Intelligent vehicle data collection method and system based on big data artificial intelligence analysis

Abstract

The present invention discloses a smart car data collection method and system based on big data artificial intelligence analysis, including obtaining sensor data and status data of the driving process, integrating the scan data, performing image extraction on the environment image to obtain driving environment data, building a user driving environment data sharing platform and matching regional driving environment data, screening the regional driving environment data, determining the vehicle state factor according to the screening result of the regional driving environment data, building a smart car-road scoring model according to the regional driving environment data and the vehicle state factor, inputting the sensor data and status data of the driving process into the smart car-road scoring model to obtain the car-road score, and performing smart car data collection according to the car-road score. This method can not only improve the efficiency and accuracy of smart car data collection, but also has good interpretability, and can be directly applied to smart car data collection systems.

Description

Technical Field

[0001] The present invention relates to the field of data collection, and in particular to a smart car data collection method and system based on big data artificial intelligence analysis. Background Art

[0002] With the rapid development of the automobile industry and the rise of intelligent transportation systems, smart cars are playing an increasingly important role in the field of modern transportation. Most smart cars are equipped with various advanced sensors, which continuously generate massive amounts of data during driving. Effectively collecting, processing and utilizing this data is of great significance to improving driving safety, optimizing driving efficiency and reducing energy consumption.

[0003] Traditional data collection methods mostly rely on a single sensor or device, which makes it difficult to fully capture the complex information during driving. There are also defects such as low data processing efficiency, insufficient data value mining, and lack of comprehensive analysis of different driving environments and vehicle states. At the same time, the storage and management of massive data also face huge challenges. The lack of effective sharing and integration mechanisms leads to the inability to fully interact and utilize data between different regions and vehicles, making it difficult to achieve collaborative optimization of intelligent transportation systems. Therefore, based on big data and machine learning technology, a smart car data collection method and system based on big data artificial intelligence analysis is designed to overcome the shortcomings of existing smart car data collection methods, provide a strong data basis for the optimization control of smart cars, intelligent decision-making, and the overall planning of intelligent transportation systems, and is of great significance to promoting the further development of the field of intelligent transportation. Summary of the invention

[0004] The purpose of the present invention is to provide a smart car data collection method and system based on big data artificial intelligence analysis.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Acquire sensor data and status data during driving, and pre-process the sensor data and status data; the sensor data includes environment images and scanning data;

[0008] Integrate the scanned data to determine the driving road condition and the driving vehicle condition, perform image extraction on the environment image to obtain driving environment data, and combine the driving environment data, the driving road condition and the status data into user driving environment data;

[0009] Building a user driving environment data sharing platform, uploading the user driving environment data and matching the regional driving environment data, screening the regional driving environment data, and determining the vehicle state factor according to the screening result of the regional driving environment data;

[0010] An intelligent vehicle-road scoring model is constructed according to the regional driving environment data and the vehicle status factor, the sensor data and status data of the driving process are input into the intelligent vehicle-road scoring model to obtain a vehicle-road score, and intelligent vehicle data is collected according to the vehicle-road score.

[0011] Furthermore, the method for obtaining the sensor data and status data of the driving process includes:

[0012] The method of obtaining sensor data includes: collecting environmental images of the driving process through cameras around the vehicle body, and collecting distance scanning data of the driving process through laser radars, millimeter wave radars and ultrasonic radars at various locations on the vehicle body;

[0013] Methods for obtaining status data include: collecting driving positioning data through GPS sensors, collecting driving motion status and posture data through IMU sensors, and directly obtaining vehicle status data through the vehicle system.

[0014] Furthermore, the method of integrating the scan data to determine the road condition and vehicle condition includes:

[0015] Divide the scanned data into training and testing sets;

[0016] Constructing a driving data model, wherein the driving data model includes a decision tree, a feature fusion layer, and a stacked classifier;

[0017] The decision tree performs feature selection on the input scan data and divides the scan data into two categories: driving road conditions and driving vehicle conditions;

[0018] The feature fusion layer includes a multi-head attention mechanism and a feature fusion strategy. It uses a multi-head self-attention mechanism to capture the spatial relationship between scanned data at different angles, processes multiple distance data of different categories in parallel, and uses a Concatenate strategy to splice and fuse feature vectors of different dimensions. The expression is:

[0019]

[0020]

[0021]

