Cockpit risk intervention strategy generation method and system based on driver physiological state recognition

Abstract

The application provides a cockpit risk intervention strategy generation method and system based on driver physiological state recognition, which first acquires continuous physiological state data of a driver when driving, real-time environmental data in the cockpit and driving operation behavior data, constructs a physiological environment behavior linkage feature matrix through multi-dimensional linkage analysis; calls a pre-trained cockpit risk evolution prediction artificial intelligence model to predict the risk evolution trend of the matrix, generates a cockpit risk evolution trajectory of the driver in the current and future multiple time periods; generates an initial cockpit risk intervention strategy based on the trajectory and a preset intervention strategy library; collects real-time state data in the cockpit and dynamically compares the trajectory, adapts the initial strategy to the time period and adjusts the measures, obtains an optimized cockpit risk intervention strategy and pushes the strategy to a cockpit control unit to record execution process data. The application can accurately predict cockpit risks and dynamically adjust intervention strategies, effectively improving driving safety.

Description

Technical Field

[0001] This invention relates to the field of driving safety technology, and more specifically, to a method and system for generating cockpit risk intervention strategies based on driver physiological state recognition. Background Technology

[0002] In automotive driving scenarios, ensuring the safety of drivers and passengers is paramount. Traditional driving safety measures primarily focus on monitoring the vehicle's mechanical performance and simple driving behavior, such as using vehicle sensors to monitor engine and braking system status, and using cameras to detect driver fatigue or violations. However, these methods have significant limitations. Firstly, focusing solely on vehicle mechanical performance and simple driving behaviors fails to fully consider the impact of complex physiological changes on driving safety. For example, slowed reactions or operational errors caused by emotional fluctuations or physical discomfort may go undetected. Secondly, real-time environmental factors within the cabin, such as temperature, humidity, and air quality, indirectly affect the driver's physiological state and driving actions, but traditional methods fail to effectively correlate these environmental factors with the driver's physiological state and driving behavior. Therefore, current technologies struggle to accurately predict potential risks within the cabin and develop effective intervention strategies, failing to meet the ever-increasing demands for driving safety. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for generating cockpit risk intervention strategies based on driver physiological state recognition, the method comprising:

[0004] Acquire continuous physiological data of the driver during driving, real-time environmental data in the cabin, and driving operation behavior data;

[0005] The continuous physiological state data, the real-time environmental data in the cockpit, and the driving operation behavior data are subjected to multi-dimensional linkage analysis and processing to construct a physiological-environment-behavior linkage feature matrix.

[0006] The pre-trained cockpit risk evolution prediction artificial intelligence model is invoked to perform risk evolution trend prediction processing on the physiological environment behavior linkage feature matrix, generating the cockpit risk evolution trajectory of the driver at the present and multiple different time periods after the present.

[0007] Based on the cabin risk evolution trajectory and a pre-set intervention strategy library, an initial cabin risk intervention strategy is generated for the risk diffusion path.

[0008] Real-time status data inside the cockpit is collected, and the real-time status data is dynamically compared with the cockpit risk evolution trajectory. The initial cockpit risk intervention strategy is adapted to different time periods and the measures are adjusted to obtain an optimized cockpit risk intervention strategy. The optimized cockpit risk intervention strategy is pushed to the cockpit control unit and the data of the intervention strategy execution process is recorded.

[0009] In another aspect, embodiments of the present invention also provide a cockpit risk intervention strategy generation system based on driver physiological state recognition, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0010] Based on the above, this invention acquires continuous physiological state data of the driver during driving, real-time environmental data within the cabin, and driving operation behavior data. It then performs multi-dimensional linkage analysis to construct a physiological-environment-behavior linkage feature matrix, accurately depicting the complex relationships between various factors during driving. A pre-trained cabin risk evolution prediction artificial intelligence model is invoked to predict the risk evolution trend of the linkage feature matrix, generating cabin risk evolution trajectories for the driver at current and future time periods, providing a forward-looking basis for risk intervention. An initial cabin risk intervention strategy is generated based on the risk evolution trajectory and a pre-set intervention strategy library. By dynamically comparing real-time cabin state data with the risk evolution trajectory, the initial strategy is adapted to different time periods and adjusted accordingly, resulting in an optimized cabin risk intervention strategy. This optimized strategy is then pushed to the cabin control unit to record execution process data. This allows for dynamic adjustment of the intervention strategy based on real-time conditions, effectively improving the accuracy and timeliness of cabin risk intervention and significantly enhancing driving safety. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the execution flow of the cockpit risk intervention strategy generation method based on driver physiological state recognition provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of exemplary hardware and software components of a cockpit risk intervention strategy generation system based on driver physiological state recognition provided in an embodiment of the present invention. Detailed Implementation

[0013] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for generating cockpit risk intervention strategies based on driver physiological state recognition, according to an embodiment of the present invention. The following is a detailed description of this method for generating cockpit risk intervention strategies based on driver physiological state recognition.

[0014] Step S110: Acquire continuous physiological state data, real-time environmental data in the cabin, and driving operation behavior data of the driver during the driving process.

[0015] In this embodiment, continuous physiological state data is acquired through wearable physiological monitoring devices installed on the driver. These devices can collect real-time physiological indicators such as heart rate, respiratory rate, skin conductance, and electroencephalogram (EEG) signals. To protect the driver's sensitive privacy data, data encryption transmission technology is used during data acquisition. The raw physiological data is encrypted in real time using a symmetric encryption algorithm that conforms to the standards of the State Cryptography Administration, ensuring that the data is not leaked during transmission. Real-time environmental data within the cockpit is collected by environmental sensors distributed in different locations within the cockpit, including temperature sensors, humidity sensors, air quality sensors, light sensors, and noise sensors. These sensors can monitor environmental parameters in different areas of the cockpit in real time. Driving operation behavior data is acquired through the vehicle's CAN bus interface and motion capture devices installed in the driver's area. The CAN bus interface is used to collect operation data such as accelerator pedal position, brake pedal travel, steering wheel angle, and gear information. The motion capture devices are used to identify the driver's hand and foot movements.

[0016] Step S120: Perform multi-dimensional linkage analysis and processing on the continuous physiological state data, the real-time environmental data in the cockpit, and the driving operation behavior data to construct a physiological environment behavior linkage feature matrix.

[0017] In this embodiment, this step involves in-depth analysis and feature extraction of the three types of data collected in order to construct a linkage feature matrix that reflects the relationship between the three.

[0018] Step S121: Perform time window division processing on the continuous physiological state data, divide it into multiple physiological data windows according to fixed time intervals, calculate the rate of change of each physiological indicator within the physiological data window, and obtain the time-series change characteristics of physiological indicators.

[0019] In this embodiment, the fixed time interval of the time window is set to 5 minutes, meaning that data from every 5 minutes constitutes a physiological data window. For example, starting from the moment the driver begins driving, the first physiological data window contains physiological data from minutes 0 to 5, the second physiological data window contains physiological data from minutes 5 to 10, and so on. For each physiological data window, the rate of change of each physiological indicator within that window is calculated. Specifically, for a given physiological indicator, the indicator value at the end of the window is subtracted from the indicator value at the beginning of the window, and then divided by the window duration (i.e., 5 minutes) to obtain the average rate of change of that physiological indicator within that window. The average rates of change of all physiological indicators are arranged in a preset indicator order to obtain the time-series change characteristics of the physiological indicators. For example, if the physiological indicators include heart rate, respiratory rate, and skin conductance, then the time-series change characteristics of the physiological indicators are a three-dimensional vector, with each dimension corresponding to the average rate of change of one physiological indicator.

