Vehicle safety assessment method and system based on artificial intelligence

Abstract

The present invention discloses a vehicle safety assessment method and system based on artificial intelligence, and the method includes data collection, data preprocessing, historical trajectory safety analysis, intention trajectory safety analysis and vehicle safety assessment. The present invention relates to the field of vehicle safety assessment technology, and specifically refers to a vehicle safety assessment method and system based on artificial intelligence. The present invention adopts a two-way vehicle driving safety analysis strategy that combines historical trajectory analysis and intention trajectory analysis, and then uses an artificial intelligence method improved for specific tasks; the attention sequence optimization long short-term memory neural network combined with weighted prediction of safety indicators is used to perform historical trajectory safety analysis, which improves the prediction depth of historical trajectory safety analysis, and at the same time improves the overall usability and reference of safety analysis results; artificial data label optimization and preprocessing optimization are used, and the integrated network of deep neural network combined with self-attention mechanism and long short-term memory neural network is used to perform intention trajectory safety analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle safety assessment, and in particular to a vehicle safety assessment method and system based on artificial intelligence. Background Art

[0002] Artificial intelligence-based vehicle safety assessment methods use machine learning and data analysis technologies to automatically assess the safety performance of vehicles by processing and analyzing large amounts of vehicle operation data, driving behavior data, and environmental data. These methods can identify potential safety hazards, predict possible accidents, and provide preventive measures, thereby improving overall traffic safety, reducing the occurrence of traffic accidents, and improving the safety level of drivers and passengers.

[0003] However, among the existing vehicle safety assessment methods, there are traditional methods for vehicle safety assessment that focus on a large amount of analysis of historical driving process data, thereby realizing a general safety analysis of vehicle driving. However, this analysis method cannot cope with the specific changes in the driving trajectory during the vehicle driving process. At the same time, for real-time driving vehicles, the lane change decisions and collision avoidance strategies executed by the driver cannot be analyzed, predicted and controlled in a timely and effective manner, thereby reducing the timeliness and availability of vehicle safety analysis. Technical problems; In the existing historical trajectory safety analysis methods, there are many data for historical trajectory safety analysis, which are time series data. Although the traditional long short-term memory neural network can process time series data and features, the general or simple binary classification and multi-classification methods cannot effectively provide highly interpretable result feedback for historical trajectory safety analysis, resulting in the lack of intuitiveness and effectiveness of safety analysis. Technical problems; In the existing intention trajectory safety analysis methods, there are intention trajectory safety analysis methods that need to ensure the prediction accuracy of the model and also need to consider real-time support to improve prediction efficiency, while the traditional technology uses large-scale data solutions that need to be improved in timeliness, and reducing the data scale may lead to insufficient accuracy. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a vehicle safety assessment method and system based on artificial intelligence. In the existing vehicle safety assessment methods, there are traditional methods for vehicle safety assessment that focus on a large amount of analysis of historical driving process data, thereby realizing a general safety analysis of vehicle driving. However, this analysis method cannot cope with the specific driving trajectory changes during vehicle driving. At the same time, for real-time driving vehicles, the lane change decisions and collision avoidance strategies executed by the driver cannot be timely and effectively analyzed, predicted and controlled for feedback, thereby reducing the timeliness and availability of vehicle safety analysis. The present solution creatively adopts a two-way vehicle driving safety analysis strategy that combines historical trajectory analysis and intention trajectory analysis, and then uses an artificial intelligence method improved for specific tasks to realize the inclusion of the lane change decisions and collision avoidance strategies executed by the driver into the consideration scope of vehicle safety analysis, thereby providing a more comprehensive mining of the analysis depth of vehicle safety analysis; in the existing historical trajectory safety analysis methods, there are many historical trajectory safety analysis data that are time series data, and although the traditional long short-term memory neural network can process time series data and features, the general or simple binary classification and multi-classification methods cannot effectively In order to solve the technical problem of providing highly interpretable result feedback for historical trajectory safety analysis, which leads to the lack of intuitiveness and effectiveness of safety analysis, this solution creatively adopts the attention sequence optimization long short-term memory neural network combined with weighted prediction of safety indicators to perform historical trajectory safety analysis. By constructing four different safety indicators and using them as prediction results and weighted processing respectively, the prediction depth of historical trajectory safety analysis is improved, and the overall availability and reference of safety analysis results are also improved. In view of the existing intention trajectory safety analysis methods, there is a technical problem that the intention trajectory safety analysis method needs to ensure the prediction accuracy of the model and consider real-time support to improve the prediction efficiency, while the traditional technology uses large-scale data solutions in terms of timeliness, and reducing the data scale may lead to insufficient accuracy. This solution creatively adopts artificial data label optimization and preprocessing optimization, combines the deep neural network with self-attention mechanism and the integrated network of long short-term memory neural network to perform intention trajectory safety analysis, and performs intention trajectory safety analysis through artificial annotation-assisted integrated network, thereby improving the efficiency and accuracy of prediction at the same time, providing strong data support for the subsequent two-way prediction integration results, and also providing a feasible and referenceable expansion idea for the direction of vehicle safety analysis.

[0005] The technical solution adopted by the present invention is as follows: The vehicle safety assessment method based on artificial intelligence provided by the present invention comprises the following steps:

[0006] Step S1: data collection;

[0007] Step S2: data preprocessing;

[0008] Step S3: historical trajectory safety analysis;

[0009] Step S4: security analysis of intention trajectory;

[0010] Step S5: Vehicle safety assessment.

[0011] Further, in step S1, the data collection is used to collect the original data set required for vehicle safety assessment, specifically from the driving record data, vehicle-mounted sensors and vehicle accident data set, to obtain the original data for safety assessment through data collection;

[0012] The safety assessment raw data specifically includes vehicle trajectory data, vehicle sensor data, environmental sensor data and historical accident data.

[0013] Furthermore, in step S2, the data preprocessing is used to optimize the original data, specifically, manually performing data label optimization, data cleaning and data alignment operations on the original security assessment data in sequence to obtain an optimized security assessment data set.

