Self-adaptive adjustment method based on driving behavior prediction

Abstract

The invention discloses a self-adaptive adjustment method based on driving behavior prediction, and relates to the technical field related to intelligent driving, and the method comprises the steps: carrying out the driving behavior data collection of a driver, and building a driving behavior data set; external environment data mapped with the driving behavior data set are obtained, and after data cleaning is conducted on the external environment data and the driving behavior data set, driving features and environment features are extracted; inputting into a deep learning prediction model, and executing deep learning prediction model training; and performing driving behavior prediction of the driver through the trained deep learning prediction model, and establishing feedback early warning according to a driving behavior prediction result. The technical problems that in the prior art, a training strategy cannot be dynamically adjusted according to individual behavior characteristics of a driver and environment data, so that the driving safety is not high, the driving behavior prediction precision is insufficient, and the examination passing rate is low are solved. The technical effects of improving the accuracy of driving behavior prediction, the driving safety and the driving test passing rate are achieved.

Description

Technical Field

[0001] The present application relates to the technical field related to intelligent driving, and in particular to an adaptive adjustment method based on driving behavior prediction. Background Art

[0002] Driving test training is a means of traffic management to ensure that drivers can drive motor vehicles safely and effectively. Traditional driving behavior training mostly relies on face-to-face teaching and simple driving simulation training, combining theoretical learning and practical operation training to enable students to master vehicle operation skills and emergency response capabilities. However, there are many shortcomings. With the complexity of the traffic environment and the increase in individual differences among drivers, traditional methods have been unable to meet the needs of modern driving test training. In addition, in intelligent transportation, autonomous driving, and assisted driving systems, driver behavior analysis and prediction have become a key link in improving driving safety and driving experience. However, the existing driving test training model ignores the personalized analysis and dynamic adjustment of driver behavior, making it impossible to conduct effective learning and targeted training based on the driving behavior characteristics of individual drivers. At the same time, it fails to fully consider the impact of the external environment, making the driving test training efficiency and driving safety insufficient.

[0003] At the current stage, relevant technologies have the problem of being unable to dynamically adjust training strategies based on individual driver behavior characteristics and environmental data, resulting in low driving safety, insufficient driving behavior prediction accuracy and low test pass rate. Summary of the invention

[0004] The present application solves the technical problem in the prior art that the training strategy cannot be dynamically adjusted according to the driver's individual behavior characteristics and environmental data, resulting in low driving safety, insufficient driving behavior prediction accuracy and low test pass rate by providing an adaptive adjustment method based on driving behavior prediction, thereby achieving the technical effect of improving the accuracy of driving behavior prediction, driving safety and driving test pass rate.

[0005] The present application provides an adaptive adjustment method based on driving behavior prediction, comprising: collecting driving behavior data of a driver to establish a driving behavior data set, wherein the driving behavior data set includes steering wheel angle data, brake pedal data, accelerator pedal data, and vehicle status data; acquiring external environment data mapped to the driving behavior data set, and extracting driving characteristics and environmental characteristics after data cleaning of the external environment data and the driving behavior data set; inputting the driving characteristics and the environmental characteristics as input data into a deep learning prediction model, and executing deep learning prediction model training, wherein, before inputting into the deep learning prediction model, performing data quality analysis of the input data through a preprocessing sub-model, and using the data quality analysis result to perform incremental supplementation processing of the input data; predicting the driver's driving behavior through the trained deep learning prediction model, and establishing a feedback warning based on the driving behavior prediction result.

[0006] In a possible implementation, the adaptive adjustment method based on driving behavior prediction also performs the following processing: utilizing the sample quantity analysis layer of the preprocessing sub-model to evaluate the sample quantity of the input data, and establish a sample quantity evaluation result; utilizing the richness analysis layer of the preprocessing sub-model to analyze the sample thickness of the input data, and establish a sample thickness analysis result; and jointly analyzing the sample quantity evaluation result and the sample thickness analysis result, and using the joint analysis result as the data quality analysis result.

[0007] In a possible implementation, the adaptive adjustment method based on driving behavior prediction also performs the following processing: establishing a first data supplement requirement based on the data quality analysis result; establishing a mapping between the driving characteristics and the environmental characteristics, and extracting environment-adaptive driving characteristics based on the driving characteristics and the environmental characteristics, and establishing a second data supplement requirement based on the extraction result; using the second data supplement requirement to perform a supplementary similarity analysis of the local database, using the supplementary similarity analysis result and the first data supplement requirement to perform adaptation analysis, and establish an incremental requirement; performing federal encrypted data matching based on the incremental requirement to establish a matching result; and completing incremental supplement processing based on the supplementary similarity analysis result and the matching result.

[0008] In a possible implementation, the adaptive adjustment method based on driving behavior prediction also performs the following processing: establishing a time correction layer; when the deep learning prediction model is trained by the input data after incremental supplementation processing, the timestamp of the original input data is extracted through the time correction layer, and the timing update concern is constructed using the timestamp; the training constraints of the deep learning prediction model are performed using the timing update concern to complete the training of the deep learning prediction model.

