleration 1.5m / s 2 , maximum vehicle acceleration 3m / s 2, the current fuel consumption of the motor vehicle is 8L / 100km, the standard fuel consumption of the motor vehicle is 7L / 100km, the current number of passengers in the motor vehicle is 4, the rated number of passengers in the motor vehicle is 5, the current gear of the motor vehicle is 4, the maximum gear is 6, the current / maximum braking intensity is 0.8 / 1, the friction parameter is 0.7, the system update degree is 0.9, the system error rate is 0.05%, the system correlation is 0.8, the system response time is 0.5s, the motor vehicle control intensity is 0.6, the current cumulative running time of the motor vehicle is 100h, the prescribed cumulative running time of the motor vehicle is 500h, the cumulative running time of the motor vehicle from the last warranty) and the structural status Status data (1. Component operation status: left front tire pressure is 2.25bar, other tire pressures are 2.33bar, airbag pressure is 1.8MPa, engine water temperature is 95℃, speed is 2000r / min, exhaust parameters are normal, seat belts are locked, windows are closed; 2. Maintenance log: unified maintenance 3 months ago, the maintenance content is to replace the engine oil and filter, and routine inspection of other components, the left front / left rear / right front / right rear tire wear index is 80 / 91 / 92 / 91, the engine quality parameters are consistent with the model and performance parameters, and the airbag quality parameters are consistent with the production batch and pressure test standards); The traffic information obtained through the V2X communication module is light rain / visibility 500m, and the distance information is collected through the on-board laser radar (obstacle in front 12m, vehicle behind 20m, streetlight in front left 2.8m, guardrail on the right 1.5m / pedestrian 4.2m).

[0018] In this embodiment, the method for determining the driving factor includes: Determine a first warning threshold, when any driving state data is greater than the first warning threshold, directly warn according to the corresponding driving state data; determine a first marking threshold according to the first warning threshold, when all driving state data are less than the first warning threshold and any driving state data is greater than the first marking threshold, mark the driving state data to obtain a driving hazard marking value; when all driving state data are less than the first marking threshold, determine that the motor vehicle is safe to drive; When any driving state data is greater than the first marking threshold, the driving state data is divided into first driving data and second driving data by using the clustering method. The first driving data is input into the first driving function to obtain the first driving factor, and the second driving data is input into the second driving function to obtain the second driving factor. The expression is: in is the first travel factor, is the second travel factor, is the speed weight, To start the weight, is the current driving speed of the vehicle, is the current acceleration of the vehicle, is the maximum vehicle acceleration, is the current fuel consumption of the vehicle, is the standard fuel consumption of motor vehicles, is the current number of passengers in the motor vehicle, is the rated number of passengers of a motor vehicle, is the motor vehicle gear index, is the current gear position of the motor vehicle, The maximum gear position of the motor vehicle, is the braking weight, To stabilize the weight, To control the weight, is the fatigue weight, is the current braking strength of the vehicle, is the friction parameter of the motor vehicle, is the maximum braking strength of the motor vehicle, For system update level, is the system error rate, is the system correlation, is the system response time, for Random numbers, is the motor vehicle control intensity, is the current accumulated running time of the motor vehicle, The accumulated running time of the motor vehicle is stipulated. The accumulated running time of the motor vehicle from the last maintenance; When all driving status data are less than the first marking threshold, the first driving factor and the second driving factor are set to 1; In the actual evaluation, the first driving data includes driving, acceleration, fuel consumption, gear position, and number of passengers, and the second driving data includes braking parameters, handling parameters, stability parameters, and fatigue parameters; Direct warning, marking and driving factor calculation operations on the local vehicle side: If the current fuel consumption of 8L / 100km is greater than the corresponding first warning threshold of 7.5L / 100km and the first marking threshold of 7L / 100km, and the remaining driving data are all less than the corresponding first warning threshold, a warning is directly issued, and the warning content is excessive fuel consumption, and all driving data are marked to obtain driving hazard marking values; Take speed weight , launch weight , Braking weight , stable weight , control weight 3. Fatigue Weight 7, Calculate the first travel factor =1.0691, the second travel factor =1.0906.