[0020] Step S122: Perform spatial region division processing on the real-time environmental data inside the cabin, divide the cabin interior space into multiple environmental monitoring areas, collect environmental parameter data within the environmental monitoring areas, calculate the environmental parameter difference values ​​between different environmental monitoring areas, and obtain the spatial distribution characteristics of environmental parameters.

[0021] In this embodiment, the interior space of the cockpit is divided according to the actual structure of the cockpit. For example, the cockpit is divided into six environmental monitoring zones: the driver's head area, the driver's chest area, the driver's leg area, the front passenger area, the left rear seat area, and the right rear seat area. Each environmental monitoring zone is equipped with a corresponding environmental sensor to collect environmental parameter data for that zone. For each environmental parameter (such as temperature, humidity, etc.), the parameter difference value between different environmental monitoring zones is calculated. Specifically, for a given environmental parameter, the parameter values ​​of any two environmental monitoring zones are subtracted, and the absolute value is taken as the difference value between the two zones. The difference values ​​of all environmental parameters across all environmental monitoring zones are arranged according to a preset zone pair order and parameter order to obtain the spatial distribution characteristics of the environmental parameters. For example, for the temperature parameter, the temperature difference value between the driver's head area and the driver's chest area, the temperature difference value between the driver's head area and the driver's leg area, etc., are calculated, and then these difference values ​​are arranged in a certain order.

[0022] Step S123: Perform action sequence extraction processing on the driving operation behavior data, extract the driving operation action sequence during continuous driving, identify the operation type and operation duration corresponding to the driving operation action sequence, and obtain driving operation behavior features.

[0023] In this embodiment, the extraction of driving operation action sequences is achieved based on time series segmentation and recognition. First, the driving operation behavior data is arranged in chronological order, and then segmented into individual operation actions according to the start and end times of the operation actions. Operation type identification is achieved through matching with a preset operation action template library. This library stores feature templates for various typical driving operations (such as acceleration, deceleration, steering, gear shifting, turning on lights, honking the horn, etc.). The extracted operation actions are matched with templates in the library, and the operation type corresponding to the template with the highest matching degree is the operation type of that operation action. The operation duration is obtained by calculating the time difference between the end time and the start time of the operation action. Arranging the identified operation types and their corresponding operation durations in chronological order yields the driving operation behavior features. For example, the driving operation behavior features can be represented as a sequence, where each element consists of an operation type and an operation duration, such as (acceleration, 5 seconds), (steering, 3 seconds), (deceleration, 2 seconds), etc.

[0024] Step S124: Calculate the co-variation coefficient of the difference in environmental parameters between the physiological data window and the environmental monitoring area within the corresponding time interval, and at the same time calculate the correlation matching degree between the physiological data window and the driving operation behavior characteristics within the corresponding time interval.

[0025] In this embodiment, the time interval corresponding to the difference in environmental parameters between the physiological data window and the environmental monitoring area refers to the time range corresponding to the physiological data window, that is, each physiological data window corresponds to the difference in environmental parameters within the same time range.

[0026] Step S1241: Extract the rate of change of each physiological indicator in the physiological data window, calculate the mean of the rate of change of each physiological indicator, and obtain the average rate of change of the physiological indicators.

[0027] In this embodiment, a physiological data window includes the rate of change of multiple physiological indicators, such as the rate of change of heart rate and respiratory rate. The mean rate of change of each physiological indicator is calculated by summing the rates of change of each physiological indicator and dividing by the number of physiological indicators.

[0028] Step S1242: Extract the environmental parameter difference values ​​of different environmental monitoring areas within the corresponding time interval, calculate the mean value of the environmental parameter difference values ​​of different environmental monitoring areas, and obtain the average environmental parameter difference value.

[0029] In this embodiment, the environmental parameter difference value between different environmental monitoring areas within the corresponding time interval refers to the environmental parameter difference value calculated between different environmental monitoring areas within the same time range as the physiological data window. The mean of these difference values ​​is calculated by summing all the environmental parameter difference values ​​and dividing by the number of difference values ​​to obtain the average environmental parameter difference value.

[0030] Step S1243: Calculate the ratio of the average rate of change of the physiological indicators to the average difference of the environmental parameters, and correct the ratio by combining the collection frequency of the two types of data to obtain the synergistic change coefficient.

[0031] In this embodiment, the unit of the average rate of change of physiological indicators is (indicator units / minute), and the unit of the average difference value of environmental parameters is (environmental parameter units). The unit of their ratio is (indicator units / (minute·environmental parameter units)). Since the collection frequencies of physiological and environmental data may differ, this ratio needs to be corrected. The correction method is as follows: divide the collection frequency of physiological data by the collection frequency of environmental data to obtain a frequency correction coefficient. Then, multiply the ratio of the average rate of change of physiological indicators to the average difference value of environmental parameters by the frequency correction coefficient to obtain the co-change coefficient. The co-change coefficient reflects the co-change relationship between the average rate of change of physiological indicators and the average difference value of environmental parameters; the larger the coefficient, the higher the degree of co-change between the two.

[0032] Step S1244: Encode the operation type and operation duration in the driving operation behavior characteristics to obtain a driving operation encoding sequence.

[0033] In this embodiment, the operation type encoding adopts a one-hot encoding method, mapping each operation type to a binary vector. The length of the vector is equal to the total number of operation types, with only one element being 1 and the rest being 0. The position of the 1 corresponds to the operation type. The operation duration encoding uses normalization processing, mapping the operation duration to a value between 0 and 1. The normalization method is to divide the operation duration by a preset maximum operation duration to obtain the normalized operation duration. The operation type encoding vector and the normalized operation duration of each operation action are combined to form the code for that operation action. Arranging all the operation action codes in chronological order yields the driving operation code sequence.

[0034] Step S1245: Encode the rate of change of physiological indicators within the physiological data window to obtain the physiological indicator coding sequence.

[0035] In this embodiment, the encoding of physiological indicator change rates also employs normalization processing, mapping each physiological indicator change rate to a value between 0 and 1. The normalization method is as follows: divide each physiological indicator change rate by a preset maximum change rate for that indicator to obtain the normalized physiological indicator change rate. The normalized physiological indicator change rates are then arranged in a preset indicator order to form a vector. This vector represents the encoding of the physiological indicator change rates within the physiological data window. Arranging the encodings of all physiological data windows in chronological order yields the physiological indicator encoding sequence.

[0036] Step S1246: Calculate the sequence similarity between the driving operation coding sequence and the physiological indicator coding sequence, and use the sequence similarity as the correlation matching degree between the physiological data window and the driving operation behavior features.

[0037] In this embodiment, the Dynamic Time Warping (DTW) algorithm is used to calculate sequence similarity. This algorithm can non-linearly align two sequences on the time axis and calculate their similarity. The specific calculation process is as follows: First, a distance matrix is ​​constructed, where the rows correspond to elements of the driving operation encoding sequence, and the columns correspond to elements of the physiological indicator encoding sequence. Each element in the matrix represents the distance (e.g., Euclidean distance) between two corresponding sequence elements. Then, a path is found from the top left corner to the bottom right corner of the distance matrix that minimizes the sum of distances along the path. The reciprocal of this minimum distance sum is the sequence similarity. The association matching degree reflects the degree of association between the physiological data window and the driving operation behavior features within the corresponding time interval. A higher matching degree indicates a stronger association between the two.

[0038] Step S125: Integrate the temporal variation characteristics of the physiological indicators, the spatial distribution characteristics of the environmental parameters, the characteristics of the driving operation behavior, and the corresponding synergistic variation coefficients and correlation matching degrees in a preset dimension order to construct a physiological-environment-behavior linkage feature matrix.