[0022] in Query vector in the attention mechanism Attention output on key-value pairs, is the focus content key vector, is the attention content value vector, is the activation function, For the A vector of attention content keys, For the A vector of attention content values, is the number of key-value pairs, is the scaling factor, is the regularization strength hyperparameter, is the L2 regularization term, It is a multi-head attention mechanism representation, used to connect multiple attentions. is the initial parameter matrix, For the Head query vector parameter matrix, For the Header attention content key vector parameter matrix, For the Head attention content value vector parameter matrix, For the Head output;

[0023] The stacked classifier processes the output data of the feature fusion layer, including the base model and the meta-model. The base models independently learn the distance data for distance prediction. The meta-model makes a final decision on the distance prediction results of the base models and outputs the road conditions and vehicle conditions. The base models include random forest, K nearest neighbor and support vector machine.

[0024] The cross entropy loss function is used to distinguish between driving road conditions and driving vehicle condition data, grid search is used to tune the hyperparameters of the driving data model, and the test set data is used to evaluate the model;

[0025] The scanned data to be integrated is input into the driving data model to obtain the driving road condition and driving vehicle condition.

[0026] Furthermore, the method for matching the regional driving environment data includes:

[0027] The driving environment data, the road conditions and the GPS positioning information are combined into the user driving environment data;

[0028] Build a user driving environment data sharing platform. The user driving environment data sharing platform uses the user data sharing contract to conduct real-time interaction of driving environment data between regional users. It uses a symmetric encryption algorithm to encrypt user data before transmission, and uses an asymmetric algorithm to verify user identity and authorize users to read data.

[0029] According to the user driving environment data matching user driving environment data sharing platform database, define the user driving environment data vector as , define the driving environment data vector of other users in the area in the database as ,in is a vector dimension, and the regional driving environment data is determined according to the data similarity. The expression is:

[0030]

[0031] in Driving environment data vector for this user Driving environment data vector with other users in the area The comprehensive similarity of is the difference fluctuation ratio, is the amplitude difference ratio, and , For vector With vector The covariance of For vector The standard deviation of For vector The standard deviation of

[0032] Select the data vector related to the user's driving environment The driving environment data vectors of other users in the m regions with the highest comprehensive similarity , calculate the regional driving environment data vector , according to the regional driving environment data vector Determine regional driving environment data.

[0033] Furthermore, the method for screening the regional driving environment data includes:

[0034] The equal volume hypersphere method is used to screen the regional driving environment data. The hypersphere is assumed to be , is the radius of the hypersphere, is the hypersphere dimension, For the volume of the hypersphere, the equivalent radius is optimized by searching for the minimum star deviation , the expression is:

[0035]

[0036] in To search for the best uniform effective representative point set obtained by the minimum star deviation, is the reference point set, is the optimal equivalent radius, The radius is Hypersphere The number of valid representative points in is the total number of data points, Hypersphere volume, For cube The volume of is the characteristic scale;

[0037] Determine the screening conditions of the equal volume hypersphere method, the expression is:

[0038]

[0039]

[0040] in For the Points in The sum of the squares of the coordinates of each dimension in the dimensional space, is the number of selected representative points.

[0041] Furthermore, the method for determining the vehicle state factor according to the regional driving environment data screening result includes:

[0042] Determine the state data at the corresponding time based on the regional driving environment data screening results, and input the state data into the state influence function to obtain the vehicle state factor, which is expressed as:

[0043]

[0044] in is the vehicle state factor, is the out-of-control probability weight, is the speed state factor weight, is the weight of safe driving factors, User usage weight, The number of users of the assisted driving system, is the rated number of users, The probability of BUG in the auxiliary driving system. is the network delay rate, is the current vehicle driving speed, Set the driving speed for the street at the current moment, is the current vehicle acceleration, Score the current vehicle posture, For the vehicle safety protection structure category, is the number of vehicle safety protection structure categories, Provides a safety score for the vehicle's protective structure. is the category of factors affecting vehicle driving, is the number of categories of factors affecting vehicle driving, is the current vehicle driving influencing factor value, is the standard value of the current vehicle driving influencing factors, is the current vehicle travel time, Score tire pressure, Score the tank oil pressure. Rate the fuel tank capacity.