[0014] Further, in step S3, the historical trajectory safety analysis is used to perform automated collision precursor safety analysis in combination with historical data of the vehicle's driving trajectory, specifically, based on the optimized safety assessment data set, the attention sequence optimized long short-term memory neural network combined with the weighted prediction of the safety index is used to perform historical trajectory safety analysis to obtain historical trajectory safety assessment data;

[0015] The attention sequence optimization long short-term memory neural network combined with the weighted prediction of the safety index specifically includes a safety index predefined unit, an attention sequence subnet, a self-attention feedforward subnet and a long short-term memory subnet;

[0016] The safety index predefinition unit is used to set the safety index of the specific predicted vehicle collision trajectory and perform specific safety analysis and prediction based on the safety index;

[0017] The attention sequence subnet is used to construct a codec structure to optimize the data information flow processing capability of time series data and enhance the processing performance of subsequent long short-term memory neural networks for complex data;

[0018] The self-attention feedforward subnetwork is used to optimize the information transfer between subnetworks and improve the processing speed of sequence data;

[0019] The long short-term memory subnet is used as a basic subnet for historical trajectory security analysis and prediction tasks;

[0020] The step of using the attention sequence optimized long short-term memory neural network combined with the weighted prediction of the safety index to perform the historical trajectory safety analysis and obtain the historical trajectory safety assessment data includes:

[0021] Step S31: pre-definition of safety indicators, specifically defining a combination of safety analysis prediction indicators, and performing weighted pre-design of prediction results to obtain a pre-defined safety analysis prediction indicator set, including the following steps:

[0022] Step S311: defining a collision remaining time prediction index, which is used to indicate the remaining time when the current vehicle collides with the previous vehicle in the position reference sequence data if the current vehicle continues to maintain the current speed, specifically calculating a collision remaining time judgment parameter, and defining the collision remaining time prediction index based on the collision remaining time judgment parameter;

[0023] Step S312: defining a collision risk prediction index, which is used to determine the risk probability of a collision between the current vehicle and the previous vehicle based on the relative speed and relative acceleration, specifically calculating a collision risk judgment parameter, and defining a collision risk prediction index based on the collision risk judgment parameter;

[0024] Step S313: defining a collision avoidance deceleration rate prediction index, which is used to represent the minimum deceleration rate required to avoid a collision when the speed of the current vehicle is higher than that of the previous vehicle, specifically calculating a collision avoidance deceleration rate judgment parameter, and defining the collision avoidance deceleration rate prediction index based on the collision avoidance deceleration rate judgment parameter;

[0025] Step S314: defining an emergency deceleration safety distance prediction index, which is used to represent the distance between the previous vehicle and the current vehicle when the two vehicles are completely stopped when both vehicles are performing emergency deceleration, specifically calculating an emergency deceleration potential collision judgment parameter, and defining the emergency deceleration safety distance prediction index based on the emergency deceleration potential collision judgment parameter;

[0026] Step S315: defining a safety analysis prediction indicator set, specifically, defining a safety analysis prediction indicator set by defining a collision remaining time prediction indicator, defining a collision risk prediction indicator, defining a collision avoidance deceleration rate prediction indicator, and defining an emergency deceleration safety distance prediction indicator, to obtain a predefined safety analysis prediction indicator set;

[0027] Step S316: Prediction result weighted pre-design, specifically, performing weighted calculation pre-design of prediction results according to the predefined safety analysis prediction indicator set, and obtaining a predefined weighted safety analysis prediction output, the calculation formula is:

[0028] ;

[0029] In the formula, HPOUTPUT is the predefined weighted safety analysis prediction output, I is the total number of vehicles to be analyzed in the optimized safety assessment dataset, i is the vehicle index, is the collision remaining time weight, TT i is the collision remaining time judgment parameter, is the collision risk weight, MT i is the collision risk judgment parameter, is the collision avoidance deceleration rate weight, DA i is the collision avoidance deceleration rate judgment parameter, is the potential collision weight of emergency deceleration, PI i It is the potential collision judgment parameter for emergency deceleration;

[0030] Step S32: constructing an attention sequence subnet, specifically constructing a sequence-to-sequence model combined with an attention mechanism including an encoder, an attention classifier, and a decoder as the attention sequence subnet, and obtaining encoding and decoding sequence feature data of the historical trajectory data by constructing the attention sequence subnet;

[0031] Step S33: construct a self-attention feedforward subnet, specifically, construct a self-attention mechanism to calculate the query matrix, key matrix and value matrix, encode the sequence data through position encoding, and construct a transformer model by constructing a feedforward layer to obtain the output of the self-attention feedforward subnet;

[0032] Step S34: constructing a long short-term memory subnet, specifically constructing a standard long short-term memory neural network as a long short-term memory subnet, and constructing the long short-term memory subnet as a basic subnet for historical trajectory security analysis and prediction tasks;

[0033] The calculation formula for constructing the standard long short-term memory neural network is:

[0034] ;

[0035] In the formula, f t is the output of the forget gate at time t, sig(·) is the S-type activation function, W f is the forget gate weight, h t-1 is the temporal hidden state at the previous moment t-1, h L is the original input feature sequence, which is used to represent the initial feature dataset in the optimized security assessment dataset, [·,·] is the vector concatenation operator, and B f is the forget gate weight;

[0036] I t is the input gate output at time t, W i is the input gate weight, B i is the input gate weight;

[0037] O t is the output gate output at time t, W o is the output gate weight, B o is the output gate weight;

[0038] is the candidate cell state at time t, tanh(·) is the hyperbolic tangent function, W c is the cell state weight, B c is the cell state bias term;

[0039] C t is the cell state at time t, is the element-wise multiplication operator;

[0040] h t is the temporal hidden state at time t, which is used to represent the final output of the standard long short-term memory neural network;

[0041] Step S35: historical trajectory security analysis model training, specifically, through the pre-definition of the security indicators, the construction of the attention sequence subnet, the construction of the self-attention feedforward subnet and the construction of the long short-term memory subnet, the historical trajectory security analysis model training is performed to obtain the historical trajectory security analysis model Model HP ;

[0042] Step S36: historical trajectory safety analysis, specifically, using the historical trajectory safety analysis model Model based on the optimized safety assessment data set HP , conduct historical trajectory safety analysis and obtain historical trajectory safety assessment data;

[0043] The historical trajectory safety assessment data specifically includes collision remaining time prediction data, collision risk prediction data, collision avoidance deceleration rate prediction data and emergency deceleration safety distance prediction data.