[0039] In this embodiment, the preset dimensional order is pre-defined based on the importance and correlation of the features. For example, physiological indicator temporal variation features are arranged first, followed by environmental parameter spatial distribution features, then driving operation behavior features, and finally the synergistic variation coefficient and correlation matching degree. During the integration process, each feature is treated as a row vector or column vector of a matrix, arranged according to the preset dimensional order to form a two-dimensional matrix. For example, if the physiological indicator temporal variation features are an m-dimensional vector, the environmental parameter spatial distribution features are an n-dimensional vector, the driving operation behavior features are a p-dimensional vector, and the synergistic variation coefficient and correlation matching degree are each a scalar, then the dimension of the physiological-environment-behavior linkage feature matrix is ​​(m+n+p+2)×k, where k is the number of physiological data windows.

[0040] Step S130: Call the pre-trained cockpit risk evolution prediction artificial intelligence model to perform risk evolution trend prediction processing on the physiological environment behavior linkage feature matrix, and generate the cockpit risk evolution trajectory of the driver at the present and multiple different time periods after the present.

[0041] In this embodiment, the cabin risk evolution prediction artificial intelligence model is a pre-trained deep learning model that can predict the cabin risk evolution at different future time periods based on the input physiological environment behavior linkage feature matrix.

[0042] Step S131: Input the physiological environment behavior linkage feature matrix into the feature preprocessing module of the cabin risk evolution prediction artificial intelligence model, and perform feature scale unification processing on features of different dimensions in the matrix to obtain a scale-unified linkage feature matrix.

[0043] In this embodiment, the main function of the feature preprocessing module is to standardize the input linked feature matrix to eliminate the dimensional differences between features of different dimensions. The standardization method is as follows: for each feature dimension in the matrix, calculate the mean and standard deviation of all feature values ​​in that dimension; then subtract the mean of that dimension from each feature value and divide by the standard deviation of that dimension to obtain the standardized feature value. Through this process, the mean of features of different dimensions within the matrix is ​​0, and the standard deviation is 1, thereby achieving a unified feature scale.

[0044] Step S132: Transmit the scale-unified linkage feature matrix to the temporal evolution module of the cabin risk evolution prediction artificial intelligence model, perform sequential correlation analysis on the features in the matrix in the time dimension, extract the transmission rules of features between different time windows, and obtain the feature temporal evolution rules.

[0045] In this embodiment, the temporal evolution module employs a recurrent neural network (RNN) structure, specifically a long short-term memory (LSTM) network. A scale-uniform linked feature matrix is ​​input into the LSTM network in chronological order. The LSTM network, through its internal memory units, captures the long-term dependencies of the sequence data over time. During the forward propagation of the LSTM network, the input feature matrix first passes through an input gate, a forget gate, and an output gate. The input gate controls how new feature information enters the memory units, the forget gate controls how old information in the memory units is forgotten, and the output gate controls how the information in the memory units is output to the next time step. Through iterative calculations across multiple time steps, the LSTM network learns the feature transfer patterns between different time windows, i.e., the feature temporal evolution patterns. These patterns can be represented as a series of weight parameters, reflecting the degree of influence of the features of the previous time window on the features of the next time window.

[0046] Step S133: Input the feature temporal evolution law into the spatial correlation module of the cabin risk evolution prediction artificial intelligence model, and combine it with the spatial region identifiers marked in the matrix to analyze the mutual influence law between environmental parameter features, physiological indicator features, and driving operation behavior features in different spatial regions, and obtain the feature spatial correlation law.

[0047] In this embodiment, the spatial association module adopts a graph neural network (GNN) structure. The nodes in the graph represent different spatial regions within the cockpit, and the features of each node include environmental parameters, physiological indicators, and driving behavior characteristics related to that region. Spatial region identifiers in the matrix are used to determine the connectivity between graph nodes; that is, there are edge connections between adjacent spatial region nodes. By learning from the graph structure data, the GNN network can analyze the mutual influence relationships between the features of different nodes (spatial regions). During the computation process of the GNN network, each node updates its feature representation based on the features of its neighboring nodes. Through multiple iterations, the network learns the mutual influence patterns between features in different spatial regions, i.e., the feature spatial association patterns. These feature spatial association patterns can also be represented as a series of weight parameters, which reflect the degree of influence between features in different spatial regions.

[0048] Step S134: Construct a risk evolution prediction model based on the temporal evolution law of the features and the spatial correlation law of the features. Input the linkage feature matrix data of the current time period, predict the changes of the linkage feature matrix in multiple different time periods after the current time, and obtain the predicted linkage feature matrix sequence.

[0049] In this embodiment, the risk evolution prediction model is constructed by fusing the characteristic temporal evolution rules learned by the temporal evolution module and the characteristic spatial correlation rules learned by the spatial correlation module.

[0050] Step S1341: Input the time transmission coefficient in the feature time-series evolution law and the spatial influence coefficient in the feature spatial correlation law into the model construction module as the core parameters of the risk evolution prediction model.

[0051] In this embodiment, the temporal transfer coefficient is a weight parameter obtained by the temporal evolution module (LSTM network) during the learning of the temporal evolution law of features. It reflects the strength of feature transfer from the previous time window to the next time window. The spatial influence coefficient is a weight parameter obtained by the spatial association module (GNN network) during the learning of the spatial association law of features. It reflects the strength of mutual influence between features in different spatial regions. The model building module integrates the above-mentioned temporal transfer coefficient and spatial influence coefficient into the network structure of the risk evolution prediction model as the core parameters of the model.

[0052] Step S1342: Construct a time-dimensional prediction submodule in the risk evolution prediction model, and establish the transmission relationship between the characteristics of the current time period and the characteristics of the next time period based on the time transmission coefficient.

[0053] In this embodiment, the structure of the time dimension prediction submodule is similar to that of the LSTM network in the time series evolution module, but the parameters have been initialized using time transfer coefficients. After the linkage feature matrix data of the current time period is input into the time dimension prediction submodule, the submodule calculates the predicted value of each element in the feature matrix of the next time period based on the time transfer coefficients, thereby establishing the transfer relationship between the features of the current time period and the features of the next time period.

[0054] Step S1343: Construct a spatial dimension prediction submodule in the risk evolution prediction model, and establish the mutual influence relationship between the characteristics of different spatial regions based on the spatial influence coefficient.

[0055] In this embodiment, the structure of the spatial dimension prediction submodule is similar to that of the GNN network in the spatial association module, and the parameters are initialized using spatial influence coefficients. The spatial dimension prediction submodule receives the feature matrix data for the next time period output by the temporal dimension prediction submodule, combines it with spatial region identifiers, calculates the mutual influence between features of different spatial regions based on spatial influence coefficients, corrects the feature matrix data, and further optimizes the prediction results of the feature matrix for the next time period.

[0056] Step S1344: Input the linkage feature matrix data of the current time period into the time dimension prediction submodule and the spatial dimension prediction submodule, and start the two submodules to perform collaborative prediction to obtain the predicted linkage feature matrix of the next time period.

[0057] In this embodiment, the linkage feature matrix data for the current time period is simultaneously input into both the time-dimension prediction submodule and the spatial-dimension prediction submodule. The time-dimension prediction submodule predicts the feature matrix in the time dimension based on the time transfer coefficient, obtaining a preliminary prediction result for the feature matrix of the next time period; the spatial-dimension prediction submodule then corrects the preliminary prediction result in the spatial dimension based on the spatial influence coefficient. Then, the time-dimension prediction result and the spatial-dimension correction result are fused using a weighted average method, where the weights of the time-dimension prediction result and the spatial-dimension correction result are determined based on empirical parameters obtained during model training. The fused result is the predicted linkage feature matrix for the next time period.