[0019] In this embodiment, the method for performing direct warning and marking operations according to the environmental collision index includes: The first edge computing includes image processing and environmental data correction, and the specific steps are: The vehicle environment image is processed by a CNN lightweight model to obtain vehicle environment features; the vehicle environment features include a first vehicle environment feature, a second vehicle environment feature, a third vehicle environment feature and a fourth vehicle environment feature; the first vehicle environment feature is related to the road meteorological environment; the second vehicle environment feature is related to the distance between the road and obstacles; the third vehicle environment feature is related to traffic signs and danger sources; the fourth vehicle environment feature is related to external defects of the vehicle; The CNN lightweight model runs in real time on the edge computing node, and specifically includes an input layer, a convolution layer, a pooling layer, and an output layer. Model pruning and quantization are used to reduce the number of model parameters and the amount of calculation. The convolution layer includes deep convolution and point-by-point convolution, and a residual network is used to connect the input layer and the convolution layer. The output layer includes 4 output nodes, and the Softmax function is used to convert the output into a probability distribution. The cross entropy loss function is used to evaluate the difference between the predicted probability and the true label, and the classification probability vector of the vehicle environment feature is output. Taking the intersection of the first vehicle environment feature and the traffic weather information as the road meteorological feature, and fusing the road meteorological feature with the first vehicle environment feature to obtain the first road meteorological environment feature; Performing data comparison between the second vehicle environment feature and the vehicle environment distance information to obtain a distance difference, performing feature crossover between the distance difference and the second vehicle environment feature to obtain a first road obstacle distance feature, and performing feature fusion between the first road obstacle distance feature and the vehicle environment distance information to obtain a second road obstacle distance feature; Comparing the fourth vehicle environment feature with the standard vehicle appearance feature to obtain a vehicle exterior defect feature, and obtaining a vehicle appearance evaluation index according to the vehicle exterior defect feature; The environmental collision index is composed of a first road meteorological environment characteristic, a second road obstacle distance characteristic and a third vehicle environment characteristic; Determine a second warning threshold, when any environmental collision indicator is greater than the second warning threshold, issue a direct warning according to the corresponding environmental collision indicator; determine a second marking threshold according to the second warning threshold, when all environmental collision indicators are less than the second warning threshold and any environmental collision indicator is greater than the second marking threshold, mark the environmental collision indicator to obtain a collision hazard marking value; when all environmental collision indicators are less than the second marking threshold, determine that the motor vehicle is safe to drive; In the actual evaluation, CNN processing is performed on the local vehicle to obtain environmental features: the first vehicle environmental feature (slippery road 0.6, fine particles 50μg / m 3 , visibility 600m, moderate rain), second vehicle environment characteristics (front / stationary / 10m, rear / moving / 20m, left / stationary / 3m, right / moving / 4m, 5 surrounding objects), third vehicle environment characteristics (2 traffic signs / ordinary signal indications, no dangerous sources), fourth vehicle environment characteristics (no obvious scratches on the exterior, lights, windows and other parts intact, tire wear / similarity to the original tire 0.9); The first edge computing is performed on the local vehicle: the first road meteorological environment characteristics (slippery road 0.6, fine particles 50μg / m 3 , visibility 600m, moderate rain, rainfall 5mm / h), the second road obstacle distance characteristics (front / stationary / obstacle / 12m, rear / moving / vehicle / 20m, left / stationary / streetlight / 3m, right / moving / pedestrian / 4m, right / stationary / railing / 1.5m, 5 surrounding objects), vehicle appearance evaluation index 0.98; Among them, all environmental collision indicators are less than the corresponding second warning threshold. At this time, the road slippery 0.6 is greater than the second marking threshold 0.5, and the distance of the obstacle ahead 12m is greater than the marking threshold 20m. All environmental collision indicators are marked to obtain the collision hazard marking value.