[0045] Furthermore, the method for collecting smart car data according to the vehicle-road score includes:

[0046] An intelligent vehicle-road scoring model is constructed based on the regional driving environment data and the vehicle status factor, the sensor data and status data of the driving process are input into the intelligent vehicle-road scoring model to obtain a vehicle-road score, and intelligent vehicle data is collected based on the vehicle-road score

[0047] The abnormal data of regional driving environment data and the corresponding vehicle status factors are combined into a vehicle-road data set, and the vehicle-road data set is divided into a training set and a test set;

[0048] Construct an intelligent vehicle-road scoring model based on isolation forest. The input layer preprocesses the input data and selects features. The isolation forest model outputs an anomaly score for each data point. The output layer assigns an overall score to the entire data group and extracts abnormal data labels based on the anomaly score and preset threshold.

[0049] The training set is used to train the isolation forest model, the cross entropy loss function is used for multi-category anomaly classification, the RMSprop optimizer is used to adjust the model learning rate, the early stopping strategy is used to control the training process, and the test set is used to evaluate the model accuracy;

[0050] The sensor data and status data of the driving process are input into the intelligent vehicle-road scoring model to obtain the vehicle-road score and the corresponding abnormal label, and the intelligent vehicle data collection strategy is selected according to the vehicle-road score.

[0051] The second aspect is the intelligent vehicle data collection system based on big data artificial intelligence analysis, including:

[0052] Data acquisition module: including GPS sensor, IMU sensor, intelligent driving camera, laser radar, ultrasonic radar; used to collect sensor data and status data during driving, and pre-process the sensor data and status data;

[0053] Image extraction module: used for performing image extraction on the environment image to obtain driving environment data;

[0054] Data processing module: used to integrate the scanned data to determine the road condition and vehicle condition, to screen the abnormal data of the regional driving environment data and determine the state data at the corresponding moment, and to determine the vehicle state factor according to the state data;

[0055] Scoring model module: constructing an intelligent vehicle-road scoring model according to the regional driving environment data and the vehicle status factor, and inputting the sensor data and status data of the driving process into the intelligent vehicle-road scoring model to obtain a vehicle-road score;

[0056] Intelligent supervision module: used to store, view and manage the sensor data, status data and the vehicle-road score of the driving process, select the intelligent vehicle data collection strategy according to the vehicle-road score and perform data collection.

[0057] The beneficial effects of the present invention are:

[0058] The present invention is a smart car data collection method and system based on big data artificial intelligence analysis. Compared with the prior art, the present invention has the following technical effects:

[0059] The present invention can improve the data preprocessing capabilities and enhance the model adaptability in smart car data collection through data fusion, image extraction, data interaction, abnormal data screening, construction of vehicle status factors, and construction of scoring models, thereby improving the efficiency of smart car data collection. The smart car data collection technology is optimized, which can greatly save resources and improve work efficiency. The data collection of smart cars is realized, which is of great significance for optimizing driving efficiency and improving driving safety. It can adapt to different smart car data collection systems based on big data artificial intelligence analysis and different users' smart car data collection needs based on big data artificial intelligence analysis, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 The present invention is a flowchart of the steps of the smart car data collection method based on big data artificial intelligence analysis. DETAILED DESCRIPTION

[0061] 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.

[0062] The smart car data collection method and system based on big data artificial intelligence analysis of the present invention include the following steps:

[0063] like Figure 1 As shown, in this embodiment, the following steps are included:

[0064] Acquire sensor data and status data during driving, and pre-process the sensor data and status data; the sensor data includes environment images and scanning data;

[0065] Integrate the scanned data to determine the driving road condition and the driving vehicle condition, perform image extraction on the environment image to obtain driving environment data, and combine the driving environment data, the driving road condition and the status data into user driving environment data;

[0066] Building a user driving environment data sharing platform, uploading the user driving environment data and matching the regional driving environment data, screening the regional driving environment data, and determining the vehicle state factor according to the screening result of the regional driving environment data;

[0067] An intelligent vehicle-road scoring model is constructed according to the regional driving environment data and the vehicle status factor, the sensor data and status data of the driving process are input into the intelligent vehicle-road scoring model to obtain a vehicle-road score, and intelligent vehicle data is collected according to the vehicle-road score.