[0058] Step S1345: Repeat the process of inputting the linkage feature matrix data of the current time period into the time dimension prediction submodule and the spatial dimension prediction submodule, and simultaneously start the two submodules to perform collaborative prediction to obtain the prediction linkage feature matrix of the next time period. This process is repeated to obtain the prediction linkage feature matrices of multiple different time periods after the current time period, and then arranged in chronological order to form a prediction linkage feature matrix sequence.

[0059] In this embodiment, for example, if it is necessary to predict the cabin risk evolution within one hour after the current time, the hour is divided into 6 time periods, each lasting 10 minutes. First, the linkage feature matrix data of the current time period is input into two sub-modules to obtain the predicted linkage feature matrix for the first prediction time period (0-10 minutes after the current time). Then, the predicted linkage feature matrix of the first prediction time period is used as the linkage feature matrix data for the current time period and input into the two sub-modules again to obtain the predicted linkage feature matrix for the second prediction time period (10-20 minutes after the current time). This process continues until the predicted linkage feature matrix for the sixth prediction time period is obtained. Arranging these 6 predicted linkage feature matrices in chronological order forms the predicted linkage feature matrix sequence.

[0060] Step S135: Perform risk type matching and diffusion path analysis on the predicted linkage feature matrix sequence to identify the risk type and diffusion direction of the risk in the spatial area corresponding to different prediction periods, and generate the cabin risk evolution trajectory.

[0061] In this embodiment, risk type matching is achieved by matching each predicted linkage feature matrix in the predicted linkage feature matrix sequence with a preset risk type feature template library. The risk type feature template library stores feature templates for various typical cabin risk types (such as driver fatigue risk, driver emotional agitation risk, cabin environment deterioration risk, etc.), with each feature template corresponding to a risk type. The similarity between the predicted linkage feature matrix and each template in the template library is calculated, and the risk type corresponding to the template with the highest similarity is the risk type for that predicted time period. The diffusion path analysis is based on the spatial correlation rules of features. According to the mutual influence rules between features of different spatial regions, the possibility and direction of risk diffusion from one spatial region to other spatial regions are analyzed. For example, if the physiological indicators of the driver's head region are abnormal, and the spatial correlation weight between this region and the driving operation area is high, then the risk may diffuse from the head region to the driving operation area. Arranging the risk types and corresponding risk diffusion directions of different predicted time periods in chronological order generates the cabin risk evolution trajectory. The cabin risk evolution trajectory is a sequence of data containing information such as time, risk type, and diffusion direction.

[0062] Step S140: Based on the cockpit risk evolution trajectory and the preset intervention strategy library, generate an initial cockpit risk intervention strategy for the risk diffusion path.

[0063] In this embodiment, the preset intervention strategy library is a database that stores various cabin risk intervention measures. These intervention measures are summarized based on historical experience and expert knowledge, and each intervention measure corresponds to a specific risk type and risk diffusion path.

[0064] Step S141: Analyze the cabin risk evolution trajectory, extract the risk type, risk diffusion direction and spatial area involved in different time periods, and determine the risk elements and intervention coverage for different time periods.

[0065] In this embodiment, the cockpit risk evolution trajectory includes information such as the risk type, risk diffusion direction, and spatial area involved in each predicted time period. The analysis process involves extracting and analyzing the above information to determine the risk elements that need intervention in each time period (such as abnormal physiological indicators, abnormal environmental parameters, or abnormal driving behavior that lead to the risk) and the spatial area that the intervention measures need to cover. For example, if the risk type in a predicted time period is driver fatigue risk, the risk diffusion direction is from the driver's head area to the driving operation area, and the spatial area involved in the risk is the driver's head area and the driving operation area, then the risk element that needs intervention is the abnormal EEG signal of the driver (reflecting fatigue), and the intervention coverage area is the driver's head area and the driving operation area.

[0066] Step S142: Invoke the preset intervention strategy library, and retrieve the matching basic intervention measures in the intervention strategy library according to the risk factors and the intervention coverage, to form a set of basic intervention measures.

[0067] In this embodiment, the retrieval of the intervention strategy library employs a combination of keyword retrieval and feature matching. First, search keywords are generated based on the extracted risk factors and intervention coverage, such as "fatigue risk," "abnormal EEG signals," "head region," and "driving operation area." Then, these keywords are used to conduct a preliminary search of the intervention strategy library, yielding a batch of candidate intervention measures. Next, detailed feature matching is performed on the characteristics of the candidate intervention measures against the risk factors and intervention coverage, calculating the matching degree. Candidate intervention measures with a matching degree higher than a preset threshold are selected into the basic intervention measure set.

[0068] Step S143: Collect execution data from the historical application process of different measures within the basic intervention measure set, and extract the measure activation response time and measure effect data from the execution data.

[0069] In this embodiment, the execution data of the historical application process of the measures is stored in the historical database of the intervention strategy library. Each execution data record contains detailed information about the application of a certain intervention measure, including the measure name, activation time, effective time, and effect evaluation. When collecting execution data, the corresponding execution data record is retrieved from the historical database based on the measure name in the basic intervention measure set. The measure activation response time is extracted from the execution data record, which is the length of time from the issuance of the activation command to the actual effectiveness of the measure (effective time minus activation time). The measure effect data includes evaluation data such as the degree of improvement of risk indicators and the recovery status of drivers after the application of the measure.

[0070] Step S144: Based on the time intervals of different periods in the cabin risk evolution trajectory, and in conjunction with the activation response time of the basic intervention measures, determine the activation lead time corresponding to the basic intervention measures, and control the activation time of the measures through the activation lead time.

[0071] In this embodiment, the lead time refers to the amount of time required to initiate intervention measures in advance so that the intervention measures can take effect in a timely manner during the period when the risk occurs.

[0072] Step S1441: Extract the start and end times of different time periods in the cabin risk evolution trajectory, calculate the time interval between two adjacent time periods, and obtain the time interval sequence.

[0073] In this embodiment, each time period in the cabin risk evolution trajectory has a clearly defined start and end time. For example, the first time period starts at the current time + 0 minutes and ends at the current time + 10 minutes; the second time period starts at the current time + 10 minutes and ends at the current time + 20 minutes. The time interval between two adjacent time periods is the start time of the latter time period minus the start time of the former time period. In the example above, the time interval between adjacent time periods is 10 minutes. Arranging all the time intervals of adjacent time periods in chronological order yields the time interval sequence.

[0074] Step S1442: Extract the time from the issuance of the activation command to the actual effect of the measure from the historical execution data of the basic intervention measures, and use it as the activation response time.

[0075] In this embodiment, the method for extracting the activation response time is the same as in step S143, and the activation response time of the basic intervention measures is obtained from the historical execution data records.

[0076] Step S1443: For the risk period corresponding to the basic intervention measures, determine the start time of the risk period corresponding to the basic intervention measures, and take the start time of the risk period corresponding to the basic intervention measures as the target time for the measures to take effect.

[0077] In this embodiment, the risk period corresponding to the basic intervention measure refers to the period during which the risk to be intervened occurs. The start time of this period can be determined from the cabin risk evolution trajectory. For example, if the basic intervention measure is to intervene in the risk of the second predicted period, then the start time of this risk period is the current time + 10 minutes, which is taken as the target time for the measure to take effect.

[0078] Step S1444: Calculate the difference between the target time and the current time, and subtract the activation response time to obtain the initial value of the activation lead time for the basic intervention measures.

[0079] In this embodiment, the difference between the target time and the current time is the time length from the current time to the target time when the measure takes effect. Subtracting the activation response time gives the amount of time required to activate the intervention measure in advance, i.e., the initial activation lead time. For example, if the target time is the current time + 10 minutes and the activation response time is 2 minutes, then the initial activation lead time is 10 minutes - 2 minutes = 8 minutes. That is, the intervention measure needs to be activated at the current time + 8 minutes to ensure it takes effect at the target time.