[0009] In a possible implementation, the adaptive adjustment method based on driving behavior prediction also performs the following processing: using the timestamp to perform time trend analysis on the original input data to establish a driver behavior trend curve under the same environment; configuring a sliding window to perform window anomaly recognition on the driver behavior trend curve to establish a window anomaly recognition result; establishing a penalty concern based on the window anomaly recognition result, and constructing the timing update concern based on the penalty concern.

[0010] In a possible implementation, the adaptive adjustment method based on driving behavior prediction also performs the following processing: obtaining the current driving scenario and real-time vehicle status data; constructing a basic danger scenario based on the driving scenario and the real-time vehicle status data; using the basic danger scenario and the driving behavior prediction results to identify driving risks and establish a driving risk level; and establishing the feedback warning based on the driving risk level.

[0011] In a possible implementation, the adaptive adjustment method based on driving behavior prediction also performs the following processing: creating a response window based on the driving risk level; performing behavior feedback monitoring on the driver in the response window, and if the behavior feedback monitoring result fails to meet a preset response threshold, then terminating the execution of a safety response plan in the response window, and performing intelligent vehicle driving through the safety response plan.

[0012] In a possible implementation, the adaptive adjustment method based on driving behavior prediction further performs the following processing: performing warning statistics on feedback warnings to generate a driver optimization report; and performing driver teaching management through the driver optimization report.

[0013] It is proposed to collect the driving behavior data of the driver and establish a driving behavior data set through the adaptive adjustment method based on driving behavior prediction proposed in this application; obtain the external environment data mapped with the driving behavior data set, extract the driving characteristics and environmental characteristics after data cleaning of the external environment data and driving behavior data set; input into the deep learning prediction model, execute the deep learning prediction model training; predict the driver's driving behavior through the trained deep learning prediction model, and establish feedback warning according to the driving behavior prediction results. It solves the technical problem that the training strategy cannot be dynamically adjusted according to the individual behavior characteristics and environmental data of the driver, resulting in low driving safety, insufficient driving behavior prediction accuracy and low test pass rate in the prior art, and achieves the technical effect of improving the accuracy of driving behavior prediction, driving safety and driving test pass rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0015] Figure 1 A schematic diagram of a flow chart of an adaptive adjustment method based on driving behavior prediction provided in an embodiment of the present application;

[0016] Figure 2 A schematic diagram of a process for performing data quality analysis in an adaptive adjustment method based on driving behavior prediction provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0018] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0019] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0020] The present application embodiment provides an adaptive adjustment method based on driving behavior prediction, such as Figure 1As shown, the method includes:

[0021] Step S100, collecting driving behavior data of the driver to establish a driving behavior data set, wherein the driving behavior data set includes steering wheel angle data, brake pedal data, accelerator pedal data, and vehicle status data.

[0022] Preferably, collecting driving behavior data of the driver means collecting and recording data related to the driver's operating behavior in real time during driving through various sensors, monitoring equipment or vehicle-mounted systems to form a driving behavior data set, wherein the driving behavior data set includes steering wheel angle data, brake pedal data, accelerator pedal data, and vehicle status data. Specifically, the steering wheel angle data refers to the rotation angle of the steering wheel and its changes when the driver operates the steering wheel. The steering wheel angle data can reflect the driver's steering behavior during driving, such as turning left, turning right, changing lanes, etc. By recording the changes in the steering wheel angle, the driver's driving habits can be analyzed, such as the rapidity and stability of steering, and whether there is oversteering or insufficient steering and other behavioral characteristics; the brake pedal data refers to the driver's stepping on the brake pedal during driving. The brake pedal data can reflect the driver's braking behavior characteristics, such as the intensity, frequency, duration, etc. of braking. By analyzing the brake pedal data, the driver's reaction speed, whether the braking is smooth, and whether the timing of braking is reasonable (such as whether it is too hasty or too late) can be understood, and then the driver's safe driving level can be evaluated.

[0023] Preferably, the accelerator pedal data records the driver's operation of the accelerator pedal during driving, including the accelerator pedaling depth and changes. The accelerator pedal data can reflect the driver's acceleration behavior, such as the smoothness of acceleration, the speed of acceleration, etc. By analyzing the throttle data, the driver's driving style can be evaluated, such as whether there are behaviors such as sudden acceleration and rapid acceleration, which is very important for judging whether the driver has good driving habits; the vehicle status data usually includes the current speed of the vehicle, engine speed, gear status, fuel consumption, temperature and other information. The vehicle status data provides detailed information about the vehicle's own operating status, which can help understand the driver's performance under different driving conditions. For example, when the vehicle is at different speeds or different driving modes, the driver's behavior may be different. By analyzing these data, the driver's driving mode can be more comprehensively understood. Monitoring and analyzing the driver's operation from multiple dimensions and establishing a driving behavior data set not only helps to identify the driver's operating habits, but also provides rich input features for the subsequent deep learning prediction model. Through in-depth analysis of the driver's driving behavior, personalized driving behavior prediction and adjustment can be achieved, thereby improving the effectiveness of driving training and road safety.

