Dangerous driving risk quantitative evaluation and early warning grading method considering risk superposition

Abstract

The invention discloses a risk superposition-considered dangerous driving risk quantitative evaluation and early warning grading method, and relates to the field of aircraft driving safety, and the method comprises the steps: selecting a pilot dangerous driving single-source risk characterization index, and carrying out the quantification; constructing a pilot dangerous driving single-source risk quantitative evaluation model and a pilot dangerous driving multi-source risk superposition model; predicting the dangerous driving comprehensive risk of the pilot; grading the dangerous driving comprehensive risk of the pilot; sending a grading early warning signal to the pilot according to the prediction structure; the pilot dangerous driving risk quantitative evaluation model is constructed, graded early warning is performed on the result, the pilot dangerous driving behavior characteristics are comprehensively analyzed, and the pilot dangerous driving risk can be efficiently sensed, predicted and early warned, so that the aircraft accident rate is reduced, and the air transportation safety is improved.

Description

Technical Field

[0001] The present invention relates to the field of aircraft driving safety, and in particular to a method for quantitatively assessing and grading dangerous driving risks and early warnings taking into account risk superposition. Background Art

[0002] With the development of my country's economy and aviation industry, aviation transportation has become an important part of my country's transportation industry, and airplanes have become an important way for people to travel. Therefore, how to identify, detect, quantify and evaluate the risk of dangerous driving by pilots, and conduct reasonable early warning and control, is of great practical significance for regulating pilot driving behavior and ensuring aviation safety.

[0003] At present, domestic and foreign research on pilot fatigue and various emotional states mostly uses medical methods to evaluate individual fatigue and emotional levels by testing pilots' physiological and biochemical indicators. Although this method has high accuracy and clear monitoring indicators, the testing conditions are harsh and the process is complicated, so it is not suitable for use. At the same time, current research also lacks analysis of the transformation and superposition mechanism of pilot dangerous driving risks. The accuracy of the measured risks is low, and it cannot effectively predict the comprehensive risks of pilots at a certain moment in the future, resulting in untimely early warning and control measures and failure to ensure flight safety.

[0004] In order to solve the problems of harsh testing conditions, complicated processes, and untimely early warning and control measures in existing testing, the present invention aims to provide a method for quantitative assessment and early warning classification of dangerous driving risks taking into account risk superposition. Summary of the invention

[0005] In order to make up for the shortcomings of the existing technical problems, the present invention provides a method for quantitative assessment and early warning classification of dangerous driving risks taking into account risk superposition. Based on the strong correlation between the pilot's facial features and his fatigue driving state and emotional driving state, the pilot's fatigue driving risk, emotional driving risk and comprehensive risk are effectively assessed according to the video image information of the pilot's state. The test conditions are simple and easy to implement, and real-time assessment and early warning can be carried out, thereby improving the safety level of aviation transportation.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The method for quantitative assessment and early warning classification of dangerous driving risk considering risk superposition includes the following steps:

[0008] (1) Dangerous driving includes fatigue driving and emotional driving. Considering the strong correlation between pilots’ facial features and their fatigue driving status and emotional driving status, important measurement indicators representing fatigue driving risk and emotional driving risk are determined from facial features respectively;

[0009] (2) Based on the facial geometric features and 68 key points of the human face, quantitative analysis is performed on the important metrics that characterize the risk of fatigue driving and the risk of emotional driving in step 1;

[0010] (3) Construct a single-source risk quantitative assessment model for pilots' dangerous driving. The single-source risk quantitative assessment model for pilots' dangerous driving obtains the pilots' fatigue driving risk value and emotional driving risk value based on the quantified indicators.

[0011] (4) Based on the principle of inclusion and exclusion and the idea of ​​time series, a multi-source risk superposition model of pilots' dangerous driving is constructed. The multi-source risk superposition model of pilots' dangerous driving can quantitatively evaluate the multi-source risk superposition effect of pilots' dangerous driving according to the risk value of pilots' fatigue driving and emotional driving.

[0012] (5) constructing a prediction model for the comprehensive risk of dangerous driving of the pilot, obtaining the pilot's facial image in real time, and obtaining the pilot's fatigue driving risk value FR, emotional driving risk value ER, and superimposed risk value according to steps (1) to (4), and inputting the fatigue driving risk value FR, emotional driving risk value ER, and superimposed risk value into the prediction model as feature variables after standardization, to obtain the pilot's comprehensive risk prediction value of dangerous driving;

[0013] (6) Obtaining comprehensive risk values ​​of dangerous driving of multiple pilots, performing cluster analysis on the comprehensive risk values ​​of dangerous driving of pilots, classifying the comprehensive risk of dangerous driving of pilots into levels, and obtaining the threshold value of each level of risk;

[0014] (7) Based on the pilot's comprehensive risk value of dangerous driving obtained in step (5) and the threshold value of each level of risk, the risk level of the pilot is determined. When the pilot's risk level reaches the warning requirement, a graded warning signal is issued according to the level of the pilot's comprehensive risk of dangerous driving, and control measures are taken.