[0080] Step S1445: Based on the time interval sequence, if the time interval between adjacent time periods is less than the activation response time, the initial value of the activation lead time for the basic intervention measures is adjusted, and the activation lead time for the basic intervention measures is finally determined.

[0081] In this embodiment, if the time interval between adjacent time periods is less than the activation response time, it indicates that the time interval between the occurrence of risks in the two adjacent time periods is short. The intervention measures in the previous time period may not have fully taken effect before the risk in the next time period has already occurred. In this case, the initial value of the activation lead time for the basic intervention measures needs to be adjusted to avoid conflicts between intervention measures. The adjustment method is as follows: increase the initial value of the activation lead time by an adjustment amount. The size of the adjustment amount is determined based on the difference between the time interval between adjacent time periods and the activation response time; the smaller the difference, the larger the adjustment amount. For example, if the time interval between adjacent time periods is 1 minute, the activation response time is 2 minutes, and the difference is -1 minute, then the adjustment amount can be set to 1 minute. The initial activation lead time is 8 minutes, and the adjusted activation lead time is 8 minutes + 1 minute = 9 minutes.

[0082] Step S145: Integrate the lead time and coverage of basic intervention measures corresponding to different time periods in chronological order to construct an initial cabin risk intervention strategy.

[0083] In this embodiment, each prediction period corresponds to one or more basic intervention measures, and each basic intervention measure has its activation lead time and intervention coverage area. The above information is integrated in chronological order according to the prediction period to form a strategy document containing information such as time, intervention measures, activation lead time, and intervention coverage area, i.e., the initial cabin risk intervention strategy. For example, the initial cabin risk intervention strategy can be represented as: Period 1 (0-10 minutes after the current time), Risk type: Fatigue risk, Intervention measure: Play energizing music, Activation lead time: 5 minutes, Intervention coverage area: Driver's head area; Period 2 (10-20 minutes after the current time), Risk type: Environmental deterioration risk, Intervention measure: Turn on the air conditioning for ventilation, Activation lead time: 3 minutes, Intervention coverage area: Entire cabin area, etc.

[0084] Step S150: Collect real-time status data in the cockpit, dynamically compare the real-time status data with the cockpit risk evolution trajectory, adapt the initial cockpit risk intervention strategy to the time period and adjust the measures to obtain an optimized cockpit risk intervention strategy, push the optimized cockpit risk intervention strategy to the cockpit control unit and record the data of the intervention strategy execution process.

[0085] In this embodiment, this step is a dynamic adjustment process. By collecting real-time state data inside the cockpit and comparing it with the predicted cockpit risk evolution trajectory, deviations between the actual situation and the predicted situation can be detected in a timely manner, thereby adjusting and optimizing the initial intervention strategy to improve the effectiveness and accuracy of the intervention strategy.

[0086] Step S151: Collect real-time status data in the cockpit, dynamically compare the real-time status data with the cockpit risk evolution trajectory, and perform time-period adaptation and measure adjustment processing on the initial cockpit risk intervention strategy to obtain an optimized cockpit risk intervention strategy.

[0087] In this embodiment, the method for collecting real-time status data in the cockpit is similar to the method for collecting real-time environmental data and driving operation behavior data in step S110, which is collected in real time through environmental sensors and motion capture devices.

[0088] Step S1511: Collect real-time physiological state data, real-time environmental data, and real-time driving operation behavior data through the monitoring equipment in the cockpit to form a real-time status data set in the cockpit.

[0089] In this embodiment, real-time physiological state data is still collected through wearable physiological monitoring devices, real-time environmental data is collected through environmental sensors, and real-time driving operation behavior data is collected through a CAN bus interface and motion capture devices. Unlike step S110, the data collection frequency is higher here to meet the real-time requirements of dynamic comparison. The three types of real-time data are combined to form a real-time in-cabin state data set.

[0090] Step S1512: Perform feature extraction processing on the real-time status data set in the cockpit to obtain real-time physiological features, real-time environmental features and real-time driving behavior features, and construct a real-time status feature matrix.

[0091] In this embodiment, the method for extracting real-time physiological features is similar to the method for extracting the temporal change features of physiological indicators in step S121. Time windows are divided according to fixed time intervals (e.g., 1 minute), and the rate of change of each physiological indicator within the window is calculated. The method for extracting real-time environmental features is similar to the method for extracting the spatial distribution features of environmental parameters in step S122, calculating the difference values ​​of environmental parameters between different environmental monitoring areas. The method for extracting real-time driving behavior features is similar to the method for extracting driving operation behavior features in step S123, extracting the sequence of driving operation actions and identifying the operation type and duration. Then, following the methods in steps S124 and S125, the co-change coefficients of real-time physiological features and real-time environmental features, and the correlation matching degree between real-time physiological features and real-time driving behavior features are calculated. These features are then integrated to construct a real-time state feature matrix.

[0092] Step S1513: Perform feature difference analysis on the real-time state feature matrix and the prediction linkage feature matrix of the corresponding time period in the cabin risk evolution trajectory, calculate the difference values ​​of different dimensions of features between the two types of matrices, and obtain the feature difference set.

[0093] In this embodiment, the corresponding time period refers to the predicted time period in which the current time corresponds to the real-time state feature matrix. For example, if the current time is the midpoint of predicted time period 2 (10-20 minutes after the current time), then the corresponding time period is predicted time period 2. The real-time state feature matrix is ​​compared with the predicted linkage feature matrix of predicted time period 2 in the cabin risk evolution trajectory. Feature difference analysis compares the feature values ​​of corresponding dimensions in the two types of matrices and calculates the difference between them. The difference is calculated by subtracting the feature value of the predicted linkage feature matrix from the feature value of the real-time state feature matrix and taking the absolute value. The difference values ​​of all dimensions are combined to form a feature difference set.

[0094] Step S1514: Determine whether the intervention measures for the corresponding time period in the initial cabin risk intervention strategy need to be adjusted based on the feature difference set. If the difference value exceeds the preset range, select an alternative intervention measure from the intervention strategy library that matches the real-time status feature matrix.

[0095] In this embodiment, the intervention measures corresponding to the time period refer to the intervention measures formulated for that time period in the initial cabin risk intervention strategy.

[0096] Step S15141: Compare different difference values ​​within the feature difference set with the preset difference threshold, and count the number of difference values ​​exceeding the difference threshold and the corresponding feature dimensions.

[0097] In this embodiment, the preset difference threshold is determined based on historical data and expert experience. The difference threshold may differ for different feature dimensions. For features related to key risks, the difference threshold is set lower to improve sensitivity to key risks. Each difference value is compared with its corresponding difference threshold. If the difference value is greater than the difference threshold, it is considered to exceed the preset range, and the number of difference values ​​and their corresponding feature dimensions are recorded.

[0098] Step S15142: If the proportion of the number of difference values ​​exceeding the difference threshold to the total number of difference values ​​reaches a preset proportion, or if the feature dimension corresponding to the difference value exceeding the difference threshold is a key risk-related dimension, then it is determined that the intervention measures for the corresponding period need to be adjusted.

[0099] In this embodiment, the preset ratio is typically set to 50%. That is, if more than half of the discrepancies exceed the difference threshold, the actual situation is considered to deviate significantly from the predicted situation, requiring adjustment of intervention measures. Key risk-related dimensions refer to feature dimensions that significantly influence the risk type and risk diffusion path, such as the driver's heart rate variability, EEG signal characteristics, and cabin oxygen concentration. If the feature dimensions corresponding to discrepancies exceeding the difference threshold include key risk-related dimensions, then regardless of their proportion, the intervention measures are determined to need adjustment.