[0024] Step S200, obtaining external environment data mapped to the driving behavior data set, performing data cleaning on the external environment data and the driving behavior data set, and extracting driving features and environmental features.

[0025] Preferably, on the basis of driving behavior data, data related to the driving environment (such as weather, roads, traffic, etc.) are collected to help understand and analyze the driver's performance in different situations. These external environment data can affect the driver's decision and behavior. Specifically, the external environment data may include traffic conditions, road conditions, weather conditions, time and geographical location, etc. Among them, traffic conditions, such as traffic density, congestion, traffic light status (green light, red light), traffic signs, etc., such as a congested traffic environment may cause the driver to brake frequently; road conditions, such as the flatness, slope, curves, slippery road surface, etc. of the road, all of which will affect the driver's operating behavior, such as the need to slow down when making sharp turns; weather conditions, such as rain, snow, haze, wind speed, etc., weather conditions directly affect the driver's operating method, such as on rainy days the driver usually takes less acceleration and braking force; time and geographical location, such as day or night, city or country, etc., the driver's driving habits and risk awareness may be different at different times and places.

[0026] Preferably, data cleaning refers to the cleaning of the acquired driving behavior data and external environment data, which usually includes removing noise data. For example, some sensors may produce errors due to failures or interference, and the cleaning process needs to remove these outliers; processing missing data. Due to incompleteness in the collection process or sensor problems, some data may be missing, and appropriate methods (such as interpolation, mean filling, etc.) are used to fill the missing data; abnormal data detection, for example, sudden extreme values ​​in brake pedal data may be due to sensor failure or abnormal operation, and these anomalies need to be identified and processed; data standardization, data from different sources may have different dimensions and ranges, and the data cleaning process may include standardizing the data to make them have the same dimension.

[0027] Preferably, driving characteristics and environmental characteristics are extracted from the cleaned driving behavior data and external environment data. Specifically, driving characteristics may include driving style characteristics, which extract the driver's driving style by analyzing the operation modes of the accelerator, brake, and steering wheel, such as behavioral characteristics such as sudden acceleration, sudden braking, and frequent lane changes; reaction time characteristics, such as the driver's reaction speed to sudden situations (such as deceleration of the vehicle in front or changes in traffic lights). Drivers with longer reaction times may exhibit lower safety; operating stability characteristics, which analyze the stability of the use of the accelerator and brake to determine whether the driver often accelerates or brakes violently, and then evaluate his driving stability. Environmental characteristics may include traffic density. Under high traffic density, the driver's braking frequency may increase, and the characteristic data may include the value of traffic density or flow; road conditions and weather. For example, when driving on a slippery road, the driver may reduce the acceleration or braking force. These data may include road types (such as highways, urban roads) and weather conditions (such as rainy days, snowy days, etc.); time characteristics. Driving behavior may vary with different time periods, such as the difference in driving behavior between rush hour in the morning and rush hour in the evening, or the difference in driving patterns between daytime and nighttime; the extracted driving characteristics and environmental characteristics can be used to predict driver behavior, optimize driving training and improve driving safety.

[0028] Step S300, input the driving characteristics and the environmental characteristics as input data into a deep learning prediction model, and execute deep learning prediction model training, wherein, before inputting into the deep learning prediction model, a data quality analysis of the input data is performed through a preprocessing sub-model, and the data quality analysis result is used to perform incremental supplementation processing on the input data.

[0029] Preferably, before the driving characteristics and environmental characteristics are input as input data into the deep learning prediction model, a data quality analysis is performed on the input data using a preprocessing sub-model, wherein the preprocessing sub-model is used to ensure the quality and practicality of the data, that is, to check and evaluate the input driving characteristics and environmental characteristics, obtain data quality analysis results, provide detailed feedback on data quality and data volume, provide problematic data features, and guide how to correct these problems. For example, if some features (such as brake pedal data) have a large number of missing values, the data quality analysis can identify this and provide a treatment plan. Specifically, the data quality analysis results are used to perform incremental supplementation processing on the input data, that is, the data is repaired and supplemented according to the data quality analysis results, which may include supplementing missing data and filling missing data using various interpolation methods (such as mean filling, nearest neighbor interpolation, linear interpolation, etc.); filling abnormal data. If some data points are considered to be abnormal, use appropriate techniques (such as data interpolation, using model prediction, etc.) to replace these abnormal values; in some cases, if the amount of data is insufficient, additional training samples can be generated by data enhancement methods, such as expanding the data set by noise injection, data translation, rotation, etc.