[0020] In this embodiment, the method for performing the collision warning and safety performance inspection includes: The deep collision analysis includes cloud collision prediction and edge collision simulation; The specific operation of the edge collision simulation is as follows: constructing an edge collision simulation model according to the environmental collision index, the first driving data and the KITTI map data, and inputting the data to be simulated into the edge collision simulation model to generate a potential collision scene; the edge collision simulation model runs in real time on the edge computing node, including a GAN adversarial network and transfer learning; the GAN adversarial network uses a generator to receive KITTI map data and environmental collision indexes, and outputs a virtual road condition image, uses a discriminator to optimize the generator according to the first driving data discrimination result, and uses a perceptual loss function to improve the generation quality; the transfer learning migrates the virtual road condition image to the real scene, and uses a domain adversarial network to reduce the difference between virtual data and real data; When any environmental collision index is greater than the second marking threshold, a cloud collision prediction is performed: a cloud collision prediction model is constructed according to the environmental collision index, the first driving factor and the second driving factor, and the data to be predicted is input into the cloud collision prediction model to obtain a collision prediction result; the cloud collision prediction model runs in real time on the cloud platform, including a BP neural network and a support vector machine SVM, the BP neural network performs nonlinear regression on the input data and outputs a collision probability, and the vector machine SVM generates a hyperplane according to the input data to divide the collision category; The edge collision simulation and cloud collision prediction results are weightedly fused to output the fused collision category and fused collision probability. When the fused collision probability is greater than the collision warning probability threshold, a collision warning is issued, and the driving hazard mark value, collision hazard mark value and corresponding fused collision category are extracted. The safety performance of the motor vehicle is checked according to the driving hazard mark value and collision hazard mark value. In the actual evaluation, edge collision simulation is performed on the local vehicle side, and cloud collision prediction is performed on the cloud data platform. The edge collision simulation results include the situation where the vehicle skids and collides with the right guardrail. The cloud collision prediction results are (1. The vehicle's braking distance is too short and collides with the obstacle in front / probability 40%, 2. The vehicle skids and collides with the right guardrail / probability 60%). The collision category and probability are determined as follows: the vehicle skids and collides with the right guardrail / probability 60%, which is greater than the collision warning probability threshold of 50%. A collision warning is issued, and the braking system and vehicle tire pattern are checked in time.

[0021] In this embodiment, the method for performing the cloud out-of-control prediction and out-of-control warning includes: The hierarchical clustering method is used to divide the structural state data to obtain the component operation state data and maintenance log data; the maintenance log data includes the component quality parameters and the corresponding maintenance date; Determine the dynamic safety index of components based on component quality parameters and corresponding maintenance dates, and use the dynamic safety index of components to correct component operation status data to obtain structural operation indicators; Determine a third warning threshold, when any structural operation index is greater than the third warning threshold, issue a direct warning according to the corresponding structural operation index; determine a third marking threshold according to the third warning threshold, when all structural operation indexes are less than the third warning threshold and any structural operation index is greater than the third marking threshold, mark the structural operation index to obtain a risk of loss of control marking value; when all structural operation indexes are less than the third marking threshold, determine that the motor vehicle is safe to drive; When any structural operation index is greater than the third marking threshold, a cloud out-of-control prediction model is constructed according to the structural operation index, the first driving factor, the second driving factor and the vehicle appearance evaluation index, and the data to be predicted is input into the cloud out-of-control prediction model to obtain an out-of-control prediction result; the cloud collision prediction model runs in real time on the cloud platform, including a BP neural network and a support vector machine SVM, the BP neural network performs nonlinear regression on the input data and outputs the out-of-control probability, and the vector machine SVM generates a hyperplane according to the input data to divide the out-of-control category; When the probability of out-of-control is greater than the out-of-control warning probability threshold, an out-of-control warning is issued, and the out-of-control hidden danger mark value and the corresponding out-of-control category are extracted; In the actual evaluation, the second edge calculation is performed on the local side of the vehicle to obtain the structural operation indicators: (tires: left front-tire pressure 2.25 bar-operation index 0.89, right front / left front / left rear-tire pressure 2.33 bar-operation index 0.95, engine / water temperature 95℃ / 2000 rpm / operation index 0.91, seat belt / locked*4 / normal / operation index 0.99, window / closed / operation index 0.99, exhaust parameters / normal / operation index 0.99, airbag / 1.8~1.75MPa / operation index 0.97); The difference between the calculated 0.9 and the left front tire operating index 0.89 is greater than the third warning threshold 0, and a warning is directly issued, and the warning content is that the tire pressure is too low; Cloud out-of-control prediction is performed on the cloud data platform, and the prediction result shows that the motor vehicle is slightly out of control, with a corresponding probability of 0.2.