[0100] Step S15143: When it is determined that the intervention measures for the corresponding time period need to be adjusted, the real-time state feature matrix is ​​input into the retrieval module of the intervention strategy library to retrieve the intervention measures with the highest matching degree to the real-time state feature matrix.

[0101] In this embodiment, the retrieval module of the intervention strategy library adopts a retrieval method similar to that in step S142, combining keyword retrieval and feature matching to retrieve the intervention measures with the highest matching degree to the real-time state feature matrix from the intervention strategy library.

[0102] Step S15144: Extract historical implementation effect data of the retrieved intervention measures and calculate the effect score of the retrieved intervention measures in similar scenarios.

[0103] In this embodiment, similar scenarios refer to scenarios that share similar risk types, risk diffusion paths, and risk-involved spatial areas with the scenarios reflected in the current real-time state feature matrix. Effectiveness evaluation data of the intervention measures in these similar scenarios, such as risk indicator improvement rates and driver satisfaction, are extracted from historical performance data. Then, the evaluation data is weighted and calculated according to a preset scoring formula to obtain an effectiveness score. In the scoring formula, the weights of different evaluation indicators are determined based on their importance to the intervention effect.

[0104] Step S15145: Select the intervention with the highest effect score as the alternative intervention, and extract the execution parameter requirements of the selected alternative intervention.

[0105] In this embodiment, the execution parameters include the initiation conditions, intensity, duration, and area of ​​effect of the intervention. These parameters are crucial to ensuring the correct execution of the intervention. For example, if the alternative intervention is "adjusting the seat temperature," the execution parameters may include the target temperature, heating / cooling rate, and area of ​​effect on the seat.

[0106] Step S1515: Redetermine the activation lead time and intervention coverage of the adjusted intervention measures, and re-integrate them in chronological order with the unadjusted intervention measures to obtain an optimized cabin risk intervention strategy.

[0107] In this embodiment, the method for redetermining the activation lead time for the adjusted intervention measures (i.e., alternative intervention measures) is the same as in step S144. The activation lead time is calculated and adjusted based on the current real-time status and the time interval of the corresponding period in the cabin risk evolution trajectory, combined with the activation response time of the alternative intervention measures. The intervention coverage is determined based on the execution parameter requirements of the alternative intervention measures and the spatial area involved in the risk. The adjusted intervention measures are then rearranged and integrated with the unadjusted intervention measures in chronological order to form an optimized cabin risk intervention strategy.

[0108] Step S152: Push the optimized cockpit risk intervention strategy to the cockpit control unit and record the data of the intervention strategy execution process.

[0109] In this embodiment, the cockpit control unit is the control center of the vehicle cockpit system, which can receive and execute various intervention commands.

[0110] For example, step S1521: Perform instruction conversion processing on different intervention measures within the optimized cockpit risk intervention strategy, converting the execution requirements of different intervention measures within the optimized cockpit risk intervention strategy into control instructions that can be recognized by the cockpit control unit.

[0111] In this embodiment, different intervention measures correspond to different cockpit control devices. For example, the intervention measure of adjusting the air conditioning temperature corresponds to the air conditioning control unit, and the intervention measure of playing music corresponds to the audio control unit. The instruction conversion processing is to convert the execution requirements of the intervention measure (such as target temperature, music track, volume, etc.) into an instruction format that the corresponding control unit can recognize. The instruction format follows the communication protocol standard of each control unit.

[0112] Step S1522: The converted control commands are sent to the corresponding cockpit control unit through the communication module in the cockpit. After different control commands are sent, the cockpit control unit is waited for a reception confirmation signal.

[0113] In this embodiment, the communication module in the cockpit adopts CAN bus communication, which features high communication speed and high reliability. Control commands are sent in priority order, with important intervention commands (such as emergency braking related commands) having higher priority. After sending each control command, the communication module waits to receive a reception confirmation signal from the corresponding cockpit control unit. If no confirmation signal is received within a preset time, the control command is resent, up to a maximum of 3 times. If no confirmation signal is received again, a command transmission failure message is recorded.

[0114] Step S1523: After receiving the confirmation signal for all control commands, collect the actual start time, actual execution duration, and cabin status change data of different control commands in chronological order.

[0115] In this embodiment, the actual start-up time refers to the time when the cockpit control unit begins executing control commands, which is obtained through the feedback signal of the control unit. The actual execution duration refers to the length of time from the start to the end of the execution of the control command. After execution, the state change data in the cockpit is collected by the monitoring equipment in the cockpit, including changes in physiological state data, environmental data, and driving operation behavior data, which are used to evaluate the actual effectiveness of the intervention measures.

[0116] Step S1524: Compare the collected execution process data with the preset parameters of the optimized cabin risk intervention strategy, and record the execution parameter deviation data of different intervention measures within the optimized cabin risk intervention strategy.

[0117] In this embodiment, the preset parameters include the preset start time, preset execution duration, and preset state change target of the intervention. The actual start time is compared with the preset start time to calculate the time deviation; the actual execution duration is compared with the preset execution duration to calculate the duration deviation; and the state change data after execution is compared with the preset state change target to calculate the state deviation. The above deviation data reflects the difference between the actual execution of the intervention and the expected outcome.

[0118] Step S1525: Integrate the content of the optimized cockpit risk intervention strategy, control command information, collected execution process data and recorded execution parameter deviation data into a preset format to form an intervention strategy execution process record document, store it in the cockpit data storage unit and mark the generation time of the intervention strategy execution process record document and the risk evolution trajectory identifier corresponding to the optimized cockpit risk intervention strategy.

[0119] In this embodiment, the preset format includes the document title, content structure, data fields, etc., ensuring the standardization and readability of the recorded documents. The cockpit data storage unit uses a large-capacity solid-state drive, which features fast data read / write speed and high reliability. Labeling the generation time facilitates subsequent time-series analysis of the intervention strategy execution, and labeling the risk evolution trajectory facilitates correlation analysis between the execution process data and the corresponding risk evolution trajectory.

[0120] Step S210: The method further includes a training step for an artificial intelligence model for predicting cockpit risk evolution.

[0121] In this embodiment, this step is performed before the model is applied, and the model is trained with a large amount of historical data to improve the model's prediction accuracy.

[0122] Step S211: Collect historical driving data, including historical physiological state data of the driver, historical environmental data in the cockpit, historical driving operation behavior data, and corresponding historical cockpit risk event data.

[0123] In this embodiment, historical driving data is collected from multiple vehicles and drivers to ensure data diversity and representativeness. Historical cockpit risk event data includes information such as the time of occurrence of the risk event, the type of risk, the risk spread path, and the consequences. During the collection process, historical physiological state data that involves privacy is also encrypted using the same encryption method as in step S110.

[0124] Step S212: Preprocess historical driving data, including data cleaning, data completion, and data standardization.

[0125] In this embodiment, data cleaning removes noise and outliers from the data. Statistical outlier detection methods, such as the Z-score method, are used to identify and remove data that deviates from the mean by more than three times the standard deviation. Data completion fills in missing data. For continuous data, linear interpolation is used; for discrete data, mode imputation is used. Data standardization converts feature data with different dimensions into data with the same scale. The standardization method described in step S131 is used to ensure that the mean is 0 and the standard deviation is 1.

[0126] Step S213: Based on the preprocessed historical driving data, construct a historical physiological environment behavior linkage feature matrix according to the method in step S120, and label the risk type and risk diffusion path label corresponding to each historical physiological environment behavior linkage feature matrix according to the historical cockpit risk event data.

[0127] In this embodiment, the process of constructing the historical physiological-environmental-behavioral linkage feature matrix is ​​exactly the same as step S120, including steps such as time window division, spatial region division, action sequence extraction, collaborative change coefficient calculation, correlation matching degree calculation, and feature matrix integration. The labeling of risk type and risk diffusion path is based on historical cabin risk event data, and each historical physiological-environmental-behavioral linkage feature matrix corresponds to one or more risk type labels and risk diffusion path labels.