[0030] Preferably, the driving characteristics and environmental characteristics processed by the preprocessing sub-model are used as input data and input into a deep learning prediction model (such as a neural network) for further learning and prediction. The input data will be passed to different levels of the model (input layer, hidden layer, output layer), and the final result will be output after calculation and transformation. Specifically, at the input layer, the processed driving characteristics and environmental characteristics are sent to the model as input data, and the neural network model is calculated through multiple hidden layers. The weights and biases are continuously adjusted during the learning and training process so that the output of the model is as close to the expected value (i.e., the prediction target) as possible. The prediction error of the model is calculated by the loss function, and the model parameters are adjusted using an optimization algorithm (such as gradient descent) to reduce the error and improve the prediction accuracy. Using the preprocessed driving characteristics and environmental characteristics and the corresponding target values ​​(such as the predicted results of driving behavior), the deep learning model (neural network) will be trained and its internal parameters will be continuously adjusted so that the driver's behavior can be accurately predicted based on the input data, such as predicting the driver's driving habits (steady driving or aggressive driving), whether there is speeding, whether there is frequent emergency braking, etc., thereby improving the accuracy and reliability of driver behavior analysis.

[0031] Further, such as Figure 2As shown, step S300 also includes step S310, using the sample quantity analysis layer of the preprocessing sub-model to evaluate the sample size of the input data and establish a sample size evaluation result; step S320, using the richness analysis layer of the preprocessing sub-model to perform sample thickness analysis on the input data and establish a sample thickness analysis result; step S330, jointly analyzing the sample size evaluation result and the sample thickness analysis result, and using the joint analysis result as the data quality analysis result.

[0032] Preferably, the sample size analysis layer of the preprocessing sub-model is used to evaluate the sample size of the input data and establish a sample size evaluation result, that is, by analyzing the number of samples in the data set, it is evaluated whether the data is sufficient to support the training of the deep learning model. The sample size directly affects the learning ability and generalization ability of the model. If the number of samples is too small, the model may overfit and it is difficult to learn effective rules; if the number of samples is too large, the training process may become lengthy or the computational cost is too high. Specifically, the specific analysis includes data sufficiency, sampling balance and data extensibility, wherein data sufficiency refers to evaluating whether the sample size in the data set meets the requirements of model training, such as comparing the sample size with the complexity of the task (such as the number of features, the number of categories, the difficulty of the task, etc.) to determine whether the data is sufficient; sampling balance refers to evaluating whether the sample distribution of different categories or features is balanced. If the number of samples in some categories is too small, it may lead to insufficient model learning of the category, thereby affecting the accuracy of the model prediction; data extensibility, if the sample size is insufficient, whether the data set can be expanded by data enhancement, data supplementation, etc. to ensure sufficient samples for training.

[0033] Preferably, the richness analysis layer of the preprocessing sub-model is used to perform sample thickness analysis of the input data, and the sample thickness analysis results are established, that is, the diversity (i.e., the richness of the data) and representativeness of the evaluation data are evaluated. The higher the sample diversity of the data set, the more comprehensive the data is and the more cases are covered. The model can obtain more knowledge and patterns during the learning process, thereby improving the accuracy and robustness of the prediction. The specific analysis includes diversity coverage, uniform distribution of data, and information coverage. Diversity coverage refers to whether the evaluation data covers all possible scenarios or situations. For example, in the driving behavior prediction task, whether the data includes driving behaviors of different weather, different road conditions, and different drivers; uniform distribution of data. If the frequency of occurrence of certain samples or categories is too low, certain patterns or behaviors may not be fully learned when training the model. For example, most of the training data is driving behavior on sunny days, and driving behavior on rainy and snowy days is not involved. The prediction effect of the model on rainy and snowy days may be very poor; information coverage refers to whether the analysis data can provide sufficient feature information and whether it can fully describe all aspects of the problem. For example, in addition to the driver's operating behavior data, whether it also includes external environmental data (weather, road, traffic conditions, etc.).

[0034] Preferably, the sample size evaluation results and the sample thickness analysis results are combined for evaluation, which can comprehensively evaluate the data from two dimensions to determine whether the data is sufficiently substantial and effective, and whether it is suitable for training deep learning models. That is, by considering both the number of samples and the diversity of samples, the quality of the data set is comprehensively evaluated to ensure that the data is not only sufficient in quantity but also rich in diversity, and can fully cover various possible driving behaviors and external environmental conditions; through joint analysis, potential problems of some data sets may be discovered. For example, if the number of samples is sufficient but the diversity of samples is poor, the data set may not cover some important scenarios; conversely, if the data diversity is rich but the number of samples is insufficient, the model may overfit; and then generate data quality analysis results, which combine the advantages and disadvantages of the data set in terms of quantity, distribution, richness, representativeness, etc., to ensure that the deep learning model can be trained on the basis of high-quality data and improve the accuracy and generalization ability of prediction.

[0035] Furthermore, step S300 also includes step S340, establishing a first data supplement requirement based on the data quality analysis result; step S350, establishing a mapping between the driving characteristics and the environmental characteristics, and extracting environment-adaptive driving characteristics based on the driving characteristics and the environmental characteristics, and establishing a second data supplement requirement based on the extraction result; step S360, using the second data supplement requirement to perform a supplementary similarity analysis of the local database, using the supplementary similarity analysis result and the first data supplement requirement to perform adaptation analysis, and establish an incremental requirement; step S370, performing federal encrypted data matching according to the incremental requirement, and establishing a matching result; step S380, completing the incremental supplement processing according to the supplementary similarity analysis result and the matching result.

