Lane keeping assist system operation comfort evaluation method based on machine learning

Abstract

The present invention relates to a lane keeping assist system operation comfort evaluation method based on machine learning, including: using an autonomous driving vehicle equipped with a lane keeping assist system to conduct field tests, obtain weather environment information, road geometry and traffic marking information, driving behavior information and driver comfort test results; using a recursive feature elimination algorithm to filter the acquired information as system operation comfort characteristic parameters, and using a receiver operating characteristic curve analysis method to determine its comfort threshold; using weather environment information and road geometry and traffic marking information as input information, using a machine learning method to predict driving behavior information; based on the driving behavior information prediction results, using a machine learning method to construct a comfort evaluation model, thereby evaluating the operation comfort of the lane keeping assist system. This method can fully consider the real-world conditions when the lane keeping assist system is running, and effectively improve the accuracy of the lane keeping assist system operation comfort evaluation results.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle driving assistance system testing, and in particular relates to a method for evaluating the operating comfort of a lane keeping assistance system based on machine learning. Background Art

[0002] Lane keeping assist system is one of the key driving assistance system functions and a key component of autonomous driving technology. It assists the driver to keep the vehicle in the lane by performing lateral dynamic driving tasks, such as intervening in steering wheel control to achieve lateral vehicle manipulation, monitoring the lateral road traffic environment of the vehicle's driving path, etc. It is expected to provide an effective way to reduce the driver's lateral manipulation burden, improve the driver's situational awareness, and enhance the lateral driving safety of the vehicle. It should be noted that the willingness of vehicle users to accept such systems is a prerequisite for their successful implementation and a key design goal of the system. Comfort, as an important content to measure the vehicle user's experience of the driving assistance system, directly reflects the user's acceptance of the system's functional performance and also reflects the difference between the user's expectations and the actual system functions. The comfort of the lane keeping assist system during operation depends to a large extent on the driver's experience of the vehicle's driving behavior under the influence of the real road environment.

[0003] As the market penetration of lane keeping assist systems increases, existing studies have subjectively evaluated the comfort of system operation based on expert experience, but few studies have proposed objective evaluation methods, and existing studies have not considered the changes in vehicle driving behavior caused by different road environment factors under system operation, thereby affecting the comfort of system operation. Considering that vehicles equipped with lane keeping assist systems have many sensors that collect and store vehicle driving behavior data, machine learning methods can mine this data and quantify driving behavior under different road environment conditions, thereby providing a new solution for objectively evaluating the comfort of lane keeping assist system operation. Summary of the invention

[0004] The purpose of the present invention is to provide a lane keeping assist system operation comfort evaluation method based on machine learning. The method can comprehensively consider the weather environment, road geometry and traffic marking information when the vehicle is driving, use machine learning methods to quantify driving behavior under different road environment conditions, and objectively reflect the impact of different factors on the operation comfort of the lane keeping assist system through changes in driving behavior. At the same time, combined with the driver's subjective evaluation, the operation comfort of the lane keeping assist system is effectively evaluated as a whole, providing a reliable basis for the optimization and upgrading of the lane keeping assist system.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: a lane keeping assist system operation comfort evaluation method based on machine learning, comprising the following steps:

[0006] Step S1, conducting field tests using an autonomous driving vehicle equipped with a lane keeping assist system to obtain weather environment information, road geometry and traffic marking information, driving behavior information, and driver comfort test results;

[0007] Step S2, using a recursive feature elimination algorithm to filter the information obtained in step S1 as a system operation comfort characteristic parameter, and using a receiver operating characteristic curve analysis method to determine its comfort threshold;

[0008] Step S3, using weather environment information and road geometry and traffic marking information as input information, using a machine learning method to predict driving behavior information;

[0009] Step S4: Based on the prediction result of the driving behavior information, a comfort evaluation model is constructed by using a machine learning method, and the comfort evaluation model is used to evaluate the operation comfort of the lane keeping assist system.