[0068] In this embodiment, the method for obtaining sensor data and status data during driving includes:

[0069] The method of obtaining sensor data includes: collecting environmental images of the driving process through cameras around the vehicle body, and collecting distance scanning data of the driving process through laser radars, millimeter wave radars and ultrasonic radars at various locations on the vehicle body;

[0070] The method for obtaining status data includes: collecting driving positioning data through a GPS sensor, collecting driving motion status and posture data through an IMU sensor, and directly obtaining vehicle status data through a vehicle system; the vehicle status data includes seat belt and airbag status, tire pressure, fuel tank fuel level, driving gear, oncoming vehicle lights and driving noise.

[0071] In this embodiment, the method of integrating the scanning data to determine the driving road condition and the driving vehicle condition includes:

[0072] Divide the scanned data into training and testing sets;

[0073] Constructing a driving data model, wherein the driving data model includes a decision tree, a feature fusion layer, and a stacked classifier;

[0074] The decision tree performs feature selection on the input scan data and divides the scan data into two categories: driving road conditions and driving vehicle conditions; the driving road conditions include the distance between the vehicle and surrounding roads and intersections; the driving vehicle conditions include the distance between the vehicle and obstacles, surrounding vehicles, and guardrails;

[0075] The feature fusion layer includes a multi-head attention mechanism and a feature fusion strategy. It uses a multi-head self-attention mechanism to capture the spatial relationship between scanned data at different angles, processes multiple distance data of different categories in parallel, and uses a Concatenate strategy to splice and fuse feature vectors of different dimensions. The expression is:

[0076]

[0077]

[0078]

[0079] in Query vector in the attention mechanism Attention output on key-value pairs, is the focus content key vector, is the attention content value vector, is the activation function, For the A vector of attention content keys, For the A vector of attention content values, is the number of key-value pairs, is the scaling factor, is the regularization strength hyperparameter, is the L2 regularization term, It is a multi-head attention mechanism representation, used to connect multiple attentions. is the initial parameter matrix, For the Head query vector parameter matrix, For the Header attention content key vector parameter matrix, For the Head attention content value vector parameter matrix, For the Head output;

[0080] The stacked classifier processes the output data of the feature fusion layer, including the base model and the meta-model. The base models independently learn the distance data for distance prediction. The meta-model makes a final decision on the distance prediction results of the base models and outputs the road conditions and vehicle conditions. The base models include random forest, K nearest neighbor and support vector machine.

[0081] The cross entropy loss function is used to distinguish between driving road conditions and driving vehicle condition data, grid search is used to tune the hyperparameters of the driving data model, and the test set data is used to evaluate the model;

[0082] Input the scanned data to be integrated into the driving data model to obtain the driving road condition and driving vehicle condition;

[0083] In the actual evaluation, the sensor data of a car user's driving process is obtained, and the sensor data of the driving process is input into the trained driving data model to obtain: smart car ID: CAR12***88, timestamp: 2024-06-13T17:22:01Z, driving road conditions (the distance between the vehicle and the intersection ahead is 150.5m, the distance from the edge of the road is 2.1m, the distance from the traffic light is 15m, and the distance from the middle guardrail is 3.8m), driving conditions (the vehicle on the left is 0.8m away-speed 30km / h, the vehicle in front is 15.2m away-speed 40km / h, the vehicle behind is 20.1m away-speed 42km / h, the obstacle in front is 5.6m away, and the pedestrian on the zebra crossing in front is 10.3m away).

[0084] In this embodiment, the method of performing image extraction on the environment image to obtain driving environment data includes:

[0085] The random forest algorithm is used to divide the environmental images into training and test sets in a ratio of 7:3;

[0086] Build a driving environment image model, which includes CNN convolutional neural network, fully connected layer, target detection algorithm, image segmentation technology, stacking generalization and decision tree;

[0087] CNN convolutional neural network performs deep learning on the environment image to extract image features, and inputs the feature extraction results into the post-structure;

[0088] The post-processing structure is arranged in parallel, including a fully connected layer, a target detection algorithm, and an image segmentation technology. The fully connected layer uses a softmax activation function to identify features such as traffic lights, traffic signs, road signs, and dangerous sources in the environmental image, and uses a FocalLoss loss function to measure the loss contribution of reducing easy-to-classify samples and increasing the loss contribution of difficult-to-classify samples; the FasterR-CNN target detection algorithm is used to extract traffic light time, traffic sign data, pedestrian flow, vehicle flow, and the number of dangerous sources, and the mean absolute error is used to adjust the difference between the predicted value and the true value; the image segmentation technology distinguishes between dangerous scenes and weather scene elements, and uses a multi-category cross entropy loss function to adjust the multi-label classification scene;