[0044] Further, in step S4, the intention trajectory safety analysis is used for automatic safety prediction of the lane change decision intention, specifically, based on the optimized safety assessment data set, an integrated network of a deep neural network combined with a self-attention mechanism and a long short-term memory neural network is used to perform the intention trajectory safety analysis to obtain the intention trajectory safety assessment data;

[0045] The integrated network of the deep neural network and the long short-term memory neural network combined with the self-attention mechanism specifically includes a deep neural subnet, a long short-term memory subnet, a self-attention layer and an integrated output layer;

[0046] The deep neural subnet is used to extract high-dimensional features of the optimized security assessment data set and perform preliminary predictive analysis;

[0047] The long short-term memory subnet is used to process temporal features and serve as a basic subnet for security analysis of intention trajectories;

[0048] The self-attention layer is used to enhance the network's ability to process data features at different times;

[0049] The integrated output layer is used to integrate the output results of the deep neural subnet and the long short-term memory subnet to obtain the final prediction output;

[0050] The step of using an integrated network of a deep neural network combined with a self-attention mechanism and a long short-term memory neural network to perform intention trajectory security analysis and obtain intention trajectory security assessment data includes:

[0051] Step S41: constructing a deep neural subnet, specifically constructing a deep neural network including an input layer, a hidden layer and an output layer, and activating it through an S-type activation function to construct the deep neural subnet and obtain a predicted output of the deep neural subnet;

[0052] Step S42: constructing a long short-term memory subnet, specifically, constructing the long short-term memory subnet according to the operation of step S34, wherein the prediction output of the deep neural subnet is used as the original input feature sequence of the long short-term memory subnet to obtain the long short-term memory subnet and obtain the prediction output of the long short-term memory subnet;

[0053] Step S43: constructing a self-attention layer, specifically by calculating a query matrix, a key matrix, and a value matrix to construct a self-attention mechanism, wherein the prediction output of the long short-term memory subnetwork is specifically used as the query matrix, the key matrix, and the value matrix of the self-attention layer to construct the self-attention layer and obtain the self-attention output;

[0054] Step S44: constructing an integrated output layer, specifically, integrating the deep neural subnet prediction output, the long short-term memory subnet prediction output and the self-attention output according to the fully connected weights to obtain integrated output data, and the calculation formula is:

[0055] y=sig(W out A+B out );

[0056] Where y is the integrated output data, sig(·) is the S-type activation function, and W out is the fully connected weight of the integrated output layer, A is the self-attention output, B out is the fully connected bias term of the integrated output layer;

[0057] Step S45: Intention trajectory security analysis model training, specifically, by constructing the deep neural subnet, the long short-term memory subnet, the self-attention layer and the integrated output layer, the intention trajectory security analysis model training is performed to obtain the intention trajectory security analysis model Model AP ;

[0058] Step S46: Intention trajectory security analysis, specifically, using the intention trajectory security analysis model Model based on the optimized security assessment data set AP , perform intention trajectory security analysis and obtain intention trajectory security assessment data;

[0059] The intended trajectory safety assessment data specifically includes left lane change risk prediction data, straight driving risk prediction data along the lane, and right lane change risk prediction data.

[0060] Further, in step S5, the vehicle safety assessment is used to combine the safety analysis results of the historical trajectory and the intended trajectory to perform an overall assessment of vehicle safety, specifically combining the historical trajectory safety assessment data and the intended trajectory safety assessment data to perform a vehicle safety assessment to obtain vehicle safety assessment data;

[0061] The vehicle safety assessment data specifically includes current driving trajectory safety analysis data and lane change intention decision driving trajectory safety analysis data;

[0062] The current driving trajectory safety analysis data is used to analyze the safety of the vehicle while driving without changing lanes; the lane change intention decision driving trajectory safety analysis data is used to analyze the safety of the vehicle after making a lane change decision.

[0063] The vehicle safety assessment system based on artificial intelligence provided by the present invention includes a data collection module, a data storage module, a trajectory analysis module and a safety assessment module;

[0064] The data collection module is used for data collection, obtains original security assessment data through data collection, and sends the original security assessment data to the data storage module;

[0065] The data storage module is used for data preprocessing and storing process data, obtaining an optimized safety assessment data set through data preprocessing, and sending the optimized safety assessment data set to the trajectory analysis module;

[0066] The trajectory analysis module is used for trajectory safety analysis, obtains historical trajectory safety assessment data and intended trajectory safety assessment data through historical trajectory safety analysis and intended trajectory safety analysis, and sends the historical trajectory safety assessment data and the intended trajectory safety assessment data to the safety assessment module;

[0067] The safety assessment module is used for vehicle safety assessment, and obtains vehicle safety assessment data through vehicle safety assessment.

[0068] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0069] (1) In view of the existing vehicle safety assessment methods, there are traditional methods for vehicle safety assessment that focus on a large amount of analysis of historical driving process data, thereby achieving a general safety analysis of vehicle driving. However, this analysis method cannot cope with the specific driving trajectory changes during vehicle driving. At the same time, for real-time driving vehicles, the lane change decisions and collision avoidance strategies executed by the driver cannot be analyzed, predicted and controlled in a timely and effective manner, thereby reducing the timeliness and availability of vehicle safety analysis. This solution creatively adopts a two-way vehicle driving safety analysis strategy that combines historical trajectory analysis and intention trajectory analysis, and then uses an artificial intelligence method improved for specific tasks to realize the inclusion of the lane change decisions and collision avoidance strategies executed by the driver into the consideration scope of vehicle safety analysis, thereby providing a more comprehensive exploration of the analysis depth of vehicle safety analysis;

[0070] (2) In the existing historical trajectory safety analysis methods, the data of historical trajectory safety analysis is mostly time series data. Although the traditional long short-term memory neural network can process time series data and features, the general or simple binary classification and multi-classification methods cannot effectively provide highly interpretable result feedback for historical trajectory safety analysis, which leads to the technical problem that the intuitiveness and effectiveness of safety analysis are insufficient. This scheme creatively uses the attention sequence optimization long short-term memory neural network combined with the weighted prediction of safety indicators to perform historical trajectory safety analysis. By constructing four different safety indicators and using them as the prediction results and weighted processing respectively, the prediction depth of historical trajectory safety analysis is improved, and the overall usability and reference of safety analysis results are also improved;

[0071] (3) In the existing intention trajectory safety analysis methods, there is a technical problem that the intention trajectory safety analysis methods need to ensure the prediction accuracy of the model while considering real-time support to improve the prediction efficiency. The traditional technology uses large-scale data solutions, which need to be improved in timeliness, while reducing the data scale may lead to insufficient accuracy. This solution creatively uses artificial data label optimization and preprocessing optimization, combined with the integrated network of deep neural network with self-attention mechanism and long short-term memory neural network to perform intention trajectory safety analysis, and uses artificial labeling to assist the integrated network to perform intention trajectory safety analysis, thereby improving the prediction efficiency and accuracy at the same time, providing strong data support for the subsequent two-way prediction integration results, and also providing a feasible and reference expansion idea for the direction of vehicle safety analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic diagram of the process of the vehicle safety assessment method based on artificial intelligence provided by the present invention;

[0073] Figure 2 A schematic diagram of a vehicle safety assessment system based on artificial intelligence provided by the present invention;

[0074] Figure 3 This is a schematic diagram of the process of data preprocessing in step S2;

[0075] Figure 4 This is a flow chart of historical trajectory safety analysis in step S3;

[0076] Figure 5 Schematic diagram of the process of intention trajectory safety analysis in step S4.