[0010] Further, in step S1, the weather environment information includes: weather type and its corresponding weather attribute information; wherein, when the weather type is rainy, the corresponding weather attribute information includes rainfall level and rainfall intensity; when the weather type is snowy, the corresponding weather attribute information includes snowfall level and snowfall intensity; when the weather type is foggy, the corresponding weather attribute information includes fog intensity level and visibility;

[0011] The road geometry and traffic marking information includes: road geometry information and matching traffic marking information; wherein the road geometry information includes road grade, design speed, route direction, slope, curvature radius, road section type, lane width, shoulder width, lane position and number of lanes; the traffic marking information includes indication and warning marking types, lane marking combination types, lane marking gaps, lane marking quality, marking maintenance conditions and marking retro-reflectivity; the road section types include straight lines, flat curves, curved slopes and intersections;

[0012] The driving behavior information includes at least: longitudinal driving speed, lateral driving speed, lateral deviation and rate, heading angle deviation and rate, lateral acceleration and rate of change, steering wheel rotation angle and rate of change, steering wheel torque, and driver takeover frequency and duration;

[0013] The driver comfort test results include: comfort and discomfort.

[0014] Furthermore, the implementation method of step S2 is as follows:

[0015] Step S21, performing data cleaning on the weather environment information, road geometry and traffic marking information, and driving behavior information obtained in step S1;

[0016] Step S22, using weather environment information, road geometry and traffic marking information, and driving behavior information as input features and normalizing them, and using driver comfort test results as output features and normalizing them;

[0017] Step S23, selecting an extreme gradient enhancement model, using a recursive feature elimination algorithm combined with ten-fold cross validation to screen the input information as a system operation comfort feature parameter, and sorting its feature importance;

[0018] Step S24: based on the standardized driver comfort test results, draw receiver operating characteristic curves for each standardized system operation comfort characteristic parameter;

[0019] Step S25, analyzing each receiver operating characteristic curve, comparing the Youden index of each system operation comfort characteristic parameter within the value range, and selecting the parameter value corresponding to the maximum Youden index as the comfort threshold of the system operation comfort characteristic parameter.

[0020] Furthermore, the data cleaning includes removing outliers and filling in missing values.

[0021] Furthermore, the implementation method of step S3 is as follows:

[0022] Step S31, using the weather environment information and road geometry and traffic marking information after data cleaning as input features and standardizing them, using the driving behavior information as output features and standardizing them, and using the input features and the output features that match them as sample pairs;

[0023] Step S32: randomly divide all sample pairs obtained in step S31 according to a set ratio, and use them as a training set and a test set of the machine learning model respectively;

[0024] Step S33, selecting a plurality of machine learning models capable of realizing multi-input-multi-output regression prediction tasks from a classic machine learning model library, respectively using ten-fold cross validation and grid search to optimize model parameters and construct a regression prediction machine learning model; wherein the machine learning models capable of realizing multi-input-multi-output regression prediction tasks include a multi-layer feedforward neural network, a Transformer model, and a convolutional neural network trained according to an error back propagation algorithm;

[0025] Step S34: Select a regression prediction machine learning model performance evaluation index to evaluate the prediction effect of the regression prediction machine learning model, and rank the regression prediction machine learning models from best to worst according to the prediction effect;

[0026] Step S35, extracting the system operation comfort characteristic parameters and their comfort thresholds belonging to the weather environment information and the road geometry and traffic marking information obtained in step S2, and inputting them into the regression prediction machine learning model to obtain the driving behavior information prediction value;

[0027] Step S36, extracting the system operation comfort characteristic parameters and the comfort threshold value belonging to the driving behavior information obtained in step S2, comparing them with the corresponding driving behavior information prediction value obtained in step S35, and sorting the regression prediction machine learning models from small to large according to the absolute value of the difference between the two;

[0028] Step S37, extract the regression prediction machine learning model with the smallest sum of the rankings obtained in step S34 and step S36 as the machine learning model for finally predicting driving behavior information, and predict the driving behavior information through the machine learning model.

[0029] Furthermore, the performance evaluation indicators of the regression prediction machine learning model include root mean square error, mean absolute error and mean absolute percentage error.

[0030] Furthermore, the implementation method of step S4 is as follows:

[0031] Step S41, using the driving behavior information predicted in step S3 as input features and normalizing them, using the driver comfort test results as output features and normalizing them, and using the input features and the output features that match them as sample pairs;

[0032] Step S42: randomly divide all sample pairs obtained in step S41 according to a set ratio, and use them as a training set and a test set of the machine learning model respectively;

[0033] Step S43, selecting a plurality of machine learning models capable of realizing multi-input-single-output classification prediction tasks from a classic machine learning model library, respectively using ten-fold cross validation and grid search to optimize model parameters and construct a classification prediction machine learning model; wherein the machine learning models capable of realizing multi-input-single-output classification prediction tasks include logistic regression, random forest, decision tree, support vector machine, extreme gradient boosting model and neural network;