[0089] The stacking generalization combines the results of feature recognition, data extraction and scene recognition to obtain a comprehensive recognition result, uses a decision tree to classify the driving environment data recognition results, and uses a structural loss function to optimize the decision tree structure; the road signals include traffic light status, traffic signs and road signs; the road traffic includes pedestrian traffic and vehicle traffic; the road hazards include hazard source types, hazard source quantity and hazard scenarios;

[0090] The grid search technique was used to tune the model hyperparameters, and the test set was used to evaluate the accuracy of the model;

[0091] Inputting the environment image to be processed into the driving environment image model to obtain driving environment data;

[0092] In the actual evaluation, the driving environment image model is used to process the vehicle environment image to obtain: road signals (green light ahead, 12s remaining on the signal light, 45km / h speed limit on the road sign, school ahead, dual lanes), road traffic (8 people passing through the zebra crossing ahead / dense pedestrians, 10 non-motor vehicles on the non-motor vehicle lane on the right front, 8 non-motor vehicles on the right rear, 12 vehicles in sight ahead / slightly crowded, 6 vehicles in sight behind / unobstructed), road hazards (obstacles - temporary fault signs, 1 obstacle, hazard sources - a fully loaded large truck 20m ahead, no dangerous accidents or scenes), and actual weather conditions (light rain, slippery ground, good visibility).

[0093] In this embodiment, the method for matching the regional driving environment data includes:

[0094] The driving environment data, the road conditions and the GPS positioning information are combined into the user driving environment data;

[0095] Build a user driving environment data sharing platform. The user driving environment data sharing platform uses the user data sharing contract to conduct real-time interaction of driving environment data between regional users. It uses a symmetric encryption algorithm to encrypt user data before transmission, and uses an asymmetric algorithm to verify user identity and authorize users to read data.

[0096] According to the user driving environment data matching user driving environment data sharing platform database, define the user driving environment data vector as , define the driving environment data vector of other users in the area in the database as ,in is a vector dimension, and the regional driving environment data is determined according to the data similarity. The expression is:

[0097]

[0098] in Driving environment data vector for this user Driving environment data vector with other users in the area The comprehensive similarity of is the difference fluctuation ratio, is the amplitude difference ratio, and , For vector With vector The covariance of For vector The standard deviation of For vector The standard deviation of

[0099] Select the data vector related to the user's driving environment The driving environment data vectors of other users in the m regions with the highest comprehensive similarity , calculate the regional driving environment data vector , according to the regional driving environment data vector Determine regional driving environment data;

[0100] In the actual evaluation, the user driving environment data is uploaded to the user driving environment data sharing platform and the user driving environment data of the nearby area is obtained: 18m in front of the user / left lane, driving road conditions (the distance between the vehicle and the front intersection-unknown, the distance to the edge of the road 5.6m, the distance to the traffic light 100m, the distance to the middle guardrail 0.3m), driving conditions (the vehicle on the right is 0.8m away-speed 36km / h, the vehicle in front is 20m away-speed 12km / h, the vehicle behind is 10m away-speed 25 km / h, no obstacles), road signals (green light ahead, 22s remaining on the signal light, 20km / h speed limit on the road sign, school ahead, dual carriageway), road traffic (25 people on the zebra crossing ahead / pedestrian congestion, 22 non-motor vehicles on the non-motor vehicle lane ahead, 19 non-motor vehicles on the right rear, 18 vehicles in sight ahead / congestion, 10 vehicles in sight behind / minor congestion), road hazards (no obstacles, a fully loaded large truck on the right side of the hazard source, no dangerous accidents or scenes), actual weather conditions (light rain, slippery ground, good visibility).

[0101] In this embodiment, the method for screening the regional driving environment data includes:

[0102] The equal volume hypersphere method is used to screen the regional driving environment data. The hypersphere is assumed to be , is the radius of the hypersphere, is the hypersphere dimension, For the volume of the hypersphere, the equivalent radius is optimized by searching for the minimum star deviation , the expression is:

[0103]

[0104] in To search for the best uniform effective representative point set obtained by the minimum star deviation, is the reference point set, is the optimal equivalent radius, The radius is Hypersphere The number of valid representative points in is the total number of data points, Hypersphere volume, For cube The volume of is the characteristic scale;

[0105] Determine the screening conditions of the equal volume hypersphere method, the expression is:

[0106]

[0107]

[0108] in For the Points in The sum of the squares of the coordinates of each dimension in the dimensional space, is the number of selected representative points;

[0109] In the actual evaluation, the abnormal data in the driving environment data of this area are: there are pedestrians walking on the zebra crossing in front of the user's vehicle, there is an obstacle 5.6m ahead, there is a fully loaded / unslowed large truck 20m ahead, it is drizzling and the ground is slippery, and there is pedestrian congestion at the school ahead.