[0128] Step S214: Divide the labeled historical physiological environment behavior linkage feature matrix into training set, validation set and test set, with a division ratio of 7:2:1.

[0129] In this embodiment, the training set is used for model parameter learning, the validation set is used for model hyperparameter tuning, and the test set is used to evaluate the model's generalization performance. The partitioning process uses random sampling to ensure consistent data distribution across all sets.

[0130] Step S215: Initialize the parameters of the cabin risk evolution prediction artificial intelligence model, including the network parameters of the temporal evolution module and the spatial correlation module, and set hyperparameters such as the number of training rounds, learning rate, and batch size.

[0131] In this embodiment, the model parameters are initialized using the Xavier initialization method to ensure that the variance of the input and output data of each layer of the network is as consistent as possible, avoiding the vanishing or exploding gradient problem. The number of training epochs is set to 100, the initial learning rate is set to 0.001, and an adaptive learning rate adjustment strategy, such as the Adam optimizer, is used to automatically adjust the learning rate based on the loss function value during training. The batch size is set to 32, meaning that 32 samples are input for each training iteration.

[0132] Step S216: Train the model using the training set, input the historical physiological environment behavior linkage feature matrix into the model to obtain the risk type and risk diffusion path predicted by the model, calculate the loss function value between the prediction result and the label, and update the model parameters through the backpropagation algorithm.

[0133] In this embodiment, the cross-entropy loss function is used to measure the difference between the predicted result and the true label in the classification task. The backpropagation algorithm calculates the gradient of the loss function with respect to the parameters of each layer of the model, and then uses gradient descent to update the parameters, causing the loss function value to continuously decrease. During training, the loss function value is calculated and the model parameters are updated after each batch of data is trained.

[0134] Step S217: After a certain number of training rounds, validate the model using a validation set, calculate the model's prediction accuracy on the validation set, and adjust the model's hyperparameters, such as learning rate, number of network layers, and number of hidden units, based on the validation accuracy.

[0135] In this embodiment, the model is validated using a validation set every 10 training epochs. If the validation accuracy fails to improve after multiple consecutive iterations, the learning rate is reduced or other hyperparameters are adjusted. Validation with the validation set helps prevent the model from overfitting the training data and improves its generalization ability.

[0136] Step S218: After training, use the test set to perform a final evaluation of the model, and calculate the model's prediction accuracy, precision, recall and other metrics on the test set. If the evaluation metrics meet the preset requirements, the model training is complete; otherwise, adjust the model structure or hyperparameters and retrain.

[0137] In this embodiment, the preset requirements are a prediction accuracy of no less than 90%, and a precision and recall of no less than 85%. If the test set evaluation metrics meet these requirements, the model can be put into practical application; otherwise, it is necessary to analyze the problems of the model, such as insufficient network structure or insufficient training data, and make corresponding adjustments and retrain.

[0138] Figure 2 The illustration shows exemplary hardware and software components of a cockpit risk intervention strategy generation system 100 based on driver physiological state recognition, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the cockpit risk intervention strategy generation system 100 based on driver physiological state recognition and to perform the functions in this application.

[0139] The cockpit risk intervention strategy generation system 100 based on driver physiological state recognition can be a general-purpose server or a special-purpose server; both can be used to implement the cockpit risk intervention strategy generation method based on driver physiological state recognition of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0140] For example, a cockpit risk intervention strategy generation system 100 based on driver physiological state recognition may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the cockpit risk intervention strategy generation system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The cockpit risk intervention strategy generation system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0141] For ease of explanation, only one processor is described in the cockpit risk intervention strategy generation system 100 based on driver physiological state recognition. However, it should be noted that the cockpit risk intervention strategy generation system 100 based on driver physiological state recognition in this application may also include multiple processors. Therefore, the steps performed by one processor as described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the cockpit risk intervention strategy generation system 100 based on driver physiological state recognition performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0142] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned method for generating cockpit risk intervention strategies based on driver physiological state recognition is implemented.

[0143] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A cockpit risk intervention strategy generation method based on driver physiological state recognition, characterized in that, The method comprises: acquiring continuous physiological state data, real-time environment data in the cabin and driving operation behavior data during driving of the driver; performing multi-dimensional linkage analysis and processing on the continuous physiological state data, the real-time environment data in the cabin and the driving operation behavior data, and constructing a physiological environment behavior linkage feature matrix; calling a pre-trained cabin risk evolution prediction artificial intelligence model to perform risk evolution trend prediction processing on the physiological environment behavior linkage feature matrix, and generating a cabin risk evolution trajectory of the driver at the current time and in multiple different time periods after the current time; based on the cabin risk evolution trajectory and a pre-set intervention strategy library, generating an initial cabin risk intervention strategy for the risk diffusion path; collecting real-time state data in the cabin, dynamically comparing the real-time state data with the cabin risk evolution trajectory, performing time period adaptation and measure adjustment processing on the initial cabin risk intervention strategy, obtaining an optimized cabin risk intervention strategy, pushing the optimized cabin risk intervention strategy to a cabin control unit and recording intervention strategy execution process data; the multi-dimensional linkage analysis and processing on the continuous physiological state data, the real-time environment data in the cabin and the driving operation behavior data, and the construction of the physiological environment behavior linkage feature matrix, comprises: performing time window division processing on the continuous physiological state data, dividing the physiological data into multiple physiological data windows at a fixed time interval, calculating the change rate of each physiological index in the physiological data window, and obtaining physiological index time sequence change characteristics; performing spatial region division processing on the real-time environment data in the cabin, dividing the internal space of the cabin into multiple environment monitoring regions, collecting environment parameter data in the environment monitoring regions, calculating environment parameter difference values between different environment monitoring regions, and obtaining environment parameter spatial distribution characteristics; performing action sequence extraction processing on the driving operation behavior data, extracting driving operation action sequences in the continuous driving process, identifying the operation type and operation duration corresponding to the driving operation action sequence, and obtaining driving operation behavior characteristics; calculating the cooperative change coefficient of the physiological data window and the environment parameter difference value of the corresponding time interval and environment monitoring region, and simultaneously calculating the correlation matching degree of the physiological data window and the driving operation behavior characteristics in the corresponding time interval; integrating the physiological index time sequence change characteristics, the environment parameter spatial distribution characteristics, the driving operation behavior characteristics and the corresponding cooperative change coefficient and correlation matching degree according to a pre-set dimension order to construct a physiological environment behavior linkage feature matrix.

2. The method of claim 1, wherein, the calling of the pre-trained cabin risk evolution prediction artificial intelligence model to perform risk evolution trend prediction processing on the physiological environment behavior linkage feature matrix to generate a cabin risk evolution trajectory of the driver at the current time and in multiple different time periods after the current time, comprises: inputting the physiological environment behavior linkage feature matrix into a feature preprocessing module of the cabin risk evolution prediction artificial intelligence model, performing feature scale unification processing on different dimension features in the matrix to obtain a scale-unified linkage feature matrix; The scale-unified linkage feature matrix is transmitted to a time evolution module of the cabin risk evolution prediction artificial intelligence model, sequence correlation analysis of features in the matrix is performed in the time dimension, transfer rules of features between different time windows are extracted, and a feature time sequence evolution rule is obtained; The feature time sequence evolution rule is input into a spatial correlation module of the cabin risk evolution prediction artificial intelligence model, in combination with the spatial region identifier labeled in the matrix, mutual influence rules of environmental parameter features and physiological index features, and driving operation behavior features in different spatial regions are analyzed, and a feature spatial correlation rule is obtained; Based on the feature time sequence evolution rule and the feature spatial correlation rule, a risk evolution prediction model is constructed, linkage feature matrix data of a current period is input, changes of linkage feature matrices of multiple different periods after the current period are predicted, and a predicted linkage feature matrix sequence is obtained; Risk type matching and diffusion path analysis processing are performed on the predicted linkage feature matrix sequence, risk types corresponding to different prediction periods and diffusion directions of risks in spatial regions are identified, and a cabin risk evolution trajectory is generated.