[0034] Step S44: selecting a classification prediction machine learning model performance evaluation index to evaluate the prediction effect of the classification prediction machine learning model, and ranking the classification prediction machine learning models from best to worst according to the prediction effect;

[0035] Step S45, extracting the system operation comfort characteristic parameters and the comfort threshold value belonging to the driving behavior information obtained in step S2, and inputting them into the classification prediction machine learning model to obtain the predicted value of the driver comfort test result;

[0036] Step S46, comparing the predicted value of the driver comfort test result obtained in step S45 with the "uncomfortable" result in the driver comfort test result after standardization in step S41, and sorting the classification prediction machine learning models from small to large according to the absolute value of the difference between the two;

[0037] Step S47, extract the classification prediction machine learning model with the smallest sum of the rankings obtained in step S44 and step S46 as the final comfort evaluation model for evaluating the operating comfort of the lane keeping assist system, and evaluate the operating comfort of the lane keeping assist system through the comfort evaluation model.

[0038] Furthermore, the performance evaluation indicators of the classification prediction machine learning model include accuracy, precision, recall, F1 score and confusion matrix.

[0039] The present invention also provides a computer device, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, and the above method is implemented when the computer program instructions are executed by the processor.

[0040] The present invention also provides a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the above method is implemented.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] (1) The present invention utilizes weather environment, road geometry, traffic markings and driving behavior information, combined with machine learning methods to identify key factors that determine the operating comfort of the lane keeping assist system, and quantifies driving behavior under different road environment conditions from an objective perspective, which can avoid the drawbacks of subjective evaluation and make the evaluation results more accurate and effective, thus filling a gap in this field;

[0043] (2) The present invention fully considers the real-world conditions when the lane keeping assist system is operating, and provides a comprehensive evaluation perspective for the operating comfort of the lane keeping assist system, which makes up for the deficiency that the existing evaluation methods have not yet considered the changes in vehicle driving behavior caused by different road environment factors when the system is operating.

[0044] (3) The lane keeping assist system operation comfort evaluation method proposed in the present invention can adapt to different road environments and driving conditions, evaluate the system performance under different conditions, and provide a basis for the optimization and upgrading of the lane keeping assist system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart for implementing a method for evaluating the running comfort of a lane keeping assist system based on machine learning provided by an embodiment of the present invention;

[0046] Figure 2 is a flow chart of determining a comfort threshold value of a characteristic parameter in an embodiment of the present invention;

[0047] Figure 3 is a flow chart of using a machine learning method to predict driving behavior information in an embodiment of the present invention;

[0048] Figure 4 It is a flow chart of constructing a comfort assessment model using a machine learning method in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0050] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0051] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0052] like Figure 1 As shown, this embodiment provides a lane keeping assist system running comfort evaluation method based on machine learning, comprising the following steps:

[0053] Step S1: Conduct field testing using an autonomous driving vehicle equipped with a lane keeping assist system to obtain weather environment information, road geometry and traffic marking information, driving behavior information, and driver comfort test results.

[0054] In step S1, the weather environment information includes: weather type and its corresponding weather attribute information; when the weather type is rainy, the corresponding weather attribute information includes rainfall level and rainfall intensity; when the weather type is snowy, the corresponding weather attribute information includes snowfall level and snowfall intensity; when the weather type is foggy, the corresponding weather attribute information includes fog intensity level and visibility.

[0055] The road geometry and traffic marking information include: road geometry information and matching traffic marking information; wherein the road geometry information includes road grade, design speed, route direction, slope, curvature radius, road section type, lane width, shoulder width, lane position and number of lanes; the traffic marking information includes indication and warning marking types, lane marking combination types, lane marking gaps, lane marking quality, marking maintenance conditions and marking retro-reflectivity; and the road section types include straight lines, flat curves, curved slopes and intersections.

[0056] The driving behavior information includes at least: longitudinal driving speed, lateral driving speed, lateral deviation and rate, heading angle deviation and rate, lateral acceleration and rate of change, steering wheel rotation angle and rate of change, steering wheel torque and driver takeover frequency and duration.

[0057] The driver comfort test results include: comfort and discomfort.