[0022] In this embodiment, the method for determining the root cause of the hidden danger and the intervention indicator includes: The first hidden danger mark value, the out-of-control category and the out-of-control probability form a causal set, and the causal set is divided into a training set and a test set according to a ratio of 7:3; the first hidden danger mark value includes a driving hidden danger mark value and an out-of-control hidden danger mark value; A causal graph model is constructed, wherein the causal graph model includes a PC algorithm, a causal attention module and a Do-Calculus algorithm; the constructed causal graph model is run in real time on a cloud platform; the PC algorithm is used to infer a causal graph structure according to a causal set; the causal attention module is used to annotate the direction and weight of the causal edge to obtain a causal graph model; the Do-Calculus algorithm is used to calculate the intervention effect to determine the intervention index, which is used to infer the root cause of the hidden danger; the intervention index represents the first hidden danger mark value that intervenes in a certain type of out-of-control effect; the hidden danger root cause represents the relationship between the first hidden danger mark value that causes a certain type of out-of-control situation and the corresponding hidden danger level; The importance of the causal attention module is expressed as: in The first hidden danger mark value Corresponding out-of-control category The original importance of attention, The first hidden danger mark value Corresponding out-of-control category The normalized importance of attention, The first hidden danger mark value Corresponding out-of-control category The probabilistic average causal effect of The first hidden danger mark value Corresponding out-of-control category The expected average causal effect of Out of control category Based on the shared weights of the probabilistic average causal effects, Out of control category Based on the shared weights of the expected average causal effect, is the regularization strength, Out of control category The probability of losing control, is the number of out-of-control categories, is the standard deviation of the causal association strength; The F1 score was used to evaluate the prediction performance of the model, and the test set was used to verify the accuracy of the causal graph model; Inputting the state data of the motor vehicle to be tested into the cause-effect diagram model to determine the root cause of the hidden danger and the intervention index, and performing safety performance testing according to the root cause of the hidden danger and the intervention index; In the actual assessment, the cloud out of control prediction result (minor out of control / probability 0.2) was causally traced based on the causal graph model, and the root cause of the hidden danger was determined to be "insufficient left tire pressure", and the intervention indicators were "excessive fuel consumption" and "brake system performance". The tire pressure, tire wear, engine and braking system were tested based on the root cause of the hidden danger and the intervention indicators.

[0023] Second, the AI-based motor vehicle safety performance testing system includes: Perception network module: used to construct a motor vehicle perception network, and use the motor vehicle perception network to obtain motor vehicle perception data; Edge computing module: used to divide the driving state data and determine the driving factor according to the division result; used to perform a first edge computing on the environmental data to obtain an environmental collision index; used to perform a second edge computing on the structural state data to obtain a structural operation index; Cloud computing module: used for performing in-depth collision analysis according to the environmental collision index and the driving factor; and performing cloud out-of-control prediction according to the structural operation index and the driving factor; Causal graph model module: used to construct a causal graph model according to the cloud out-of-control prediction results, and input the state data of the motor vehicle to be detected into the causal graph model to determine the root cause of the hidden danger and the intervention indicator; Intelligent supervision module: used to store, view and manage the driving status data, driving factors, collision indicators and structural operation indicators of motor vehicles, perform direct warning and marking operations based on the driving status data, the environmental collision indicators and the structural operation indicators, perform collision warnings based on the deep collision analysis results, perform out-of-control warnings based on the cloud out-of-control prediction results, and perform safety performance detection based on the root causes of hidden dangers and intervention indicators.