[0077] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0079] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0080] Example 1, see Figure 1 The vehicle safety assessment method based on artificial intelligence provided by the present invention comprises the following steps:

[0081] Step S1: data collection;

[0082] Step S2: data preprocessing;

[0083] Step S3: historical trajectory safety analysis;

[0084] Step S4: security analysis of intention trajectory;

[0085] Step S5: Vehicle safety assessment.

[0086] By performing the above operations, in the existing vehicle safety assessment methods, there are traditional methods for vehicle safety assessment that focus on a large amount of analysis of historical driving process data, thereby realizing a general safety analysis of vehicle driving. However, this analysis method cannot cope with the specific driving trajectory changes during vehicle driving. At the same time, for real-time driving vehicles, the lane change decisions and collision avoidance strategies executed by the driver cannot be timely and effectively analyzed, predicted and controlled for feedback, thereby reducing the timeliness and availability of vehicle safety analysis. Technical problems, this solution creatively adopts a two-way vehicle driving safety analysis strategy that combines historical trajectory analysis and intention trajectory analysis, and then uses artificial intelligence methods improved for specific tasks, so as to realize the inclusion of the lane change decisions and collision avoidance strategies executed by the driver into the consideration scope of vehicle safety analysis, thereby providing a more sufficient exploration of the analysis depth of vehicle safety analysis.

[0087] Example 2, see Figure 1 and Figure 2 In step S1, the data collection is used to collect the original data set required for vehicle safety assessment, specifically from the driving record data, vehicle-mounted sensors and vehicle accident data set, to obtain the original data for safety assessment through data collection;

[0088] The safety assessment raw data specifically includes vehicle trajectory data, vehicle sensor data, environmental sensor data and historical accident data;

[0089] The vehicle trajectory data specifically includes speed sequence data, acceleration sequence data, direction sequence data and position reference sequence data;

[0090] The vehicle sensor data specifically includes brake status data, tire pressure data and engine status data;

[0091] The environmental sensor data specifically includes basic weather data, basic road condition data and traffic flow reference data;

[0092] The historical accident data specifically includes accident type data, accident time data, accident location data and accident environment data.

[0093] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to optimize the original data, specifically, manually perform data label optimization, data cleaning and data alignment operations on the original security assessment data in sequence to obtain an optimized security assessment data set, including the following steps:

[0094] Step S21: manual data label optimization, specifically, labeling the vehicle driving intention trajectory data according to the vehicle trajectory data and vehicle sensor data in the safety assessment original data to obtain vehicle intention trajectory data, and incorporating the vehicle intention trajectory data into the safety assessment original data to obtain label optimized original data;

[0095] The vehicle intention trajectory data specifically includes left lane change data, straight driving data along the lane, right lane change data and lane change sign data;

[0096] The lane change sign data specifically includes lane change start data, lane change middle data and lane change end data;

[0097] The lane change start data specifically includes start vehicle trajectory data and start vehicle sensor data;

[0098] The lane change mid-segment data specifically includes mid-segment vehicle trajectory data and mid-segment vehicle sensor data;

[0099] The lane change terminal data specifically includes terminal vehicle trajectory data and terminal vehicle sensor data;

[0100] Step S22: data cleaning, specifically, sequentially performing missing value processing, outlier processing and noise processing on the label optimized original data, and performing data standardization to obtain a cleaned and optimized data set;

[0101] Step S23: data alignment, specifically, aligning the time series data in the cleaned and optimized data set in time steps to obtain an aligned and optimized data set;

[0102] Step S24: preprocessing optimization, specifically, performing data preprocessing optimization through the artificial data label optimization, the data cleaning and the data alignment, and obtaining an aligned optimized data set, performing preliminary feature extraction using an artificial feature extraction method, and obtaining an optimized security assessment data set;

[0103] The optimized security assessment dataset specifically includes an alignment optimization dataset and an initial feature dataset;

[0104] The initial feature data set specifically includes features of the number of emergency brakes, the number of sharp turns, the range of vehicle speed fluctuations, and lane change reference features.

[0105] Example 4, see Figure 1 , Figure 2 and Figure 4This embodiment is based on the above embodiment. In step S3, the historical trajectory safety analysis is used to perform automated collision precursor safety analysis in combination with historical data of vehicle driving trajectories. Specifically, the historical trajectory safety analysis is performed using an attention sequence optimized long short-term memory neural network combined with weighted prediction of safety indicators based on the optimized safety assessment data set to obtain historical trajectory safety assessment data;

[0106] The attention sequence optimization long short-term memory neural network combined with the weighted prediction of the safety index specifically includes a safety index predefined unit, an attention sequence subnet, a self-attention feedforward subnet and a long short-term memory subnet;

[0107] The safety index predefinition unit is used to set the safety index of the specific predicted vehicle collision trajectory and perform specific safety analysis and prediction based on the safety index;

[0108] The attention sequence subnet is used to construct a codec structure to optimize the data information flow processing capability of time series data and enhance the processing performance of subsequent long short-term memory neural networks for complex data;

[0109] The self-attention feedforward subnetwork is used to optimize the information transfer between subnetworks and improve the processing speed of sequence data;

[0110] The long short-term memory subnet is used as a basic subnet for historical trajectory security analysis and prediction tasks;

[0111] The step of using the attention sequence optimized long short-term memory neural network combined with the weighted prediction of the safety index to perform the historical trajectory safety analysis and obtain the historical trajectory safety assessment data includes:

[0112] Step S31: pre-definition of safety indicators, specifically defining a combination of safety analysis prediction indicators, and performing weighted pre-design of prediction results to obtain a pre-defined safety analysis prediction indicator set, including the following steps:

[0113] Step S311: defining a collision remaining time prediction index, which is used to indicate the remaining time when the current vehicle collides with the previous vehicle in the position reference sequence data if the current vehicle continues to maintain the current speed, specifically calculating a collision remaining time judgment parameter, and defining the collision remaining time prediction index based on the collision remaining time judgment parameter;

[0114] The calculation formula of the collision remaining time judgment parameter is:

[0115] ;

[0116] Where, TT i is the collision remaining time judgment parameter, is the infinity symbol, used to indicate no collision safety risk, i is the vehicle index, x i-1 is the position reference sequence data of the previous vehicle, x i is the current vehicle position reference sequence data, v i is the current vehicle speed data, v i-1 is the speed data of the previous vehicle, is the length of the current vehicle;

[0117] Step S312: defining a collision risk prediction index, which is used to determine the risk probability of a collision between the current vehicle and the previous vehicle based on the relative speed and relative acceleration, specifically calculating a collision risk judgment parameter, and defining a collision risk prediction index based on the collision risk judgment parameter;