[0036] Preferably, a first data supplementation requirement is established based on the data quality analysis result, that is, for the insufficient quantity or diversity of the data set, a first data supplementation requirement is established, that is, the amount of data and sample types that need to be increased, which may include supplementing the number of samples by collecting more samples. If the samples of certain scenes or environments are insufficient, the data of specific scenes (such as rainy days, night driving, etc.) are added to enhance the representativeness and generalization ability of the data set. For example, if the data quality analysis results show that there are fewer sample data with lower traffic density, then the first data supplementation requirement may be to increase the driving behavior data with higher traffic density; by analyzing the relationship between the driver's operating behavior (such as braking, accelerator, steering wheel, etc.) and the external environmental data (such as traffic conditions, weather, road conditions, etc.), a mapping relationship between these two types of features is established. For example, frequent braking driving behavior may be related to environmental characteristics with higher traffic density. Then, based on the mapping of driver behavior and environmental data, features that can reflect driver behavior under environmental changes are extracted to help the driver respond in different external environments. For example, when driving in the rain, the driver may take more conservative acceleration and braking behaviors. A second data supplementation requirement is established based on the extraction results, that is, based on the extracted features, it is identified which environmental conditions have insufficient data samples and need to be supplemented.

[0037] Preferably, the second data supplementation requirement is used to perform analysis on the existing local data set to identify samples similar to the second data supplementation requirement. For example, if the second data supplementation requirement requires the addition of data on driving in rainy days, the supplementary similarity analysis will search the local database for driving behavior samples similar to driving in rainy days (such as driving behavior on wet roads or in low visibility conditions) to help determine which parts of the existing data can be matched with the newly added supplementary data, reduce redundancy and unnecessary repeated data, and thus improve the efficiency and accuracy of data supplementation; then an adaptation analysis is performed. Specifically, the supplementary similarity analysis results are combined with the first data supplementation requirement to identify which features or samples are most needed in the data supplementation process and which can be replaced by existing data. Through the adaptation analysis, it can be ensured that data supplementation can meet the sample size and diversity requirements while being able to supplement the most useful data in a targeted manner, and then establish incremental requirements, that is, further data supplementation requirements obtained through the adaptation analysis, which help guide the actual data supplementation process.

[0038] Preferably, federated encrypted data matching is performed according to incremental needs, wherein federated learning is a distributed machine learning method in which data is retained on local devices for processing rather than being centrally stored on a server. When data supplementation is required, federated encrypted data matching is used to match data between different devices or databases while ensuring the privacy and security of the data, thereby establishing matching results, including data samples that meet the incremental needs, and finally completing the incremental supplementation processing based on the supplementary similarity analysis results and the matching results. Specifically, based on the incremental needs, the supplementary similarity analysis results and the matching results, data supplementation is actually performed, which may include collecting new data samples from external data sources, or extracting samples that meet the needs from existing data and adding them to the local database, that is, continuously increasing, optimizing and enriching existing data sets according to the required features, scenarios or environments to improve the training quality of the deep learning model and ultimately improve the prediction performance of the model.

[0039] Furthermore, step S300 also includes step S390, establishing a time correction layer; step S3100, when the deep learning prediction model is trained by the input data after incremental supplementation processing, the timestamp of the original input data is extracted through the time correction layer, and the timing update concern is constructed using the timestamp; step S3200, the training constraints of the deep learning prediction model are performed using the timing update concern to complete the training of the deep learning prediction model.

[0040] Preferably, a time correction layer is introduced in the deep learning model training, and the time series update attention is performed through the timestamp information to enhance the model's learning ability of time dependencies and ensure that the model can better process time series data. Specifically, the time correction layer is a layer for processing time-related input data. The main purpose is to help the model understand the importance of time information for prediction, especially in time series data. Driving behavior or environmental data are usually time series data, that is, each data point is associated with a specific timestamp. The task of the time correction layer is to extract these timestamps and combine them with the input data to ensure that the model can understand and utilize time information. For example, driving behavior data (such as braking, throttle, speed, etc.) may change over time, and some behaviors may occur more in specific time periods (for example, rush hour or nighttime). The time correction layer can help the model process these changes and incorporate them into the training process.

[0041] Preferably, each piece of input data is usually accompanied by a timestamp, indicating at which moment or time period the data was collected, and the timestamp is used to construct a time series update attention, that is, dynamically adjust the model's attention to data in different time periods or time points according to the time information. Specifically, over time, the driver's response may become better and better. For example, if a driver often brakes suddenly at the beginning of training, but as the training progresses, he gradually learns to brake more smoothly, then the attention to this scene should be gradually reduced, because the driver has learned how to deal with this situation; conversely, over time, the driver may have new problems. If a driver performs well in the early stage, but gradually develops problems such as fatigue driving over time, the model's attention should be increased accordingly to respond to this change; by associating the timestamp with the training data, the model can dynamically adjust the weight or importance of the data according to the time position of each data, and the time correction layer updates the attention to these time series data, so that the model can continuously learn and adapt to changes in driver behavior and environment.