[0015] In the present invention, important measurement indicators for characterizing fatigue driving risk include selecting eye closing degree, yawning frequency and nodding frequency;

[0016] Important measurement indicators that characterize emotional driving risks include the angle of eye corners rising, the angle of mouth corners rising, the degree of eyebrow cohesion, the degree of eye openness, and the degree of mouth openness.

[0017] In the present invention, the construction of the pilot dangerous driving single-source risk quantitative assessment model includes:

[0018] Step 3.1, set the judgment criteria for each indicator, and divide the threshold into four levels. Different levels correspond to different scores and four levels of risk. The default level 1 risk coefficient is the highest.

[0019] Step 3.2: Establish a single-source risk model based on the scores of each indicator. The formula is:

[0020]

[0021] Where: i Represents the scores of various indicators, ranging from [0, 1], i = 1, 2, 3...n, and Risk represents the risk of fatigue driving or emotional driving.

[0022] In the present invention, a multi-source risk superposition model of pilot dangerous driving is constructed based on the principle of inclusion-exclusion and the idea of ​​time series, and the steps are as follows:

[0023] Step 4.1, collecting facial data of pilots in group Q during flight over a continuous period of time;

[0024] Step 4.2, using formula (13) to calculate the pilot's fatigue driving risk value FR and emotional driving risk value ER in group Q data respectively, and count the amount of data M at level 1 fatigue driving risk, the amount of data N at level 1 emotional driving risk, and the amount of data K at both level 1 fatigue driving risk and level 1 emotional driving risk in group Q data;

[0025] Step 4.3: Calculate the superposition probability of multiple sources of risk of dangerous driving by pilots based on the principle of inclusion and exclusion. The formula is:

[0026] P(FR∪ER)=P(FR)+P(ER)-P(FR∩ER)=M / Q+N / QK / Q (14)

[0027] Thus, a quantitative assessment of the cumulative effect of multiple sources of risk in dangerous driving by pilots can be achieved.

[0028] In the present invention, the prediction model of the comprehensive risk of dangerous driving of pilots adopts the XGBoost machine learning model, and the prediction function expression of the comprehensive risk of dangerous driving of pilots is:

[0029]

[0030] Where: is the comprehensive risk prediction value of dangerous driving of the pilot in the tth iteration, It is the characteristic variable of the comprehensive risk of dangerous driving of pilots, which is composed of the fatigue driving risk value FR, the emotional driving risk value ER and the superimposed risk value after standardization. About The prediction function of .

[0031] In the present invention, after multiple iterations of the pilot's dangerous driving comprehensive risk prediction function expression, the objective function Lt reaches a set value, and the final prediction function is obtained;

[0032] The objective function is a regularized synthesis of the loss function and the suppression function model complexity, and the expression is:

[0033]

[0034] Where: Lt is the objective function value of the tth iteration, is the loss function of a single sample, representing the error between the predicted value and the true value of the comprehensive risk of dangerous driving of the pilot. is the overall loss function for all samples,

[0035] is the sum of the regularization terms of the first t iterations, representing the complexity of the model decision tree;

[0036] The objective function is expanded by a second-order Taylor, and the expression of the expanded objective function is:

[0037]

[0038] Where: g i 、h i They are First and second order derivatives, is the loss function of a single sample in the t-1 iteration.

[0039] In the present invention, in step six, the RM-K-means++ algorithm is used to perform cluster analysis on the comprehensive risk value of pilot dangerous driving. By introducing risk factors, random initialization and multiple iterative optimization, the comprehensive risk of pilot dangerous driving is divided into primary risk, secondary risk, tertiary risk and quaternary risk:

[0040] Step 6.1, initialization phase:

[0041] Get the comprehensive risk values ​​of dangerous driving of multiple pilots in different states as a point set, randomly select a data point as the first cluster center, and for each remaining data point R i , calculate its distance D(R i ), and D(R i ) as the weight, and select the next cluster center by probability. The probability formula is as follows:

[0042]

[0043] Where: j represents the number of clusters; D(R i ) is larger, the probability of being selected as the cluster center is greater, that is, P(R i ) is larger, repeat the above steps until k initial cluster centers are selected;