[0058] Among them, weather environment information can be obtained through on-site collection or statistical analysis of public weather information data. Road geometry and traffic marking information can be collected on-site or provided by the road design department. Driving behavior information can be obtained through on-site collection. The driver comfort test results can be obtained by recruiting a certain number of drivers as subjects, and recruiting an observer with rich driving experience (driving experience> 20 years) to sit in the co-pilot seat, and asking and recording the comfort level of the subjects at regular intervals.

[0059] Step S2: using a recursive feature elimination algorithm to filter the information obtained in step S1 as a system operation comfort characteristic parameter, and using a receiver operating characteristic curve analysis method to determine its comfort threshold.

[0060] like Figure 2 As shown, the implementation method of step S2 is as follows:

[0061] Step S21, performing data cleaning on the weather environment information, road geometry and traffic marking information, and driving behavior information obtained in step S1.

[0062] Among them, data cleaning includes removing outliers and filling missing values, which can be achieved based on the Pandas library, NumPy library, matplotlib library, seaborn library, etc. in the Python programming language.

[0063] Step S22: using weather environment information, road geometry and traffic marking information, and driving behavior information as input features and standardizing them, and using driver comfort test results as output features and standardizing them.

[0064] Step S23: Select an extreme gradient enhancement model, use a recursive feature elimination algorithm combined with ten-fold cross validation to screen the input information as the system operation comfort feature parameter, and sort its feature importance.

[0065] Among them, feature screening and importance sorting can be achieved based on the Scikit-learn library and XGBoost library in the Python programming language.

[0066] Step S24: based on the standardized driver comfort test results, draw receiver operating characteristic curves for each standardized system operation comfort characteristic parameter.

[0067] Among them, the drawing of the receiver operating characteristic curve can be achieved by relying on the matplotlib library, seaborn library, etc. in the Python programming language.

[0068] In this embodiment, for each possible comfort classification threshold, the true positive rate and false positive rate of the model are calculated, and the calculation formula is as follows:

[0069] ;

[0070] ;

[0071] Among them, TP represents true positive, FN represents false negative, FP represents false positive, and TN represents true negative.

[0072] The (FPR, TPR) point under each threshold is plotted in the coordinate system, with the horizontal axis representing FPR and the vertical axis representing TPR, to obtain the receiver operating characteristic curve.

[0073] Step S25, analyzing each receiver operating characteristic curve, comparing the Youden index of each system operation comfort characteristic parameter within the value range, and selecting the parameter value corresponding to the maximum Youden index as the comfort threshold of the system operation comfort characteristic parameter.

[0074] Among them, the calculation of the Youden index can be implemented based on the Scikit-learn library in the Python programming language.

[0075] Youden Index is used to evaluate the performance of the classification model at a specific classification threshold. The calculation formula is as follows:

[0076] Youden = TPR − FPR;

[0077] Among them, TPR is the true positive rate, FPR is the false positive rate, and the Youden index ranges from -1 to 1. The larger the value, the better the performance of the model at that threshold. When the Youden index reaches the maximum value, the corresponding threshold is the optimal classification threshold.

[0078] Step S3: using weather environment information, road geometry and traffic marking information as input information, and using machine learning methods to predict driving behavior information.

[0079] like Figure 3 As shown, the implementation method of step S3 is as follows:

[0080] Step S31, using the weather environment information and road geometry and traffic marking information after data cleaning in step S21 as input features and standardizing them, using the driving behavior information as output features and standardizing them, and using the input features and the output features matched therewith as sample pairs.

[0081] Among them, data loading and processing can be implemented based on the Pandas library in the Python programming language, and feature standardization can be implemented based on the StandardScaler in the Scikit-learn library in the Python programming language.

[0082] Step S32: randomly divide all sample pairs obtained in step S31 into training sets and test sets according to a ratio of 8:2, respectively, as training sets and test sets of the machine learning model.

[0083] Among them, data set division can be achieved by relying on the train_test_split in the Scikit-learn library in the Python programming language.

[0084] Step S33: select a variety of machine learning models that can achieve multi-input and multi-output regression prediction tasks from the classic machine learning model library, use ten-fold cross validation and grid search to optimize the model parameters and build a regression prediction machine learning model.

[0085] Among them, the machine learning models that can realize multi-input and multi-output regression prediction tasks include multi-layer feedforward neural networks, Transformer models and convolutional neural networks trained according to the error back propagation algorithm.

[0086] The screening and construction of machine learning models for multi-input and multi-output regression prediction tasks can be achieved by relying on the Scikit-learn library in the Python programming language.