[0042] Preferably, the model uses the time series update attention as a training constraint during training, that is, the degree of influence of different data points on the training is determined based on the update attention of the time series data. Specifically, if some scenarios become more important in the later stages of training (for example, the driver begins to get used to making unsafe reactions in certain environments), the model will be constrained to pay more attention to these data with larger changes, ensuring that the model can learn new driving behavior patterns in a timely manner; if the driver's behavior in certain scenarios has become stable and performs well (for example, the driver has gradually learned to drive smoothly in a specific environment), then the model's attention to these scenarios will be reduced, thereby avoiding overfitting and meaningless calculations, and focusing on optimizing the prediction ability for new scenarios; and then by dynamically adjusting the weight of each data sample, the model can more flexibly adapt to the trend of data changes over time during the training process, and finally complete the deep learning prediction model training, and pay more attention to new scenarios or driving behavior changes, thereby optimizing the model's prediction ability and adaptability, and improving the model's prediction accuracy.

[0043] Furthermore, step S3100 also includes step S3110, using the timestamp to perform time trend analysis on the original input data to establish a driver behavior trend curve under the same environment; step S3120, configuring a sliding window to perform window anomaly recognition on the driver behavior trend curve to establish a window anomaly recognition result; step S3130, establishing a penalty concern based on the window anomaly recognition result, and constructing the timing update concern based on the penalty concern.

[0044] Preferably, timestamps are used to dynamically analyze driver behavior data, and the model's focus on driver behavior is adjusted through time series analysis. Specifically, based on timestamps, the driver's behavior data (such as acceleration, braking, steering wheel turning, etc.) is analyzed for changing trends over time. Time trend analysis can reveal the evolution of driver behavior, such as whether the driver has gradually improved his driving skills, whether there is a deterioration in certain behaviors (such as fatigue driving, sudden braking, etc.). Through time trend analysis, the data can be divided into time series to form a driver behavior trend curve under the same environment, showing how the driver's behavior changes over time under similar external environmental conditions (such as similar weather, road and traffic conditions). For example, if the braking behavior is analyzed, the trend curve may show the driver's braking force at different time points, or the speed of the braking reaction.

[0045] Preferably, a sliding window is configured to perform window anomaly recognition on the driver behavior trend curve, that is, by defining a window of fixed size, sliding gradually in the time series data, analyzing the data in each window, performing anomaly detection on the driver's behavior, identifying abnormal fluctuations in the driver's behavior within a specific time period, and then establishing a window anomaly recognition result. For example, if the driver's behavior is found to be unstable (such as frequent emergency braking) within a certain time period, these behaviors are identified as abnormal; then, penalty attention is established based on the window anomaly recognition result. Specifically, when certain behaviors are marked as abnormal in the window anomaly recognition, additional penalties are imposed on these abnormal behaviors to reduce the number of these abnormalities. According to the weights in model training, the deep learning prediction model pays less attention to these abnormal behavior data. By restricting abnormal data, it helps the model reduce over-learning of overly unstable and extreme data, thereby improving the stability and generalization ability of the model. Then, based on the penalty attention, the temporal update attention is constructed. That is, according to the results of the penalty attention, the model will adjust the attention to the data in each time period. For example, if the behavior in certain time periods is marked as abnormal, the model will dynamically adjust the attention and reduce the weight of abnormal data in training, so that it can more effectively learn long-term stable driving behavior and ignore short-term fluctuations or accidental events, thereby providing more accurate and flexible predictions.

[0046] Step S400: predict the driver's driving behavior through the trained deep learning prediction model, and establish a feedback warning based on the driving behavior prediction result.

[0047] Preferably, after training, the deep learning prediction model can predict the driver's driving behavior based on the input driving characteristics and environmental characteristics (for example, the driver's throttle, brake, steering wheel operation, road conditions, weather, etc.). Specifically, driving behavior prediction may include driving style prediction, that is, predicting whether the driver tends to drive smoothly or tends to have aggressive driving styles such as sudden acceleration, sudden braking, frequent lane changes, etc.; driver reaction prediction, that is, predicting the driver's possible reactions in specific situations (for example, encountering obstacles ahead, sudden braking, or when the traffic light turns red), such as whether to brake in time, whether to change lanes quickly, etc.; dangerous behavior prediction, that is, identifying whether the driver is likely to engage in dangerous behaviors, such as fatigue driving, drunk driving, speeding, sudden braking, etc.; future driving behavior prediction, that is, based on the current driver's behavior and the surrounding environment, predicting the driver's possible behavior in the next period of time.