[0118] The calculation formula of the collision risk judgment parameter is:

[0119] ;

[0120] In the formula, MT i is the collision risk judgment parameter, where i is the vehicle index, is the relative speed between the current vehicle and the previous vehicle, is the relative acceleration between the current vehicle and the previous vehicle, is the relative distance between the current vehicle and the previous vehicle, is the length of the current vehicle;

[0121] Step S313: defining a collision avoidance deceleration rate prediction index, which is used to represent the minimum deceleration rate required to avoid a collision when the speed of the current vehicle is higher than that of the previous vehicle, specifically calculating a collision avoidance deceleration rate judgment parameter, and defining the collision avoidance deceleration rate prediction index based on the collision avoidance deceleration rate judgment parameter;

[0122] The calculation formula of the collision avoidance deceleration rate judgment parameter is:

[0123] ;

[0124] In the formula, DA i is the collision avoidance deceleration rate judgment parameter, 0 is the deceleration rate 0 identifier, which is used to indicate that there is no need to decelerate at this time, i is the vehicle index, x i-1 is the position reference sequence data of the previous vehicle, x i is the current vehicle position reference sequence data, v i is the current vehicle speed data, v i-1 is the speed data of the previous vehicle, is the length of the current vehicle;

[0125] Step S314: defining an emergency deceleration safety distance prediction index, which is used to represent the distance between the previous vehicle and the current vehicle when the two vehicles are completely stopped when both vehicles are performing emergency deceleration, specifically calculating an emergency deceleration potential collision judgment parameter, and defining the emergency deceleration safety distance prediction index based on the emergency deceleration potential collision judgment parameter;

[0126] The calculation formula of the emergency deceleration potential collision judgment parameter is:

[0127] ;

[0128] In the formula, is the emergency deceleration potential collision judgment parameter, v i is the current vehicle speed data, v i-1 is the speed data of the previous vehicle, a is the emergency stop deceleration rate parameter, is the relative distance between the current vehicle and the previous vehicle, is the emergency stop response weight parameter;

[0129] Step S315: defining a safety analysis prediction indicator set, specifically, defining a safety analysis prediction indicator set by defining a collision remaining time prediction indicator, defining a collision risk prediction indicator, defining a collision avoidance deceleration rate prediction indicator, and defining an emergency deceleration safety distance prediction indicator, to obtain a predefined safety analysis prediction indicator set;

[0130] The calculation formula of the predefined safety analysis prediction indicator set is:

[0131] SSM i ={TT i ,MT i ,DA i ,PI i};

[0132] In the formula, SSM i is a set of predefined safety analysis prediction indicators, TT i Is the collision remaining time judgment parameter, MT i is the collision risk judgment parameter, DA i is the collision avoidance deceleration rate judgment parameter, PI i It is the potential collision judgment parameter for emergency deceleration;

[0133] Step S316: Prediction result weighted pre-design, specifically, performing weighted calculation pre-design of prediction results according to the predefined safety analysis prediction indicator set, and obtaining a predefined weighted safety analysis prediction output, the calculation formula is:

[0134] ;

[0135] In the formula, HP OUTPUT is the predefined weighted safety analysis prediction output, I is the total number of vehicles to be analyzed in the optimized safety assessment dataset, i is the vehicle index, is the collision remaining time weight, TT i is the collision remaining time judgment parameter, is the collision risk weight, MT i is the collision risk judgment parameter, is the collision avoidance deceleration rate weight, DA i is the collision avoidance deceleration rate judgment parameter, is the potential collision weight of emergency deceleration, PI i It is the potential collision judgment parameter for emergency deceleration;

[0136] Step S32: constructing an attention sequence subnet, specifically constructing a sequence-to-sequence model combined with an attention mechanism including an encoder, an attention classifier, and a decoder as the attention sequence subnet, and obtaining encoding and decoding sequence feature data of the historical trajectory data by constructing the attention sequence subnet;

[0137] The calculation formula for constructing the attention sequence subnet is:

[0138] ;

[0139] In the formula, h t is the temporal hidden state at time t, f(·) is the encoder function, x t is a model input data sample, used to represent the aligned optimized dataset in the optimized security assessment dataset, h t-1 is the temporal hidden state at the previous moment t-1;

[0140] a t is the attention score at time t, softmax(·) is the attention classifier function, V T is the attention parameter transposed matrix, tanh(·) is the hyperbolic tangent function, W 1 is the hidden state attention classification weight, W 2 is the decoder state attention classification weight;

[0141] s t is the decoder state vector at time t, c t is the context state vector at time t, is the total length of the input sequence data, is the input sequence data index, is the first The attention score of the input sequence data, It is The temporal hidden state of the input sequence data;

[0142] g(·) is the decoder function, y t-1 is the decoded output at time t-1, s t-1 is the decoder state vector at time t-1;

[0143] y t is the decoding output at time t, h(·) is the output function, and specifically adopts the fully connected layer combined with the predefined weighted security analysis prediction output as the output function;

[0144] The encoder function f(·) and the decoder function g(·) are specifically both long short-term memory neural network functions LSTM(·);

[0145] Step S33: construct a self-attention feedforward subnet, specifically, construct a self-attention mechanism to calculate the query matrix, key matrix and value matrix, encode the sequence data through position encoding, and construct a transformer model by constructing a feedforward layer to obtain the output of the self-attention feedforward subnet;

[0146] Step S34: constructing a long short-term memory subnet, specifically constructing a standard long short-term memory neural network as a long short-term memory subnet, and constructing the long short-term memory subnet as a basic subnet for historical trajectory security analysis and prediction tasks;

[0147] The calculation formula for constructing the standard long short-term memory neural network is:

[0148] ;

[0149] In the formula, f t is the output of the forget gate at time t, sig(·) is the S-type activation function, W f is the forget gate weight, h t-1 is the temporal hidden state at the previous moment t-1, h L is the original input feature sequence, which is used to represent the initial feature dataset in the optimized security assessment dataset, [·,·] is the vector concatenation operator, and B f is the forget gate weight;

[0150] I t is the input gate output at time t, W i is the input gate weight, B i is the input gate weight;

[0151] O t is the output gate output at time t, W o is the output gate weight, B o is the output gate weight;

[0152] is the candidate cell state at time t, tanh(·) is the hyperbolic tangent function, W c is the cell state weight, B c is the cell state bias term;

[0153] C t is the cell state at time t, is the element-wise multiplication operator;

[0154] h t is the temporal hidden state at time t, which is used to represent the final output of the standard long short-term memory neural network;

[0155] Step S35: historical trajectory security analysis model training, specifically, through the pre-definition of the security indicators, the construction of the attention sequence subnet, the construction of the self-attention feedforward subnet and the construction of the long short-term memory subnet, the historical trajectory security analysis model training is performed to obtain the historical trajectory security analysis model Model HP ;

[0156] Step S36: historical trajectory safety analysis, specifically, using the historical trajectory safety analysis model Model based on the optimized safety assessment data set HP , conduct historical trajectory safety analysis and obtain historical trajectory safety assessment data;

[0157] The historical trajectory safety assessment data specifically includes collision remaining time prediction data, collision risk prediction data, collision avoidance deceleration rate prediction data and emergency deceleration safety distance prediction data.