3. The method of claim 1, wherein, Based on the cabin risk evolution trajectory, an initial cabin risk intervention strategy for a risk diffusion path is generated in combination with a preset intervention strategy library, including: The cabin risk evolution trajectory is analyzed, risk types, risk diffusion directions and spatial regions involved by risks corresponding to different periods are extracted, risk elements and intervention coverage ranges that need to be intervened in different periods are determined; A preset intervention strategy library is called, and according to the risk elements and the intervention coverage ranges, matched basic intervention measures in the intervention strategy library are searched, and a basic intervention measure set is formed; For different measures in the basic intervention measure set, execution data in a historical application process of the measures is collected, and measure start response time and measure effect data in the execution data are extracted; According to time intervals of different periods in the cabin risk evolution trajectory, in combination with the start response time of the basic intervention measures, a start advance amount corresponding to the basic intervention measures is determined, and the start time of the measures is controlled through the start advance amount; The start advance amount and the intervention coverage range of the basic intervention measures corresponding to different periods are integrated in time sequence, and an initial cabin risk intervention strategy is constructed.

4. The method of claim 1, wherein, The real-time state data in the cabin is collected, the real-time state data is dynamically compared with the cabin risk evolution trajectory, period adaptation and measure adjustment processing are performed on the initial cabin risk intervention strategy, and an optimized cabin risk intervention strategy is obtained, including: Real-time physiological state data, real-time environmental data and real-time driving operation behavior data are collected by a monitoring device in the cabin to form a real-time state data set in the cabin; Real-time physiological features, real-time environmental features and real-time driving behavior features are obtained through feature extraction processing on the real-time state data set in the cabin, and a real-time state feature matrix is constructed; Feature difference analysis is performed on the real-time state feature matrix and the predicted linkage feature matrix of a corresponding period in the cabin risk evolution trajectory, difference values of different dimension features between the two types of matrices are calculated, and a feature difference set is obtained; According to the feature difference set, it is judged whether the intervention measures of the initial cabin risk intervention strategy in the corresponding time period need to be adjusted. If the difference value exceeds the preset range, the alternative intervention measures matching the real-time state feature matrix in the intervention strategy library are selected; The start-up advance and the intervention coverage range of the adjusted intervention measures are re-determined, and the intervention measures are re-integrated in time sequence combined with the unadjusted intervention measures, to obtain an optimized cabin risk intervention strategy.

5. The method of claim 1, wherein, The synergistic change coefficient of the physiological data window and the environmental parameter difference value of the environmental monitoring area in the corresponding time interval is calculated, and the correlation matching degree of the physiological data window and the driving operation behavior feature in the corresponding time interval is calculated, including: The change rate of each physiological index in the physiological data window is extracted, the mean value of the change rate of each physiological index is calculated, and the average change rate of the physiological index is obtained; The environmental parameter difference value of different environmental monitoring areas in the corresponding time interval is extracted, the mean value of the environmental parameter difference value of different environmental monitoring areas is calculated, and the average environmental parameter difference value is obtained; The ratio of the average change rate of the physiological index to the average environmental parameter difference value is calculated, and the ratio is corrected combined with the collection frequency of the two types of data, to obtain the synergistic change coefficient; The operation type and operation duration in the driving operation behavior feature are encoded, to obtain a driving operation encoding sequence; The physiological index change rate in the physiological data window is encoded, to obtain a physiological index encoding sequence; The sequence similarity of the driving operation encoding sequence and the physiological index encoding sequence is calculated, and the sequence similarity is taken as the correlation matching degree of the physiological data window and the driving operation behavior feature.

6. The method of claim 2, wherein, The risk evolution prediction model is constructed based on the feature time sequence evolution rule and the feature space correlation rule, the linkage feature matrix data of the current time period is input, the linkage feature matrix change of multiple different time periods after the current time period is predicted, and a predicted linkage feature matrix sequence is obtained, including: The time transfer coefficient in the feature time sequence evolution rule and the space influence coefficient in the feature space correlation rule are input into a model construction module as core parameters of the risk evolution prediction model; A time dimension prediction submodule is constructed in the risk evolution prediction model, and a transfer relationship between the current time period feature and the next time period feature is established based on the time transfer coefficient; A space dimension prediction submodule is constructed in the risk evolution prediction model, and a mutual influence relationship between features of different space regions is established based on the space influence coefficient; The linkage feature matrix data of the current time period is input into the time dimension prediction submodule and the space dimension prediction submodule, and the two submodules are started simultaneously for collaborative prediction to obtain the predicted linkage feature matrix of the next time period; The process of inputting the linkage feature matrix data of the current time period into the time dimension prediction submodule and the space dimension prediction submodule, starting the two submodules simultaneously for collaborative prediction, and obtaining the predicted linkage feature matrix of the next time period is repeatedly executed, to obtain the predicted linkage feature matrix of multiple different time periods after the current time period in sequence, and to arrange the predicted linkage feature matrix sequence in time sequence.

7. The method of claim 3, wherein, The starting time and the ending time of different time periods in the cabin risk evolution trajectory are extracted, the time interval of adjacent two time periods is calculated, and a time interval sequence of the time periods is obtained. The time length from the issuance of the starting instruction of the basic intervention measure to the actual effectiveness is extracted from the historical execution data of the basic intervention measure as the starting response time of the basic intervention measure; The starting time of the risk period corresponding to the basic intervention measure is determined, and the starting time of the risk period corresponding to the basic intervention measure is taken as the target time of the effectiveness of the measure; The difference between the target time and the current time is calculated, and the starting response time is subtracted to obtain the starting advance amount initial value of the basic intervention measure; The starting advance amount initial value of the basic intervention measure is adjusted in combination with the time interval sequence of the time periods, and the starting advance amount of the basic intervention measure is finally determined. The feature difference set is used to determine whether the intervention measure of the initial cabin risk intervention strategy needs to be adjusted, and if the difference value exceeds the preset range, the alternative intervention measure matching the real-time state feature matrix in the intervention strategy library is selected, including:

8. The method of claim 4, wherein, The different difference values in the feature difference set are compared with the preset difference threshold, and the number of difference values exceeding the difference threshold and the corresponding feature dimensions are counted; If the proportion of the number of difference values exceeding the difference threshold to the total number of difference values reaches a preset proportion, or the feature dimensions corresponding to the difference values exceeding the difference threshold are key risk correlation dimensions, it is determined that the intervention measure of the corresponding time period needs to be adjusted; When it is determined that the intervention measure of the corresponding time period needs to be adjusted, the real-time state feature matrix is input into the retrieval module of the intervention strategy library, and the intervention measure with the highest matching degree with the real-time state feature matrix is retrieved; The historical execution effect data of the retrieved intervention measure is extracted, and the effect score of the retrieved intervention measure in a similar scene is calculated; The intervention measure with the highest effect score is selected as the alternative intervention measure, and the execution parameter requirement of the selected alternative intervention measure is extracted. The cabin risk intervention strategy generation system based on the physiological state recognition of the driver includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to realize the cabin risk intervention strategy generation method based on the physiological state recognition of the driver in any one of claims 1-8.

9. A cockpit risk intervention strategy generation system based on driver physiological state recognition, characterized in that, ​

Images