[0048] Preferably, the deep learning prediction model can predict whether sudden braking will occur, for example, if the pressure of the current brake pedal suddenly increases and the road surface is slippery, the model can predict that the driver may brake suddenly when an obstacle appears ahead; or whether there is dangerous driving, for example, if the driver frequently accelerates to close to the maximum speed of the vehicle and the surrounding environment is relatively complex (such as high traffic density, narrow roads, etc.), the model may predict that the driver is at risk of high-speed driving; the output result of the model prediction is used as the driving behavior prediction result, and a feedback warning is established based on the driving behavior prediction result, that is, based on the driving behavior results predicted by the deep learning model, a warning or suggestion is issued to the driver or relevant personnel (such as the vehicle-mounted system, traffic management system, etc.). For example, when the system detects that the driver is about to brake suddenly, the vehicle-mounted system may issue a sound prompt "Please pay attention to the braking distance" to remind the driver to maintain an appropriate distance to prevent potential dangers or bad behaviors, thereby improving driving safety, preventing traffic accidents, and avoiding bad driving habits.

[0049] Preferably, step S400 also includes step S410, acquiring the current driving scenario and real-time vehicle status data; step S420, constructing a basic danger scenario based on the driving scenario and the real-time vehicle status data; step S430, using the basic danger scenario and the driving behavior prediction result to identify driving risks and establish a driving risk level; step S440, establishing the feedback warning based on the driving risk level.

[0050] Preferably, the current driving scenario and real-time vehicle status data are obtained, including the specific environment in which the current driver is located, including external factors such as traffic density, road type, weather conditions, traffic light status, obstacles ahead, as well as the vehicle's speed, acceleration, braking status, steering wheel angle, engine speed, tire pressure, etc., to comprehensively evaluate the current driving environment and the driver's behavior, identify potential dangerous situations that may exist, and construct basic dangerous scenarios, that is, based on the external environment (such as roads, traffic, weather) and the internal status of the vehicle (such as speed, acceleration), determine which scenarios may cause the driver's behavior to be unsafe. For example, if a vehicle is driving at high speed in rainy or foggy weather and the road is slippery, the braking distance may increase. The basic dangerous scenario is high-speed driving on a slippery road.

[0051] Preferably, after obtaining the basic danger scenario, the driving risk is further identified in combination with the driving behavior prediction result. For example, by predicting the driver's behavior, it is analyzed whether he is likely to make dangerous reactions (such as sudden braking, sudden acceleration, frequent lane changes, etc.). The driver's risk level is evaluated according to the basic danger scenario and the predicted driving behavior, which may include low risk (the driver is in a relatively safe environment, the behavior is stable and the vehicle is in normal condition), medium risk (the driver's behavior begins to be unstable or there are small hidden dangers) and high risk (there are obvious dangerous behaviors, such as sudden acceleration, frequent lane changes, failure to brake in time, etc., or the driving environment itself has greater risks, such as rainy and snowy weather, high-speed driving, high traffic density, etc.). Finally, a feedback warning is established according to the driving risk level, that is, based on the identified driving risk level, a corresponding warning or reminder is issued to the driver or the vehicle system to help the driver realize the potential dangers and take preventive measures, thereby improving driving safety and the driving test pass rate and reducing the occurrence of safety accidents.

[0052] Furthermore, step S430 also includes step S431, creating a response window based on the driving risk level; step S432, performing behavior feedback monitoring on the driver in the response window, and if the behavior feedback monitoring result fails to meet a preset response threshold, then terminating execution of a safety response plan in the response window, and performing intelligent vehicle driving through the safety response plan.

[0053] Preferably, a response window is created based on the driving risk level and behavioral feedback monitoring is performed, and then the vehicle's intelligent driving control is executed through a safety response plan when the driver fails to respond in time, ensuring that the driver's behavior meets safety requirements and taking automated intervention when necessary to ensure driving safety. Specifically, the response window refers to monitoring whether the driver can make an appropriate response according to the current driving risk level within a certain time range. The length of the response window is usually related to the degree of dangerousness of the driving behavior. For example, the response window corresponding to low risk may be longer, giving the driver more time to adjust the driving behavior. Conversely, the response window corresponding to high risk will be shorter because high-risk driving behavior requires a quicker response.

[0054] Preferably, the driver's behavior is monitored in real time within the response window to monitor whether the driver slows down or brakes to deal with obstacles ahead or impending dangers, and adjusts the driving style, such as avoiding sudden acceleration or sudden braking, maintaining smooth driving, etc. If the behavior feedback monitoring result fails to meet the preset response threshold, that is, the necessary safety measures are not taken in time or fully (for example, failure to decelerate in time, failure to brake appropriately, etc.), it is judged that the driver has failed to reach the required safety response level, wherein the preset response threshold refers to the preset minimum safety response that the driver should take within the response window, and then the safety response plan is executed at the end of the response window, that is, intelligent driving control operations are automatically taken to reduce potential risks and ensure driving safety, such as automatic braking, lane keeping, deceleration control, etc., and vehicle control is performed based on these intelligent driving to ensure that the vehicle is always in a safe driving state, reduce accidents, and improve driving safety.

[0055] Furthermore, step S440 also includes step S441, performing warning statistics on feedback warnings and generating a driver optimization report; and step S442, performing driver teaching management through the driver optimization report.