[0024] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An artificial intelligence motor vehicle safety performance detection method, characterized in that: The following steps are involved: S1. Constructing a motor vehicle perception network to obtain motor vehicle perception data; the motor vehicle perception data includes environmental data and state data; the state data includes driving state data and structural state data; S2, performing direct warning and marking operations on the driving state data, dividing the driving state data, and determining a driving factor according to the division result; the driving factor includes a first driving factor and a second driving factor; S3, performing a first edge calculation on the environmental data to obtain an environmental collision index, and performing direct warning and marking operations according to the environmental collision index; S4, performing a deep collision analysis according to the environmental collision index and the driving factor, and performing a collision warning and a safety performance check according to the deep collision analysis result; S5. Perform a second edge calculation on the structure state data to obtain a structure operation index, perform a direct warning and marking operation according to the structure operation index, and perform a cloud out-of-control prediction and out-of-control warning according to the structure operation index and the driving factor; S6. Construct a causal graph model based on the cloud out-of-control prediction result, input the state data of the motor vehicle to be tested into the causal graph model to determine the root cause of the hidden danger and the intervention index, and perform safety performance testing based on the root cause of the hidden danger and the intervention index.

2. The artificial intelligence motor vehicle safety performance detection method according to claim 1 is characterized in that: The method for obtaining the motor vehicle perception data comprises: The motor vehicle perception network includes OBD terminals, V2X communication modules and vehicle-mounted sensor systems; Directly collect real-time vehicle status data through the OBD terminal; Exchange real-time traffic and weather information with cloud servers through the V2X communication module; The vehicle-mounted sensor system includes a laser radar and a high-definition camera. The high-definition camera collects vehicle environment images, and the laser radar collects vehicle environment distance information; Traffic weather information, vehicle environment image and vehicle environment distance information are combined into environmental data.

3. The artificial intelligence motor vehicle safety performance detection method according to claim 1 is characterized in that: The method for determining the driving factor includes: Determine a first warning threshold, when any driving state data is greater than the first warning threshold, directly warn according to the corresponding driving state data; determine a first marking threshold according to the first warning threshold, when all driving state data are less than the first warning threshold and any driving state data is greater than the first marking threshold, mark the driving state data to obtain a driving hazard marking value; when all driving state data are less than the first marking threshold, determine that the motor vehicle is safe to drive; When any driving state data is greater than the first marking threshold, the driving state data is divided into first driving data and second driving data by using the clustering method. The first driving data is input into the first driving function to obtain the first driving factor, and the second driving data is input into the second driving function to obtain the second driving factor. The expression is: in is the first travel factor, is the second travel factor, is the speed weight, To start the weight, is the current driving speed of the vehicle, is the current acceleration of the vehicle, is the maximum vehicle acceleration, is the current fuel consumption of the vehicle, is the standard fuel consumption of motor vehicles, is the current number of passengers in the motor vehicle, is the rated number of passengers of a motor vehicle, is the motor vehicle gear index, is the current gear position of the motor vehicle, The maximum gear position of the motor vehicle, is the braking weight, To stabilize the weight, To control the weight, is the fatigue weight, is the current braking strength of the motor vehicle, is the friction parameter of the motor vehicle, is the maximum braking strength of the motor vehicle, For system update level, is the system error rate, is the system correlation, is the system response time, for Random numbers, is the motor vehicle control intensity, is the current accumulated running time of the motor vehicle, The accumulated running time of the motor vehicle is stipulated. The accumulated running time of the motor vehicle from the last maintenance; When all driving status data are less than the first marking threshold, the first driving factor and the second driving factor are set to 1.