[0158] By performing the above operations, in view of the fact that in the existing historical trajectory safety analysis methods, the data of historical trajectory safety analysis are mostly time series data, and although the traditional long short-term memory neural network can process time series data and features, the general or simple binary classification and multi-classification methods cannot effectively provide highly interpretable result feedback for the historical trajectory safety analysis, which leads to the technical problem of insufficient intuitiveness and effectiveness of safety analysis. This scheme creatively adopts the attention sequence optimized long short-term memory neural network combined with weighted prediction of safety indicators to perform historical trajectory safety analysis. By constructing four different safety indicators and using them as prediction results and weighted processing respectively, the prediction depth of historical trajectory safety analysis is improved, and the overall availability and reference of safety analysis results are also improved.

[0159] Example 5, see Figure 1 , Figure 2 and Figure 5In step S4, the intention trajectory safety analysis is used for automatic safety prediction of the lane change decision intention, specifically, based on the optimized safety assessment data set, an integrated network of a deep neural network combined with a self-attention mechanism and a long short-term memory neural network is used to perform the intention trajectory safety analysis to obtain the intention trajectory safety assessment data;

[0160] The integrated network of the deep neural network and the long short-term memory neural network combined with the self-attention mechanism specifically includes a deep neural subnet, a long short-term memory subnet, a self-attention layer and an integrated output layer;

[0161] The deep neural subnet is used to extract high-dimensional features of the optimized security assessment data set and perform preliminary predictive analysis;

[0162] The long short-term memory subnet is used to process temporal features and serve as a basic subnet for security analysis of intention trajectories;

[0163] The self-attention layer is used to enhance the network's ability to process data features at different times;

[0164] The integrated output layer is used to integrate the output results of the deep neural subnet and the long short-term memory subnet to obtain the final prediction output;

[0165] The step of using an integrated network of a deep neural network combined with a self-attention mechanism and a long short-term memory neural network to perform intention trajectory security analysis and obtain intention trajectory security assessment data includes:

[0166] Step S41: constructing a deep neural subnet, specifically constructing a deep neural network including an input layer, a hidden layer and an output layer, and activating it through an S-type activation function to construct the deep neural subnet and obtain a predicted output of the deep neural subnet;

[0167] Step S42: constructing a long short-term memory subnet, specifically, constructing the long short-term memory subnet according to the operation of step S34, wherein the prediction output of the deep neural subnet is used as the original input feature sequence of the long short-term memory subnet to obtain the long short-term memory subnet and obtain the prediction output of the long short-term memory subnet;

[0168] Step S43: constructing a self-attention layer, specifically by calculating a query matrix, a key matrix, and a value matrix to construct a self-attention mechanism, wherein the prediction output of the long short-term memory subnetwork is specifically used as the query matrix, the key matrix, and the value matrix of the self-attention layer to construct the self-attention layer and obtain the self-attention output;

[0169] Step S44: constructing an integrated output layer, specifically, integrating the deep neural subnet prediction output, the long short-term memory subnet prediction output and the self-attention output according to the fully connected weights to obtain integrated output data, and the calculation formula is:

[0170] y=sig(W out A+B out );

[0171] Where y is the integrated output data, sig(·) is the S-type activation function, and W out is the fully connected weight of the integrated output layer, A is the self-attention output, B out is the fully connected bias term of the integrated output layer;

[0172] Step S45: Intention trajectory security analysis model training, specifically, by constructing the deep neural subnet, the long short-term memory subnet, the self-attention layer and the integrated output layer, the intention trajectory security analysis model training is performed to obtain the intention trajectory security analysis model Model AP ;

[0173] Step S46: Intention trajectory security analysis, specifically, using the intention trajectory security analysis model Model based on the optimized security assessment data set AP , perform intention trajectory security analysis and obtain intention trajectory security assessment data;

[0174] The intended trajectory safety assessment data specifically includes left lane change risk prediction data, straight driving risk prediction data along the lane, and right lane change risk prediction data.

[0175] By performing the above operations, in view of the fact that in the existing intention trajectory safety analysis methods, the intention trajectory safety analysis methods need to ensure the prediction accuracy of the model and also need to consider real-time support to improve the prediction efficiency, while the traditional technology uses large-scale data solutions in terms of timeliness to be improved, and reducing the data scale may lead to technical problems such as insufficient accuracy. This solution creatively adopts artificial data label optimization and preprocessing optimization, combines the deep neural network with self-attention mechanism and the integrated network of long short-term memory neural network to perform intention trajectory safety analysis, and uses artificial labeling to assist the integrated network to perform intention trajectory safety analysis, thereby improving the prediction efficiency and accuracy at the same time, providing strong data support for the subsequent two-way prediction integration results, and also providing a feasible and reference expansion idea for the direction of vehicle safety analysis.

[0176] Example 6, see Figure 1 and Figure 2, this embodiment is based on the above embodiment. In step S5, the vehicle safety assessment is used to combine the safety analysis results of the historical trajectory and the intended trajectory to perform an overall assessment of vehicle safety, specifically combining the historical trajectory safety assessment data and the intended trajectory safety assessment data to perform a vehicle safety assessment and obtain vehicle safety assessment data;

[0177] The vehicle safety assessment data specifically includes current driving trajectory safety analysis data and lane change intention decision driving trajectory safety analysis data;

[0178] The current driving trajectory safety analysis data is used to analyze the safety of the vehicle while driving without changing lanes; the lane change intention decision driving trajectory safety analysis data is used to analyze the safety of the vehicle after making a lane change decision.

[0179] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the vehicle safety assessment system based on artificial intelligence provided by the present invention includes a data collection module, a data storage module, a trajectory analysis module and a safety assessment module;

[0180] The data collection module is used for data collection, obtains original security assessment data through data collection, and sends the original security assessment data to the data storage module;

[0181] The data storage module is used for data preprocessing and storing process data, obtaining an optimized safety assessment data set through data preprocessing, and sending the optimized safety assessment data set to the trajectory analysis module;

[0182] The trajectory analysis module is used for trajectory safety analysis, obtains historical trajectory safety assessment data and intended trajectory safety assessment data through historical trajectory safety analysis and intended trajectory safety analysis, and sends the historical trajectory safety assessment data and the intended trajectory safety assessment data to the safety assessment module;

[0183] The safety assessment module is used for vehicle safety assessment, and obtains vehicle safety assessment data through vehicle safety assessment.