[0110] In this embodiment, the method for determining the vehicle state factor according to the regional driving environment data screening result includes:

[0111] Determine the state data at the corresponding time based on the regional driving environment data screening results, and input the state data into the state influence function to obtain the vehicle state factor, which is expressed as:

[0112]

[0113] in is the vehicle state factor, is the out-of-control probability weight, is the speed state factor weight, is the weight of safe driving factors, User usage weight, The number of users of the assisted driving system, is the rated number of users, The probability of BUG in the auxiliary driving system. is the network delay rate, is the current vehicle driving speed, Set the driving speed for the street at the current moment, is the current vehicle acceleration, Score the current vehicle posture, For the vehicle safety protection structure category, is the number of vehicle safety protection structure categories, Provides a safety score for the vehicle's protective structure. is the category of factors affecting vehicle driving, is the number of categories of factors affecting vehicle driving, is the current vehicle driving influencing factor value, is the standard value of the current vehicle driving influencing factors, is the current vehicle travel time, Score tire pressure, Score the tank oil pressure. Score the fuel tank level;

[0114] In actual evaluation, the vehicle status factor weight is , the state data of the regional driving environment data corresponding to the abnormal data at the time ( )Input state influence function to obtain vehicle state factor .

[0115] In this embodiment, the method for collecting smart car data according to the vehicle-road score includes:

[0116] An intelligent vehicle-road scoring model is constructed based on the regional driving environment data and the vehicle status factor, the sensor data and status data of the driving process are input into the intelligent vehicle-road scoring model to obtain a vehicle-road score, and intelligent vehicle data is collected based on the vehicle-road score

[0117] The abnormal data of regional driving environment data and the corresponding vehicle status factors are combined into a vehicle-road data set, and the vehicle-road data set is divided into a training set and a test set;

[0118] Construct an intelligent vehicle-road scoring model based on isolation forest. The input layer preprocesses the input data and selects features. The isolation forest model outputs an anomaly score for each data point. The output layer assigns an overall score to the entire data group and extracts abnormal data labels based on the anomaly score and preset threshold.

[0119] The training set is used to train the isolation forest model, the cross entropy loss function is used for multi-category anomaly classification, the RMSprop optimizer is used to adjust the model learning rate, the early stopping strategy is used to control the training process, and the test set is used to evaluate the model accuracy;

[0120] The sensor data and status data of the driving process are input into the intelligent vehicle-road scoring model to obtain the vehicle-road score and the corresponding abnormal label, and the intelligent vehicle data collection strategy is selected according to the vehicle-road score; when the vehicle-road score is greater than the set threshold, the driving process status data at the corresponding time point is collected; when the vehicle-road score is less than the set threshold, the sensor data and status data of the driving process at the corresponding time point are collected;

[0121] In the actual evaluation, the sensor data and status data of the driving process are input into the intelligent vehicle-road scoring model to obtain a vehicle-road score of 6.6 points, abnormal conditions (including pedestrians passing through zebra crossings, obstacles, fully loaded large trucks that do not slow down, light rain and slippery ground, school sections, and pedestrian road congestion), and the vehicle-road score is less than the vehicle-road score threshold of 8 points, and the sensor data and status data of the driving process during this period are collected.

[0122] The second aspect is the intelligent vehicle data collection system based on big data artificial intelligence analysis, including:

[0123] Data acquisition module: including GPS sensor, IMU sensor, intelligent driving camera, laser radar, ultrasonic radar; used to collect sensor data and status data during driving, and pre-process the sensor data and status data;

[0124] Image extraction module: used for performing image extraction on the environment image to obtain driving environment data;

[0125] Data processing module: used to integrate the scanned data to determine the road condition and vehicle condition, to screen the abnormal data of the regional driving environment data and determine the state data at the corresponding moment, and to determine the vehicle state factor according to the state data;

[0126] Scoring model module: constructing an intelligent vehicle-road scoring model according to the regional driving environment data and the vehicle status factor, and inputting the sensor data and status data of the driving process into the intelligent vehicle-road scoring model to obtain a vehicle-road score;

[0127] Intelligent supervision module: used to store, view and manage the sensor data, status data and the vehicle-road score of the driving process, select the intelligent vehicle data collection strategy according to the vehicle-road score and perform data collection.