4. The artificial intelligence motor vehicle safety performance detection method according to claim 1 is characterized in that: The method for performing direct warning and marking operations according to the environmental collision index comprises: The first edge computing includes image processing and environmental data correction, and the specific steps are: The vehicle environment image is processed by a CNN lightweight model to obtain vehicle environment features; the vehicle environment features include a first vehicle environment feature, a second vehicle environment feature, a third vehicle environment feature and a fourth vehicle environment feature; the first vehicle environment feature is related to the road meteorological environment; the second vehicle environment feature is related to the distance between the road and obstacles; the third vehicle environment feature is related to traffic signs and danger sources; the fourth vehicle environment feature is related to external defects of the vehicle; The CNN lightweight model runs in real time on the edge computing node, and specifically includes an input layer, a convolution layer, a pooling layer, and an output layer. Model pruning and quantization are used to reduce the number of model parameters and the amount of calculation. The convolution layer includes deep convolution and point-by-point convolution, and a residual network is used to connect the input layer and the convolution layer. The output layer includes 4 output nodes, and the Softmax function is used to convert the output into a probability distribution. The cross entropy loss function is used to evaluate the difference between the predicted probability and the true label, and the classification probability vector of the vehicle environment feature is output. Taking the intersection of the first vehicle environment feature and the traffic weather information as the road meteorological feature, and fusing the road meteorological feature with the first vehicle environment feature to obtain the first road meteorological environment feature; Performing data comparison between the second vehicle environment feature and the vehicle environment distance information to obtain a distance difference, performing feature crossover between the distance difference and the second vehicle environment feature to obtain a first road obstacle distance feature, and performing feature fusion between the first road obstacle distance feature and the vehicle environment distance information to obtain a second road obstacle distance feature; Comparing the fourth vehicle environment feature with the standard vehicle appearance feature to obtain a vehicle exterior defect feature, and obtaining a vehicle appearance evaluation index according to the vehicle exterior defect feature; The environmental collision index is composed of a first road meteorological environment characteristic, a second road obstacle distance characteristic and a third vehicle environment characteristic; A second warning threshold is determined. When any environmental collision indicator is greater than the second warning threshold, a direct warning is issued according to the corresponding environmental collision indicator. A second marking threshold is determined according to the second warning threshold. When all environmental collision indicators are less than the second warning threshold and any environmental collision indicator is greater than the second marking threshold, the environmental collision indicator is marked to obtain a collision hazard marking value. When all environmental collision indicators are less than the second marking threshold, the motor vehicle is determined to be safe to drive.

5. The artificial intelligence motor vehicle safety performance detection method according to claim 1 is characterized in that: The method for performing the collision warning and safety performance inspection includes: The deep collision analysis includes cloud collision prediction and edge collision simulation; The specific operation of the edge collision simulation is as follows: constructing an edge collision simulation model according to the environmental collision index, the first driving data and the KITTI map data, and inputting the data to be simulated into the edge collision simulation model to generate a potential collision scene; the edge collision simulation model runs in real time on the edge computing node, including a GAN adversarial network and transfer learning; the GAN adversarial network uses a generator to receive KITTI map data and environmental collision indexes, and outputs a virtual road condition image, uses a discriminator to optimize the generator according to the first driving data discrimination result, and uses a perceptual loss function to improve the generation quality; the transfer learning migrates the virtual road condition image to the real scene, and uses a domain adversarial network to reduce the difference between virtual data and real data; When any environmental collision index is greater than the second marking threshold, a cloud collision prediction is performed: a cloud collision prediction model is constructed according to the environmental collision index, the first driving factor and the second driving factor, and the data to be predicted is input into the cloud collision prediction model to obtain a collision prediction result; the cloud collision prediction model runs in real time on the cloud platform, including a BP neural network and a support vector machine SVM, the BP neural network performs nonlinear regression on the input data and outputs a collision probability, and the vector machine SVM generates a hyperplane according to the input data to divide the collision category; The edge collision simulation and cloud collision prediction results are weightedly fused to output the fused collision category and fused collision probability. When the fused collision probability is greater than the collision warning probability threshold, a collision warning is issued, and the driving hazard mark value, collision hazard mark value and corresponding fused collision category are extracted. The safety performance of the motor vehicle is checked according to the driving hazard mark value and collision hazard mark value.