[0184] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0185] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0186] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A vehicle safety assessment method based on artificial intelligence, characterized by: The method comprises the following steps: Step S1: Data collection, obtaining original safety assessment data; Step S2: data preprocessing, specifically, performing manual data label optimization, data cleaning and data alignment operations on the original security assessment data in sequence to obtain an optimized security assessment data set; Step S3: historical trajectory safety analysis, which is used to perform automated collision precursor safety analysis in combination with historical data of vehicle driving trajectories. Specifically, based on the optimized safety assessment data set, an attention sequence optimized long short-term memory neural network combined with weighted prediction of safety indicators is used to perform historical trajectory safety analysis to obtain historical trajectory safety assessment data, including the following steps: Step S31: pre-definition of security indicators; Step S32: construction of attention sequence subnet; Step S33: construction of self-attention feedforward subnet; Step S34: construction of long short-term memory subnet; Step S35: historical trajectory security analysis model training; Step S36: historical trajectory security analysis; In step S31, the safety index is predefined, specifically defining a safety analysis prediction index combination, and performing a weighted pre-design of the prediction results to obtain a predefined safety analysis prediction index set, including the following steps: Step S311: defining a collision remaining time prediction index, which is used to indicate the remaining time when the current vehicle collides with the previous vehicle in the position reference sequence data if the current vehicle continues to maintain the current speed, specifically calculating a collision remaining time judgment parameter, and defining the collision remaining time prediction index based on the collision remaining time judgment parameter; The calculation formula of the collision remaining time judgment parameter is: Where, TT i is the collision remaining time judgment parameter, ∞ is the infinity symbol, which is used to indicate no collision safety risk, i is the vehicle index, x i-1 is the position reference sequence data of the previous vehicle, x i is the current vehicle position reference sequence data, v i is the current vehicle speed data, v i-1 is the speed data of the previous vehicle, l is the length of the current vehicle; Step S312: defining a collision risk prediction index, which is used to determine the risk probability of a collision between the current vehicle and the previous vehicle based on the relative speed and relative acceleration, specifically calculating a collision risk judgment parameter, and defining a collision risk prediction index based on the collision risk judgment parameter; The calculation formula of the collision risk judgment parameter is: In the formula, MT i is the collision risk judgment parameter, where i is the vehicle index, Δv is the relative speed between the current vehicle and the previous vehicle, Δa is the relative acceleration between the current vehicle and the previous vehicle, Δx is the relative distance between the current vehicle and the previous vehicle, and l is the length of the current vehicle; Step S313: defining a collision avoidance deceleration rate prediction index, which is used to represent the minimum deceleration rate required to avoid a collision when the speed of the current vehicle is higher than that of the previous vehicle, specifically calculating a collision avoidance deceleration rate judgment parameter, and defining the collision avoidance deceleration rate prediction index based on the collision avoidance deceleration rate judgment parameter; The calculation formula of the collision avoidance deceleration rate judgment parameter is: In the formula, DA i is the collision avoidance deceleration rate judgment parameter, 0 is the deceleration rate 0 identifier, which is used to indicate that there is no need to decelerate at this time, i is the vehicle index, x i-1 is the position reference sequence data of the previous vehicle, x i is the current vehicle position reference sequence data, v i is the current vehicle speed data, v i-1 is the speed data of the previous vehicle, l is the length of the current vehicle; Step S314: defining an emergency deceleration safety distance prediction index, which is used to represent the distance between the previous vehicle and the current vehicle when the two vehicles are completely stopped when both vehicles are performing emergency deceleration, specifically calculating an emergency deceleration potential collision judgment parameter, and defining the emergency deceleration safety distance prediction index based on the emergency deceleration potential collision judgment parameter; The calculation formula of the emergency deceleration potential collision judgment parameter is: In the formula, PI i is the emergency deceleration potential collision judgment parameter, v i is the current vehicle speed data, v i-1 is the speed data of the previous vehicle, a is the emergency stop deceleration rate parameter, Δx is the relative distance between the current vehicle and the previous vehicle, and ε is the emergency stop reaction weight parameter; Step S315: defining a safety analysis prediction indicator set, specifically, defining a safety analysis prediction indicator set by defining a collision remaining time prediction indicator, defining a collision risk prediction indicator, defining a collision avoidance deceleration rate prediction indicator, and defining an emergency deceleration safety distance prediction indicator, to obtain a predefined safety analysis prediction indicator set; The calculation formula of the predefined safety analysis prediction indicator set is: SSM i ={TT i ,MT i ,YES i ,PI i }; In the formula, SSM i is a set of predefined safety analysis prediction indicators, TT i Is the collision remaining time judgment parameter, MT i is the collision risk judgment parameter, DA i is the collision avoidance deceleration rate judgment parameter, PI i It is the potential collision judgment parameter for emergency deceleration; Step S316: Prediction result weighted pre-design, specifically, performing weighted calculation pre-design of prediction results according to the predefined safety analysis prediction indicator set, and obtaining a predefined weighted safety analysis prediction output, the calculation formula is: In the formula, HP OUTPUT is the predefined weighted safety analysis prediction output, I is the total number of vehicles to be analyzed in the optimized safety assessment dataset, i is the vehicle index, α is the residual time weight for collision, and TT i is the collision remaining time judgment parameter, β is the collision risk weight, MT i is the collision risk judgment parameter, γ is the collision avoidance deceleration rate weight, DA i is the collision avoidance deceleration rate judgment parameter, δ is the potential collision weight of emergency deceleration, PI i It is the potential collision judgment parameter for emergency deceleration; Step S4: intention trajectory safety analysis, which is used for automatic safety prediction of lane change decision intention, specifically, based on the optimized safety assessment data set, an integrated network of a deep neural network combined with a self-attention mechanism and a long short-term memory neural network is used to perform intention trajectory safety analysis to obtain intention trajectory safety assessment data; Step S5: vehicle safety assessment, specifically, combining the historical trajectory safety assessment data and the intended trajectory safety assessment data to perform vehicle safety assessment and obtain vehicle safety assessment data.