[0044] Step 6.2, Risk Factor Adjustment:

[0045] When selecting the initial cluster center, the risk consequence severity σ is introduced as the risk factor to correct the distance. The corrected distance D σ (R i ) is calculated as:

[0046] D σ (R i )=D(R i )×(1+ασ(R i )) (19)

[0047] Among them, α is the adjustment parameter, σ(R i ) is the data point R i The risk value of

[0048] Step 6.3, iterative optimization:

[0049] Assignment stage: assign each data point to the nearest cluster center to form k clusters;

[0050] Update phase: Calculate the centroid of each cluster and update the cluster center. The centroid calculation formula is:

[0051]

[0052] Among them, μ j is the centroid of the jth cluster, C j are all the data points of the jth cluster;

[0053] Repeat the allocation and update phases, using the Euclidean distance to calculate and minimize each risk sample point R i The sum of the Euclidean distances to the centroid μj until the cluster center no longer changes or the preset number of iterations is reached. The Euclidean distance calculation formula is:

[0054]

[0055] Where d(R,μ) is the Euclidean distance, R ji is the i-th data point in the j-th cluster;

[0056] Step 6.4: Finally, output cluster partition C = {C1, C2, C3, C4}.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. Compared with the prior art, the present invention proposes a single-source risk quantification model based on pilot facial features and a superposition risk quantification model based on the inclusion-exclusion principle and time series thinking, which can effectively evaluate the pilot's fatigue driving risk, emotional driving risk and comprehensive risk according to the video image information of the pilot's status, thereby improving the safety level of aviation transportation.

[0059] 2. Compared with the prior art, the present invention proposes a risk grading method and a graded early warning control strategy based on the RM-K-means++ algorithm, which grades the prediction results of comprehensive risks into primary risk, secondary risk, tertiary risk and quaternary risk, and can realize comprehensive and differentiated control of pilots' dangerous driving behaviors.

[0060] 3. Compared with the existing technology, the present invention proposes a method for quantitative assessment and early warning classification of pilot dangerous driving risks covering "risk quantification, risk superposition, risk prediction, and risk classification". This method realizes the perception, prediction and early warning of two single-source risks caused by pilot dangerous driving in a short period of time - fatigue driving risk and emotional driving risk, as well as the superimposed comprehensive risks based on real-time video images. The integrated risk assessment system improves the safety of aircraft flight in all aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The figure is a flow chart of the method of the present invention.

[0062] Figure 2 Schematic diagram of 68 key points of the face in the present invention. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0064] Combination Figure 1 , a method for quantitatively assessing and grading early warning of dangerous driving considering risk superposition of the present invention is described in detail, the dangerous driving behavior of pilots refers to fatigue driving and emotional driving during the take-off and landing phases of the aircraft; the risk superposition refers to the risk superposition effect caused by the simultaneous occurrence of fatigue driving and emotional driving of the pilot as the spatiotemporal characteristics change; the method for quantitatively assessing and grading early warning of dangerous driving of pilots includes the following steps:

[0065] Step 1: Select the single-source risk characterization index of dangerous driving of pilots: Considering the strong correlation between the facial features of pilots and their fatigue driving state and emotional driving state, the present invention selects the eye closure degree P 80 , yawning frequency F Yawn , nodding frequency Fnod As an important metric to characterize the risk of fatigue driving, the eye angle θ is selected eye 、Angle of mouth corner up θ mouth , Eyebrow cohesion degree d brow , Eyes opening degree eye , mouth opening degree h mouth As an important metric to characterize the risk of emotional driving, the above facial feature information can be obtained by collecting video images of the pilot during flight through a camera.

[0066] Step 2: Based on facial geometric features and 68 key points of the face, Figure 2 As shown in the figure, the single-source risk characterization index of dangerous driving of pilots selected in step 1 of the quantitative analysis is:

[0067] Step 2.1: Quantitative analysis of facial indicators representing fatigue status:

[0068] Step 2.1.1. Mark 6 key points on the eyes and calculate the aspect ratio E of the eyes AR (i.e. the ratio of the Euclidean distance between the longitudinal landmark and the lateral landmark of the eye), calculated as:

[0069]

[0070] Determine E AR The threshold value is calculated as:

[0071] E AR =(E AR,max -E AR,min )(1-X1)+E AR,min (2)

[0072] Where: X1 is the percentage of pupil area covered by eyelids, E AR,max is the maximum eye aspect ratio, E AR,min is the minimum eye aspect ratio;

[0073] By E AR Get the number of eye closure frames and calculate the eye closure degree P from it 80 , the calculation formula is:

[0074]