[0056] Preferably, warning statistics refer to summarizing and counting all the warning information issued. By analyzing these warning data, the driving habits of the driver over a period of time can be understood, and which behavior patterns or situations lead to frequent warnings can be identified. Then, a driver optimization report is generated based on the identification results. The report summarizes the driver's driving behavior performance over a period of time and points out the risky behaviors and potential problems. It may include driver behavior evaluation, safe driving rating, optimization suggestions, etc. The driver's driving habits and risky behaviors are clearly displayed through data analysis, and the driver is provided with improvement directions and specific suggestions; the driver optimization report is used for driver teaching management, specifically including that based on the analysis results in the optimization report, the coach or manager can tailor a personalized training plan for the driver; the high-frequency risk behaviors mentioned in the report will become the focus of teaching management; based on the driver's optimization report, the coach can take behavioral correction measures, such as reminding the driver to maintain an appropriate distance between vehicles, avoid excessive acceleration in certain environments, etc. The driver can not only get instant feedback, but also improve his driving skills in continuous teaching management and gradually improve his safe driving level.

[0057] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. An adaptive adjustment method based on driving behavior prediction, characterized in that: The method comprises: Collecting driving behavior data of the driver to establish a driving behavior data set, wherein the driving behavior data set includes steering wheel angle data, brake pedal data, accelerator pedal data, and vehicle status data; Acquire external environment data mapped to the driving behavior data set, and extract driving characteristics and environmental characteristics after performing data cleaning on the external environment data and the driving behavior data set; The driving characteristics and the environmental characteristics are used as input data and input into a deep learning prediction model to perform deep learning prediction model training, wherein before inputting into the deep learning prediction model, a data quality analysis of the input data is performed by a preprocessing sub-model, and an incremental supplement processing of the input data is performed using the data quality analysis result; The driver's driving behavior is predicted by the trained deep learning prediction model, and a feedback warning is established based on the driving behavior prediction results.

2. The adaptive adjustment method based on driving behavior prediction according to claim 1, characterized in that: The data quality analysis of the input data by the preprocessing sub-model includes: Using the sample size analysis layer of the preprocessing sub-model to evaluate the sample size of input data, and establish a sample size evaluation result; Performing sample thickness analysis of input data using the richness analysis layer of the preprocessing sub-model to establish sample thickness analysis results; The sample size evaluation result and the sample thickness analysis result are jointly analyzed, and the joint analysis result is used as the data quality analysis result.

3. The adaptive adjustment method based on driving behavior prediction according to claim 2, characterized in that: The incremental supplementation process of input data using the data quality analysis result includes: Establishing a first data supplementation requirement according to the data quality analysis result; Establishing a mapping between the driving characteristics and the environmental characteristics, and extracting environment-adaptive driving characteristics based on the driving characteristics and the environmental characteristics, and establishing a second data supplementation requirement according to the extraction result; Using the second data supplementation requirement to perform supplementary similarity analysis on the local database, using the supplementary similarity analysis result and the first data supplementation requirement to perform adaptation analysis, and establishing incremental requirements; Performing federated encrypted data matching according to the incremental requirements and establishing matching results; The incremental supplementation process is completed according to the supplementary similarity analysis result and the matching result.

4. The adaptive adjustment method based on driving behavior prediction according to claim 1, characterized in that: The performing of deep learning prediction model training includes: Establishing a time correction layer; When the deep learning prediction model is trained by the input data processed by incremental supplementation, the timestamp of the original input data is extracted through the time correction layer, and the time series update attention is constructed using the timestamp; The timing update focus is used to perform training constraints on the deep learning prediction model to complete the training of the deep learning prediction model.

5. The adaptive adjustment method based on driving behavior prediction according to claim 4, characterized in that: The using the timestamp to construct a timing update concern includes: Using the timestamp to perform time trend analysis on the original input data, and establishing a driver behavior trend curve under similar environments; configuring a sliding window to perform window anomaly recognition on the driver behavior trend curve, and establishing a window anomaly recognition result; A penalty concern is established according to the window anomaly recognition result, and the timing update concern is constructed according to the penalty concern.

6. The adaptive adjustment method based on driving behavior prediction according to claim 1, characterized in that: The step of establishing feedback warning according to the driving behavior prediction result includes: Obtain current driving scenario and real-time vehicle status data; Constructing a basic danger scenario according to the driving scenario and the real-time vehicle status data; Using the basic danger scenario and the driving behavior prediction result to identify driving risks and establish a driving risk level; The feedback warning is established according to the driving risk level.

7. The adaptive adjustment method based on driving behavior prediction according to claim 6, characterized in that: The method further comprises: creating a response window based on the driving risk level; The driver's behavior feedback is monitored in the response window. If the behavior feedback monitoring result fails to meet a preset response threshold, the safety response plan is executed at the end of the response window, and the vehicle is intelligently driven through the safety response plan.

8. The adaptive adjustment method based on driving behavior prediction according to claim 1, characterized in that: The method further comprises: Collect early warning statistics on feedback warnings and generate driver optimization reports; The driver's teaching management is carried out through the driver optimization report.

Images