[0087] In this embodiment, a convolutional neural network is selected as the machine learning regressor of the model proposed in the present disclosure. Convolutional neural network is a powerful, flexible and efficient machine learning tool that provides excellent performance and a wide range of functions by simulating the working principle of biological visual systems, and is particularly suitable for solving complex regression problems with spatially structured data.

[0088] In order to extract local features from the input data, the data is convolved in the convolution layer, and the calculation formula is as follows:

[0089] ;

[0090] in, is the element in the output feature map corresponding to the kth convolution kernel, is the element in the input feature map, is the element in the kth convolution kernel, is the bias term corresponding to the kth convolution kernel, and k is the size of the convolution kernel.

[0091] In order to introduce nonlinear factors and enable the convolutional neural network to learn more complex features, an activation function is applied after the convolution layer. The calculation formula is as follows:

[0092] ;

[0093] Here, we take the ReLU activation function as an example. We set all negative values ​​to zero and retain positive values, which helps to increase the nonlinear ability of the model while simplifying the calculation and increasing the training efficiency.

[0094] The data is pooled in the pooling layer, and the calculation formula is as follows:

[0095] ;

[0096] Here we take maximum pooling as an example, and select the maximum value from the input feature map as the element of the output feature map.

[0097] A fully connected layer is added at the end of the convolutional neural network to integrate the local features extracted by the convolutional layer and the pooling layer for the final regression prediction. The calculation formula is as follows:

[0098] ;

[0099] Among them, Input j represents the jth element in the vector obtained by flattening the feature map after the pooling layer, w ij represents the element in the weight matrix of the fully connected layer, b i Represents the bias term of the fully connected layer, Output i Represents the i-th element in the output vector of the fully connected layer.

[0100] In order to train the convolutional neural network, a loss function is defined to measure the difference between the model prediction value and the true value. The calculation formula of the mean square error loss function is:

[0101] ;

[0102] Where N is the number of samples, is the true value of the i-th sample, is the predicted value of the ith sample.

[0103] Select the Adam optimizer to update the parameters of the model to train the convolutional neural network. The calculation formula is as follows:

[0104] ;

[0105] Among them, θ t is the model parameter of the tth iteration, α is the learning rate, m t is the first-order moment estimate, is the second-order moment estimate, is a small constant used to prevent division by zero errors.

[0106] Among them, the screening and construction of machine learning models for multi-input and multi-output regression prediction tasks can be implemented based on the Scikit-learn library in the Python programming language.

[0107] Step S34: Select a regression prediction machine learning model performance evaluation index to evaluate the prediction effect of the regression prediction machine learning model, and rank the regression prediction machine learning models from best to worst according to the prediction effect.

[0108] Among them, the performance evaluation indicators of regression prediction machine learning models include root mean square error, mean absolute error and mean absolute percentage error.

[0109] Among them, the evaluation of the performance of regression prediction machine learning models can be achieved based on the Scikit-learn library in the Python programming language.

[0110] Step S35: extract the system operation comfort characteristic parameters and their comfort thresholds belonging to the weather environment information, road geometry and traffic marking information obtained in step S2, and input them into the regression prediction machine learning model to obtain the driving behavior information prediction value.

[0111] Among them, the loading and use of machine learning models can be achieved based on the Scikit-learn library in the Python programming language.

[0112] Step S36: extract the system operation comfort characteristic parameters and their comfort thresholds belonging to the driving behavior information obtained in step S2, compare them with the corresponding driving behavior information prediction values ​​obtained in step S35, and sort the regression prediction machine learning models from small to large according to the absolute value of the difference between the two.

[0113] Among them, data loading and processing can be implemented by relying on the Pandas library in the Python programming language, and the calculation of the absolute value of the difference can be implemented by relying on the NumPy library in the Python programming language.

[0114] Step S37, extract the regression prediction machine learning model with the smallest sum of the rankings obtained in step S34 and step S36 as the machine learning model for finally predicting driving behavior information, and predict the driving behavior information through the machine learning model.

[0115] Among them, the calculation of the rank sum can be implemented based on the Pandas library in the Python programming language.

[0116] Step S4: Based on the prediction result of the driving behavior information, a comfort evaluation model is constructed by using a machine learning method, and the comfort evaluation model is used to evaluate the operation comfort of the lane keeping assist system.