[0075] Where: M is the total number of frames in the unit time series video, N is the number of frames with eyes closed in unit time;

[0076] Step 2.1.2: Mark 8 key points on the mouth and calculate the aspect ratio M of the mouth AR , the calculation formula is:

[0077]

[0078] Determine M AR The threshold value is calculated as:

[0079] M AR =(M AR,max -M AR,min )(1-X2)+M AR,min (5)

[0080] Where: X2 is the ratio of the mouth opening degree to the maximum opening degree, M AR,max is the maximum mouth aspect ratio, M AR,min is the minimum mouth aspect ratio;

[0081] The yawning frequency can be calculated based on the percentage of mouth open time, and the calculation formula is:

[0082]

[0083] Where: T m is the total detection duration, tm is the time the mouth is kept open;

[0084] Step 2.1.3: The calculation formula of head posture fatigue characteristic, i.e. nodding frequency, is:

[0085]

[0086] Where: M is the total number of frames of the unit time series video, t nod is the number of frames of head down per unit time;

[0087] Step 2.2: Quantitative analysis of facial indicators representing emotional states:

[0088] Step 2.2.1. Calculate the upward angle of the eye corner θ eye :

[0089]

[0090] Where: inner_corner is the key point 40 (left eye) or 43 (right eye), outer_corner is the key point 37 (left eye) or 46 (right eye);

[0091] Step 2.2.2, calculate the angle θ of the mouth corner mouth :

[0092]

[0093] Where: mouth_center is the key point 63, mouth_corner_right is the key point 55;

[0094] Step 2.2.3: Calculate the degree of eyebrow cohesion d brow:

[0095]

[0096] Where: brow_left_inner is key point 21, brow_right_inner is key point 24;

[0097] Step 2.2.4: Calculate the degree of eye opening h eye :

[0098] h eye =y upper_syelid -y lower_eyelid (11)

[0099] Where: upper_eyelid is the key point 38 (left eye) or 45 (right eye), lower_eyelid is the key point 42 (left eye) or 47 (right eye);

[0100] Step 2.2.5. Calculate the mouth opening degree h mouth :

[0101] h mouth =y upper_lip -y lower_lip (12)

[0102] Wherein: upper_lip is key point 52, lower_lip is key point 58.

[0103] Step 3: Construct a quantitative assessment model for the single-source risk of dangerous driving by pilots:

[0104] Step 3.1, set the judgment criteria for each indicator, and divide the threshold into four levels. Different levels correspond to different scores and four levels of risk. The default level 1 risk coefficient is the highest.

[0105] Step 3.2: Establish a single-source risk model based on the scores of each indicator. The formula is:

[0106]

[0107] Where: i (i=1, 2, 3…n) represents the score of each indicator, ranging from [0, 1], and Risk represents FR (fatigue driving risk) and ER (emotional driving risk);

[0108] Step 4: Construct a multi-source risk superposition model for pilots' dangerous driving based on the principle of inclusion-exclusion and time series thinking:

[0109] Step 4.1, collecting facial data of pilots in group Q during flight over a continuous period of time;

[0110] Step 4.2, using formula (13) to calculate the pilot's fatigue driving risk value FR and emotional driving risk value ER in group Q data respectively, and count the amount of data M at level 1 fatigue driving risk, the amount of data N at level 1 emotional driving risk, and the amount of data K at both level 1 fatigue driving risk and level 1 emotional driving risk in group Q data;

[0111] Step 4.3: Calculate the superposition probability of multiple sources of risk of dangerous driving by pilots based on the principle of inclusion and exclusion. The formula is:

[0112] P(FR∪ER)=P(FR)+P(ER)-P(FR∩ER)=M / Q+N / QK / Q (14)

[0113] Thus, the quantitative assessment of the superposition effect of multiple sources of risk in dangerous driving of pilots can be achieved;

[0114] Step 5: Use the XGBoost (extreme gradient boosting) machine learning model to predict the comprehensive risk of dangerous driving of pilots:

[0115] Step 5.1: The comprehensive risk prediction function expression of the pilot's dangerous driving is:

[0116]

[0117] Where: is the comprehensive risk prediction value of dangerous driving of the pilot in the tth iteration, It is the characteristic variable of the comprehensive risk of dangerous driving of pilots, which is composed of the fatigue driving risk value FR, the emotional driving risk value ER and the superimposed risk value after standardization. About The prediction function of

[0118] Step 5.2: Combine the loss function and the regularization of the suppression function model complexity into the objective function. The expression of the objective function is:

[0119]