[0128] 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. A smart car data collection method based on big data artificial intelligence analysis, characterized in that: The following steps are involved: S1, acquiring sensor data and status data of the driving process, and preprocessing the sensor data and the status data; the sensor data includes an environment image and scanning data; S2, integrating the scanned data to determine the road condition and the vehicle condition, performing image extraction on the environment image to obtain driving environment data, and combining the driving environment data, the road condition and the status data to form user driving environment data; S3, constructing a user driving environment data sharing platform, uploading the user driving environment data and matching the regional driving environment data, screening the regional driving environment data, and determining the vehicle state factor according to the screening result of the regional driving environment data; S4, constructing an intelligent vehicle-road scoring model according to the regional driving environment data and the vehicle state factor, inputting the sensor data and state data of the driving process into the intelligent vehicle-road scoring model to obtain a vehicle-road score, and performing intelligent vehicle data collection according to the vehicle-road score; Determine the state data at the corresponding time based on the regional driving environment data screening results, and input the state data into the state influence function to obtain the vehicle state factor, which is expressed as: in is the vehicle state factor, is the out-of-control probability weight, is the speed state factor weight, is the weight of safe driving factors, User usage weight, The number of users of the assisted driving system, is the rated number of users, The probability of BUG in the auxiliary driving system. is the network delay rate, is the current vehicle driving speed, Set the driving speed for the street at the current moment, is the current vehicle acceleration, Score the current vehicle posture, It is the vehicle safety protection structure category. is the number of vehicle safety protection structure categories, Provides a safety score for the vehicle's protective structure. is the category of factors affecting vehicle driving, is the number of categories of factors affecting vehicle driving, is the current vehicle driving influencing factor value, is the standard value of the current vehicle driving influencing factors, is the current vehicle travel time, Score tire pressure, Score the tank oil pressure, Score the fuel tank capacity; The abnormal data of regional driving environment data and the corresponding vehicle status factors are combined into a vehicle-road data set, and the vehicle-road data set is divided into a training set and a test set; Construct an intelligent vehicle-road scoring model based on isolation forest. The input layer preprocesses the input data and selects features. The isolation forest model outputs an anomaly score for each data point. The output layer assigns an overall score to the entire data group and extracts abnormal data labels based on the anomaly score and preset threshold. The training set is used to train the isolation forest model, the cross entropy loss function is used for multi-category anomaly classification, the RMSprop optimizer is used to adjust the model learning rate, the early stopping strategy is used to control the training process, and the test set is used to evaluate the model accuracy; The sensor data and status data of the driving process are input into the intelligent vehicle-road scoring model to obtain the vehicle-road score and the corresponding abnormal label, and the intelligent vehicle data collection strategy is selected according to the vehicle-road score.

2. The smart car data collection method based on big data artificial intelligence analysis according to claim 1 is characterized in that: The method for obtaining sensor data and status data of a driving process includes: The method of obtaining sensor data includes: collecting environmental images of the driving process through cameras around the vehicle body, and collecting distance scanning data of the driving process through laser radars, millimeter wave radars and ultrasonic radars at various locations on the vehicle body; Methods for obtaining status data include: collecting driving positioning data through GPS sensors, collecting driving motion status and posture data through IMU sensors, and directly obtaining vehicle status data through the vehicle system.