[0117] like Figure 4 As shown, the implementation method of step S4 is as follows:

[0118] Step S41: Use the driving behavior information predicted by the machine learning model obtained in step S37 as input features and standardize them, use the driver comfort test results as output features and standardize them, and use the input features and the output features that match them as sample pairs.

[0119] Among them, data loading and processing can be implemented based on the Pandas library in the Python programming language, and feature standardization can be implemented based on the StandardScaler in the Scikit-learn library in the Python programming language.

[0120] Step S42: randomly divide all sample pairs obtained in step S41 into training sets and test sets according to a ratio of 8:2, respectively, as training sets and test sets of the machine learning model.

[0121] Among them, data set division can be achieved by relying on the train_test_split in the Scikit-learn library in the Python programming language.

[0122] Step S43: Select a variety of machine learning models that can achieve multi-input-single-output classification prediction tasks from the classic machine learning model library, use ten-fold cross validation and grid search to optimize the model parameters and build a classification prediction machine learning model.

[0123] Among them, machine learning models that can achieve multi-input-single-output classification prediction tasks include logistic regression, random forest, decision tree, support vector machine, extreme gradient boosting model and neural network.

[0124] The screening and construction of machine learning models for multi-input and multi-output classification prediction tasks can be achieved by relying on the Scikit-learn library in the Python programming language.

[0125] In this embodiment, the extreme gradient boosting model is selected as the machine learning classifier of the model proposed in the present disclosure. The extreme gradient boosting model classifier is a powerful, flexible and efficient machine learning tool. It provides excellent performance and a wide range of functions through the gradient boosting framework and is suitable for solving various complex classification problems. Therefore, the extreme gradient boosting model is selected as the machine learning classifier.

[0126] The objective function of the extreme gradient boosting model consists of two parts: the loss function and the regularization term. The calculation formula is as follows:

[0127] ;

[0128] in, is the loss function used to measure the predicted value and the true value The difference between. It is a regularization term used to control the complexity of the model and prevent overfitting.

[0129] For the classification problem in this embodiment, c j Represents comfort, the i-th tree of the extreme gradient boosting model for category c j The prediction can be expressed as:

[0130] ;

[0131] in, is the kth tree for category c j The prediction score of , C is the total number of categories.

[0132] Select the class with the highest predicted probability, calculated as follows:

[0133] ;

[0134] In order to evaluate the performance of the model and avoid overfitting, this embodiment adopts ten-fold cross validation and divides the data set into 10 subsets, 9 of which are used for training and 1 subset is used for validation. This is repeated 10 times, each time using a different subset as the validation set and the remaining subsets as the training set. The average performance index of the model in the 10 validations is calculated to evaluate the stability and generalization ability of the model.

[0135] Use grid search to tune the model's hyperparameters: define the search range and step size of hyperparameters, such as maximum tree depth, learning rate, number of trees, etc., enumerate all possible hyperparameter combinations, use ten-fold cross validation to evaluate the model performance of each combination, and select the hyperparameter combination with the best performance to build the final extreme gradient boosting model.

[0136] Step S44: Select a classification prediction machine learning model performance evaluation index to evaluate the prediction effect of the classification prediction machine learning model, and rank the classification prediction machine learning models from best to worst according to the prediction effect.

[0137] Among them, the performance evaluation indicators of classification prediction machine learning models include accuracy, precision, recall, F1 score and confusion matrix.

[0138] The performance evaluation of machine learning models for classification prediction can be achieved using the Scikit-learn library in the Python programming language.

[0139] Step S45: extract the system operation comfort characteristic parameters and their comfort thresholds belonging to the driving behavior information obtained in step S2, and input them into the classification prediction machine learning model to obtain the predicted value of the driver comfort test result.

[0140] Among them, the loading and use of machine learning models can be achieved based on the Scikit-learn library in the Python programming language.

[0141] Step S46: Compare the predicted value of the driver comfort test result obtained in step S45 with the "uncomfortable" result in the standardized driver comfort test result in step S41, and sort the classification prediction machine learning models from small to large according to the absolute value of the difference between the two.

[0142] Among them, the calculation of the absolute value of the difference can be implemented based on the NumPy library in the Python programming language.

[0143] Step S47, extract the classification prediction machine learning model with the smallest sum of the rankings obtained in step S44 and step S46 as the final comfort evaluation model for evaluating the operating comfort of the lane keeping assist system, and evaluate the operating comfort of the lane keeping assist system through the comfort evaluation model.