2. The vehicle safety assessment method based on artificial intelligence according to claim 1, characterized in that: In step S3, the attention sequence optimization long short-term memory neural network combined with the weighted prediction of the safety index specifically includes a safety index predefined unit, an attention sequence subnet, a self-attention feedforward subnet and a long short-term memory subnet; The safety index predefinition unit is used to set the safety index of the specific predicted vehicle collision trajectory and perform specific safety analysis and prediction based on the safety index; The attention sequence subnet is used to construct a codec structure to optimize the data information flow processing capability of time series data and enhance the processing performance of subsequent long short-term memory neural networks for complex data; The self-attention feedforward subnetwork is used to optimize the information transfer between subnetworks and improve the processing speed of sequence data; The long short-term memory subnet is used as a basic subnet for historical trajectory security analysis and prediction tasks.

3. The vehicle safety assessment method based on artificial intelligence according to claim 2 is characterized in that: In step S32, the attention sequence subnet is constructed, specifically, a sequence-to-sequence model combined with an attention mechanism including an encoder, an attention classifier, and a decoder is constructed as the attention sequence subnet, and the encoding and decoding sequence feature data of the historical trajectory data is obtained by constructing the attention sequence subnet; In step S33, the self-attention feedforward subnetwork is constructed, specifically, a self-attention mechanism is constructed to calculate a query matrix, a key matrix, and a value matrix, sequence data is encoded by position encoding, and a transformer model is constructed by constructing a feedforward layer to obtain the output of the self-attention feedforward subnetwork; In step S34, the construction of the long short-term memory subnet is specifically to construct a standard long short-term memory neural network as the long short-term memory subnet, and by constructing the long short-term memory subnet, it is used as the basic subnet for historical trajectory security analysis and prediction tasks.

4. The vehicle safety assessment method based on artificial intelligence according to claim 3 is characterized in that: In step S35, the historical trajectory security analysis model training is specifically performed by predefining the security indicators, constructing the attention sequence subnet, constructing the self-attention feedforward subnet and constructing the long short-term memory subnet to obtain the historical trajectory security analysis model Model HP ; In step S36, the historical trajectory safety analysis is specifically performed based on the optimized safety assessment data set using the historical trajectory safety analysis model Model HP , conduct historical trajectory safety analysis and obtain historical trajectory safety assessment data; The historical trajectory safety assessment data specifically includes collision remaining time prediction data, collision risk prediction data, collision avoidance deceleration rate prediction data and emergency deceleration safety distance prediction data.

5. The vehicle safety assessment method based on artificial intelligence according to claim 4 is characterized in that: In step S4, the integrated network of the deep neural network and the long short-term memory neural network combined with the self-attention mechanism specifically includes a deep neural subnet, a long short-term memory subnet, a self-attention layer and an integrated output layer; The deep neural subnet is used to extract high-dimensional features of the optimized security assessment data set and perform preliminary predictive analysis; The long short-term memory subnet is used to process temporal features and serve as a basic subnet for security analysis of intention trajectories; The self-attention layer is used to enhance the network's ability to process data features at different times; The integrated output layer is used to integrate the output results of the deep neural subnet and the long short-term memory subnet to obtain the final prediction output; The step of using an integrated network of a deep neural network combined with a self-attention mechanism and a long short-term memory neural network to perform intention trajectory security analysis and obtain intention trajectory security assessment data includes: Step S41: constructing a deep neural subnet, specifically constructing a deep neural network including an input layer, a hidden layer and an output layer, and activating it through an S-type activation function to construct the deep neural subnet and obtain a predicted output of the deep neural subnet; Step S42: constructing a long short-term memory subnet, specifically, constructing the long short-term memory subnet according to the operation of step S34, wherein the prediction output of the deep neural subnet is used as the original input feature sequence of the long short-term memory subnet to obtain the long short-term memory subnet and obtain the prediction output of the long short-term memory subnet; Step S43: constructing a self-attention layer, specifically by calculating a query matrix, a key matrix, and a value matrix to construct a self-attention mechanism, wherein the prediction output of the long short-term memory subnetwork is specifically used as the query matrix, the key matrix, and the value matrix of the self-attention layer to construct the self-attention layer and obtain the self-attention output; Step S44: constructing an integrated output layer, specifically, integrating the deep neural subnet prediction output, the long short-term memory subnet prediction output and the self-attention output according to the fully connected weights to obtain integrated output data, and the calculation formula is: y=sig(W out A+B out ); Where y is the integrated output data, sig(·) is the S-type activation function, and W out is the fully connected weight of the integrated output layer, A is the self-attention output, B out is the fully connected bias term of the integrated output layer; Step S45: Intention trajectory security analysis model training, specifically, by constructing the deep neural subnet, the long short-term memory subnet, the self-attention layer and the integrated output layer, the intention trajectory security analysis model training is performed to obtain the intention trajectory security analysis model Model AP ; Step S46: Intention trajectory security analysis, specifically, using the intention trajectory security analysis model Model based on the optimized security assessment data set AP , perform intention trajectory security analysis and obtain intention trajectory security assessment data; The intended trajectory safety assessment data specifically includes left lane change risk prediction data, straight driving risk prediction data along the lane, and right lane change risk prediction data.

6. The vehicle safety assessment method based on artificial intelligence according to claim 5 is characterized in that: In step S5, the vehicle safety assessment is used to combine the safety analysis results of the historical trajectory and the intended trajectory to conduct an overall assessment of vehicle safety, specifically combining the historical trajectory safety assessment data and the intended trajectory safety assessment data to conduct a vehicle safety assessment to obtain vehicle safety assessment data; The vehicle safety assessment data specifically includes current driving trajectory safety analysis data and lane change intention decision driving trajectory safety analysis data; The current driving trajectory safety analysis data is used to analyze the safety of the vehicle while driving without changing lanes; the lane change intention decision driving trajectory safety analysis data is used to analyze the safety of the vehicle after making a lane change decision.

7. A vehicle safety assessment system based on artificial intelligence, used to implement the vehicle safety assessment method based on artificial intelligence as described in any one of claims 1 to 6, characterized in that: It includes data collection module, data storage module, trajectory analysis module and safety assessment module.

8. The vehicle safety assessment system based on artificial intelligence according to claim 7, characterized in that: The data collection module is used for data collection, obtains original security assessment data through data collection, and sends the original security assessment data to the data storage module; The data storage module is used for data preprocessing and storing process data, obtaining an optimized safety assessment data set through data preprocessing, and sending the optimized safety assessment data set to the trajectory analysis module; The trajectory analysis module is used for trajectory safety analysis, obtains historical trajectory safety assessment data and intended trajectory safety assessment data through historical trajectory safety analysis and intended trajectory safety analysis, and sends the historical trajectory safety assessment data and the intended trajectory safety assessment data to the safety assessment module; The safety assessment module is used for vehicle safety assessment, and obtains vehicle safety assessment data through vehicle safety assessment.

Images