3. The smart car data collection method based on big data artificial intelligence analysis according to claim 1 is characterized in that: The method of integrating the scan data to determine the road condition and vehicle condition includes: Divide the scanned data into training and testing sets; Constructing a driving data model, wherein the driving data model includes a decision tree, a feature fusion layer, and a stacked classifier; The decision tree performs feature selection on the input scan data and divides the scan data into two categories: driving road conditions and driving vehicle conditions; The feature fusion layer includes a multi-head attention mechanism and a feature fusion strategy. It uses a multi-head self-attention mechanism to capture the spatial relationship between scanned data at different angles, processes multiple distance data of different categories in parallel, and uses a Concatenate strategy to splice and fuse feature vectors of different dimensions. The expression is: in Query vector in the attention mechanism Attention output on key-value pairs, is the focus content key vector, is the attention content value vector, is the activation function, For the A vector of attention content keys, For the A vector of attention content values, is the number of key-value pairs, is the scaling factor, is the regularization strength hyperparameter, is the L2 regularization term, It is a multi-head attention mechanism representation, used to connect multiple attentions. is the initial parameter matrix, For the Head query vector parameter matrix, For the Header attention content key vector parameter matrix, For the Head attention content value vector parameter matrix, For the Head output; The stacked classifier processes the output data of the feature fusion layer, including the base model and the meta-model. The base models independently learn the distance data for distance prediction. The meta-model makes a final decision on the distance prediction results of the base models and outputs the road conditions and vehicle conditions. The base models include random forest, K nearest neighbor and support vector machine. The cross entropy loss function is used to distinguish between driving road conditions and driving vehicle condition data, grid search is used to tune the driving data model parameters, and the test set data is used to evaluate the model; The scanned data to be integrated is input into the driving data model to obtain the driving road condition and driving vehicle condition.

4. The smart car data collection method based on big data artificial intelligence analysis according to claim 1 is characterized in that: The method for matching the regional driving environment data comprises: The driving environment data, the road conditions and the GPS positioning information are combined into the user driving environment data; Build a user driving environment data sharing platform. The user driving environment data sharing platform uses the user data sharing contract to conduct real-time interaction of driving environment data between regional users. It uses a symmetric encryption algorithm to encrypt user data before transmission, and uses an asymmetric algorithm to verify user identity and authorize users to read data. According to the user driving environment data matching user driving environment data sharing platform database, define the user driving environment data vector as , define the driving environment data vector of other users in the area in the database as ,in is a vector dimension, and the regional driving environment data is determined according to the data similarity. The expression is: in Driving environment data vector for this user Driving environment data vector with other users in the area The comprehensive similarity of is the difference fluctuation ratio, is the amplitude difference ratio, and , For vector With vector The covariance of For vector The standard deviation of For vector The standard deviation of Select the data vector related to the user's driving environment The driving environment data vectors of other users in the m regions with the highest comprehensive similarity , calculate the regional driving environment data vector , according to the regional driving environment data vector Determine regional driving environment data.

5. The smart car data collection method based on big data artificial intelligence analysis according to claim 1 is characterized in that: The method for screening the regional driving environment data comprises: The equal volume hypersphere method is used to screen the regional driving environment data. The hypersphere is assumed to be , is the radius of the hypersphere, is the hypersphere dimension, For the volume of the hypersphere, the equivalent radius is optimized by searching for the minimum star deviation , the expression is: in To search for the most uniform effective representative point set obtained by the minimum star deviation, is the reference point set, is the optimal equivalent radius, The radius is Hypersphere The number of valid representative points in is the total number of data points, Hypersphere volume, For cube The volume of is the characteristic scale; Determine the screening conditions of the equal volume hypersphere method, the expression is: in For the Points in The sum of the squares of the coordinates of each dimension in the dimensional space, is the number of selected representative points.

6. A smart car data acquisition system based on big data artificial intelligence analysis, used to execute the method described in any one of claims 1 to 5, characterized in that: include: Data acquisition module: including GPS sensor, IMU sensor, smart driving camera, laser radar, ultrasonic radar; Used to collect sensor data and status data during driving, and pre-process the sensor data and status data; Image extraction module: used for performing image extraction on the environment image to obtain driving environment data; Data processing module: used to integrate the scanned data to determine the road condition and vehicle condition, to screen the abnormal data of the regional driving environment data and determine the state data at the corresponding moment, and to determine the vehicle state factor according to the state data; Data interaction module: used to upload the user driving environment data to the user driving environment data sharing platform, and read and match the regional driving environment data; Scoring model module: constructing an intelligent vehicle-road scoring model according to the regional driving environment data and the vehicle status factor, and inputting the sensor data and status data of the driving process into the intelligent vehicle-road scoring model to obtain a vehicle-road score; Intelligent supervision module: used to store, view and manage the sensor data, status data and the vehicle-road score of the driving process, select the intelligent vehicle data collection strategy according to the vehicle-road score and perform data collection.

Images