[0144] Among them, the calculation of the rank sum can be implemented based on the Pandas library in the Python programming language.

[0145] This embodiment further provides a computer device, including: at least one processor, at least one memory, and computer program instructions stored in the memory, and when the computer program instructions are executed by the processor, the above method is implemented.

[0146] This embodiment further provides a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the above method is implemented.

[0147] In summary, the present invention provides a lane keeping assist system operation comfort evaluation method based on machine learning, which can comprehensively consider the weather environment, road geometry and traffic marking information when the vehicle is driving, and use machine learning methods to quantify driving behavior under different road environment conditions, and then objectively reflect the influence of different road environment factors on the operation comfort of the lane keeping assist system through changes in driving behavior. At the same time, based on the driver's subjective evaluation, subjective and objective methods are used to effectively evaluate the operation comfort of the lane keeping assist system, providing a reliable basis for the optimization and upgrading of the lane keeping assist system; by using weather environment, road geometry and traffic markings and driving behavior information, combined with machine learning methods to identify the key factors that determine the operation comfort of the lane keeping assist system, quantifying driving behavior under different road environment conditions from an objective perspective, it can avoid the disadvantages of subjective evaluation and make the evaluation results more accurate and effective. The design method of the present invention fills the gap in this field; it provides a comprehensive evaluation perspective for the operation comfort of the lane keeping assist system, and makes up for the deficiency that the existing evaluation method has not yet considered the changes in vehicle driving behavior caused by different road environment factors under system operation.

[0148] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0149] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0150] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0151] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0152] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A lane keeping assist system operation comfort evaluation method based on machine learning, characterized in that: The following steps are involved: Step S1, conducting field tests using an autonomous driving vehicle equipped with a lane keeping assist system to obtain weather environment information, road geometry and traffic marking information, driving behavior information, and driver comfort test results; Step S2, using a recursive feature elimination algorithm to filter the information obtained in step S1 as a system operation comfort characteristic parameter, and using a receiver operating characteristic curve analysis method to determine its comfort threshold; Step S3, using weather environment information and road geometry and traffic marking information as input information, using a machine learning method to predict driving behavior information; Step S4: Based on the prediction result of the driving behavior information, a comfort evaluation model is constructed by using a machine learning method, and the comfort evaluation model is used to evaluate the operation comfort of the lane keeping assist system.

2. The lane keeping assist system running comfort evaluation method based on machine learning according to claim 1 is characterized in that: In step S1, the weather environment information includes: weather type and its corresponding weather attribute information; when the weather type is rainy, the corresponding weather attribute information includes rainfall level and rainfall intensity; when the weather type is snowy, the corresponding weather attribute information includes snowfall level and snowfall intensity; when the weather type is foggy, the corresponding weather attribute information includes fog intensity level and visibility; The road geometry and traffic marking information includes: road geometry information and matching traffic marking information; wherein the road geometry information includes road grade, design speed, route direction, slope, curvature radius, road section type, lane width, shoulder width, lane position and number of lanes; the traffic marking information includes indication and warning marking types, lane marking combination types, lane marking gaps, lane marking quality, marking maintenance conditions and marking retro-reflectivity; the road section types include straight lines, flat curves, curved slopes and intersections; The driving behavior information includes at least: longitudinal driving speed, lateral driving speed, lateral deviation and rate, heading angle deviation and rate, lateral acceleration and rate of change, steering wheel rotation angle and rate of change, steering wheel torque, and driver takeover frequency and duration; The driver comfort test results include: comfort and discomfort.

3. The lane keeping assist system running comfort evaluation method based on machine learning according to claim 1, characterized in that: The implementation method of step S2 is as follows: Step S21, performing data cleaning on the weather environment information, road geometry and traffic marking information, and driving behavior information obtained in step S1; Step S22, using weather environment information, road geometry and traffic marking information, and driving behavior information as input features and normalizing them, and using driver comfort test results as output features and normalizing them; Step S23, selecting an extreme gradient enhancement model, using a recursive feature elimination algorithm combined with ten-fold cross validation to screen the input information as a system operation comfort feature parameter, and sorting its feature importance; Step S24: based on the standardized driver comfort test results, draw receiver operating characteristic curves for each standardized system operation comfort characteristic parameter; Step S25, analyzing each receiver operating characteristic curve, comparing the Youden index of each system operation comfort characteristic parameter within the value range, and selecting the parameter value corresponding to the maximum Youden index as the comfort threshold of the system operation comfort characteristic parameter.