6. The artificial intelligence motor vehicle safety performance detection method according to claim 1 is characterized in that: The method for performing the cloud out-of-control prediction and out-of-control warning comprises: The hierarchical clustering method is used to divide the structural state data to obtain the component operation state data and maintenance log data; the maintenance log data includes the component quality parameters and the corresponding maintenance date; Determine the dynamic safety index of components based on component quality parameters and corresponding maintenance dates, and use the dynamic safety index of components to correct component operation status data to obtain structural operation indicators; Determine a third warning threshold, when any structural operation index is greater than the third warning threshold, issue a direct warning according to the corresponding structural operation index; determine a third marking threshold according to the third warning threshold, when all structural operation indexes are less than the third warning threshold and any structural operation index is greater than the third marking threshold, mark the structural operation index to obtain a risk of loss of control marking value; when all structural operation indexes are less than the third marking threshold, determine that the motor vehicle is safe to drive; When any structural operation index is greater than the third marking threshold, a cloud out-of-control prediction model is constructed according to the structural operation index, the first driving factor, the second driving factor and the vehicle appearance evaluation index, and the data to be predicted is input into the cloud out-of-control prediction model to obtain an out-of-control prediction result; the cloud collision prediction model runs in real time on the cloud platform, including a BP neural network and a support vector machine SVM, the BP neural network performs nonlinear regression on the input data and outputs the out-of-control probability, and the vector machine SVM generates a hyperplane according to the input data to divide the out-of-control category; When the probability of out-of-control is greater than the out-of-control warning probability threshold, an out-of-control warning is issued, and the out-of-control hidden danger mark value and the corresponding out-of-control category are extracted.

7. The artificial intelligence motor vehicle safety performance detection method according to claim 1 is characterized in that: Methods for determining the root causes of the hazards and intervention indicators, including: The first hidden danger mark value, the out-of-control category and the out-of-control probability form a causal set, and the causal set is divided into a training set and a test set according to a ratio of 7:3; the first hidden danger mark value includes a driving hidden danger mark value and an out-of-control hidden danger mark value; A causal graph model is constructed, wherein the causal graph model includes a PC algorithm, a causal attention module and a Do-Calculus algorithm; the constructed causal graph model is run in real time on a cloud platform; the PC algorithm is used to infer a causal graph structure according to a causal set; the causal attention module is used to annotate the direction and weight of the causal edge to obtain a causal graph model; the Do-Calculus algorithm is used to calculate the intervention effect to determine the intervention index, which is used to infer the root cause of the hidden danger; the intervention index represents the first hidden danger mark value that intervenes in a certain type of out-of-control effect; the hidden danger root cause represents the relationship between the first hidden danger mark value that causes a certain type of out-of-control situation and the corresponding hidden danger level; The importance of the causal attention module is expressed as: in The first hidden danger mark value Corresponding out-of-control category The original importance of attention, The first hidden danger mark value Corresponding out-of-control category The normalized importance of attention, The first hidden danger mark value Corresponding out-of-control category The probabilistic average causal effect of The first hidden danger mark value Corresponding out-of-control category The expected average causal effect of Out of control category Based on the shared weights of the probabilistic average causal effects, Out of control category Based on the shared weights of the expected average causal effect, is the regularization strength, Out of control category The probability of losing control, is the number of out-of-control categories, is the standard deviation of the causal association strength; The F1 score was used to evaluate the prediction performance of the model, and the test set was used to verify the accuracy of the causal graph model; The state data of the motor vehicle to be tested is input into the cause-effect diagram model to determine the root cause of the hidden danger and the intervention index, and the safety performance test is performed according to the root cause of the hidden danger and the intervention index.

8. An artificial intelligence motor vehicle safety performance detection system, used to execute the method according to any one of claims 1 to 7, characterized in that: include: Perception network module: used to construct a motor vehicle perception network, and use the motor vehicle perception network to obtain motor vehicle perception data; Edge computing module: used to divide the driving state data and determine the driving factor according to the division result; used to perform a first edge computing on the environmental data to obtain an environmental collision index; used to perform a second edge computing on the structural state data to obtain a structural operation index; Cloud computing module: used for performing in-depth collision analysis according to the environmental collision index and the driving factor; and performing cloud out-of-control prediction according to the structural operation index and the driving factor; Causal graph model module: used to construct a causal graph model according to the cloud out-of-control prediction results, and input the state data of the motor vehicle to be detected into the causal graph model to determine the root cause of the hidden danger and the intervention indicator; Intelligent supervision module: used to store, view and manage the driving status data, driving factors, collision indicators and structural operation indicators of motor vehicles, perform direct warning and marking operations based on the driving status data, the environmental collision indicators and the structural operation indicators, perform collision warnings based on the deep collision analysis results, perform out-of-control warnings based on the cloud out-of-control prediction results, and perform safety performance detection based on the root causes of hidden dangers and intervention indicators.

Images