4. The lane keeping assist system running comfort evaluation method based on machine learning according to claim 3 is characterized in that: The data cleaning includes removing outliers and filling missing values.

5. The lane keeping assist system running comfort evaluation method based on machine learning according to claim 1, characterized in that: The implementation method of step S3 is as follows: Step S31, using the weather environment information and road geometry and traffic marking information after data cleaning as input features and standardizing them, using the driving behavior information as output features and standardizing them, and using the input features and the output features that match them as sample pairs; Step S32: randomly divide all sample pairs obtained in step S31 according to a set ratio, and use them as a training set and a test set of the machine learning model respectively; Step S33, selecting a plurality of machine learning models capable of realizing multi-input-multi-output regression prediction tasks from a classic machine learning model library, respectively using ten-fold cross validation and grid search to optimize model parameters and construct a regression prediction machine learning model; wherein the machine learning models capable of realizing multi-input-multi-output regression prediction tasks include a multi-layer feedforward neural network, a Transformer model, and a convolutional neural network trained according to an error back propagation algorithm; Step S34: Select a regression prediction machine learning model performance evaluation index to evaluate the prediction effect of the regression prediction machine learning model, and rank the regression prediction machine learning models from best to worst according to the prediction effect; Step S35, extracting the system operation comfort characteristic parameters and their comfort thresholds belonging to the weather environment information and the road geometry and traffic marking information obtained in step S2, and inputting them into the regression prediction machine learning model to obtain the driving behavior information prediction value; Step S36, extracting the system operation comfort characteristic parameters and the comfort threshold value belonging to the driving behavior information obtained in step S2, comparing them with the corresponding driving behavior information prediction value obtained in step S35, and sorting the regression prediction machine learning models from small to large according to the absolute value of the difference between the two; Step S37, extract the regression prediction machine learning model with the smallest sum of the rankings obtained in step S34 and step S36 as the machine learning model for finally predicting driving behavior information, and predict the driving behavior information through the machine learning model.

6. The lane keeping assist system running comfort evaluation method based on machine learning according to claim 5 is characterized in that: The performance evaluation indicators of the regression prediction machine learning model include root mean square error, mean absolute error and mean absolute percentage error.

7. The lane keeping assist system running comfort evaluation method based on machine learning according to claim 1, characterized in that: The implementation method of step S4 is as follows: Step S41, using the driving behavior information predicted in step S3 as input features and normalizing them, using the driver comfort test results as output features and normalizing them, and using the input features and the output features that match them as sample pairs; Step S42: randomly divide all sample pairs obtained in step S41 according to a set ratio, and use them as a training set and a test set of the machine learning model respectively; Step S43, selecting a plurality of machine learning models capable of realizing multi-input-single-output classification prediction tasks from a classic machine learning model library, respectively using ten-fold cross validation and grid search to optimize model parameters and construct a classification prediction machine learning model; wherein the machine learning models capable of realizing multi-input-single-output classification prediction tasks include logistic regression, random forest, decision tree, support vector machine, extreme gradient boosting model and neural network; Step S44: selecting a classification prediction machine learning model performance evaluation index to evaluate the prediction effect of the classification prediction machine learning model, and ranking the classification prediction machine learning models from best to worst according to the prediction effect; Step S45, extracting the system operation comfort characteristic parameters and the comfort threshold value thereof belonging to the driving behavior information obtained in step S2, and inputting them into the classification prediction machine learning model to obtain the predicted value of the driver comfort test result; Step S46, comparing the predicted value of the driver comfort test result obtained in step S45 with the "uncomfortable" result in the driver comfort test result after standardization in step S41, and sorting the classification prediction machine learning models from small to large according to the absolute value of the difference between the two; Step S47, extract the classification prediction machine learning model with the smallest sum of the rankings obtained in step S44 and step S46 as the final comfort evaluation model for evaluating the operating comfort of the lane keeping assist system, and evaluate the operating comfort of the lane keeping assist system through the comfort evaluation model.

8. The lane keeping assist system running comfort evaluation method based on machine learning according to claim 7 is characterized in that: The performance evaluation indicators of the classification prediction machine learning model include accuracy, precision, recall, F1 score and confusion matrix.

9. A computer device, characterized in that: include: At least one processor, at least one memory and computer program instructions stored in the memory, when the computer program instructions are executed by the processor, implement the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images