[0120] Where: Lt is the objective function value of the tth iteration, is the loss function of a single sample, representing the error between the predicted value and the true value of the comprehensive risk of dangerous driving of the pilot. is the overall loss function for all samples,

[0121] is the sum of the regularization terms of the first t iterations, representing the complexity of the model decision tree;

[0122] Step 5.3: Perform a second-order Taylor expansion on the objective function. The expression of the expanded objective function is:

[0123]

[0124] Where: g i 、h i They are First and second order derivatives, is the loss function of a single sample in the t-1 iteration;

[0125] Step 5.4, after multiple iterations, the objective function Lt reaches the set value, indicating that the function accuracy meets the requirements. At this time, the pilot's dangerous driving comprehensive risk prediction function is the final prediction function;

[0126] Step 6: Use the RM-K-means++ algorithm to perform cluster analysis on the comprehensive risk index R of pilots' dangerous driving. By introducing risk factors, random initialization, and multiple iterative optimizations, the comprehensive risk of pilots' dangerous driving is divided into primary risk, secondary risk, tertiary risk, and quaternary risk:

[0127] Step 6.1, initialization phase:

[0128] Get the comprehensive risk values ​​of dangerous driving of multiple pilots in different states as a point set, randomly select a data point as the first cluster center, and for each remaining data point R i , calculate its distance D(R i ), and D(R i ) as the weight, and select the next cluster center by probability. The probability formula is as follows:

[0129]

[0130] Where: j represents the number of clusters; D(R i ) is larger, the probability of being selected as the cluster center is greater, that is, P(R i ) is larger, repeat the above steps until k initial cluster centers are selected;

[0131] Step 6.2, Risk Factor Adjustment:

[0132] When selecting the initial cluster center, the risk factor (risk consequence severity) σ is introduced to correct the distance. The corrected distance D σ (R i ) is calculated as:

[0133] D σ (R i )=D(R i )×(1+ασ(R i )) (19)

[0134] Among them, α is the adjustment parameter, σ(R i ) is the data point R i The risk value of

[0135] Step 6.3, iterative optimization:

[0136] Assignment stage: assign each data point to the nearest cluster center to form k clusters;

[0137] Update phase: Calculate the centroid of each cluster and update the cluster center. The centroid calculation formula is:

[0138]

[0139] Among them, μ j is the centroid of the jth cluster, C j are all the data points of the jth cluster;

[0140] Repeat the allocation and update phases, using the Euclidean distance to calculate and minimize each risk sample point R i To the center of mass μ j The sum of the Euclidean distances is calculated until the cluster center no longer changes or the preset number of iterations is reached. The Euclidean distance calculation formula is:

[0141]

[0142] Where d(R,μ) is the Euclidean distance, R ji is the i-th data point in the j-th cluster;

[0143] Step 6.4, finally output cluster partition C = {C1, C2, C3, C4};

[0144] Step 7. Obtain the comprehensive risk thresholds of dangerous driving for pilots of four risk levels through step 6. Use the facial image acquired in real time and steps 1 to 5 to predict the comprehensive risk of dangerous driving for the pilot. Determine which risk level the comprehensive risk of dangerous driving for the pilot belongs to. If the current risk level of the pilot requires an early warning, issue an early warning based on the level and take control measures. Otherwise, continue real-time monitoring until the end.

[0145] What is described above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for quantitative assessment and early warning classification of dangerous driving risks considering risk superposition, characterized in that: The following steps are involved: (1) Dangerous driving includes fatigue driving and emotional driving. Considering the strong correlation between pilots’ facial features and their fatigue driving status and emotional driving status, important measurement indicators representing fatigue driving risk and emotional driving risk are determined from facial features respectively; (2) Based on the facial geometric features and 68 key points of the human face, quantitative analysis is performed on the important metrics that characterize the risk of fatigue driving and the risk of emotional driving in step 1; (3) Construct a single-source risk quantitative assessment model for pilots' dangerous driving. The single-source risk quantitative assessment model for pilots' dangerous driving obtains the pilots' fatigue driving risk value and emotional driving risk value based on the quantified indicators. (4) Based on the principle of inclusion and exclusion and the idea of ​​time series, a multi-source risk superposition model of pilots' dangerous driving is constructed. The multi-source risk superposition model of pilots' dangerous driving can quantitatively evaluate the multi-source risk superposition effect of pilots' dangerous driving according to the risk value of pilots' fatigue driving and emotional driving. (5) constructing a prediction model for the comprehensive risk of dangerous driving of the pilot, obtaining the pilot's facial image in real time, and obtaining the pilot's fatigue driving risk value FR, emotional driving risk value ER, and superimposed risk value according to steps (1) to (4), and inputting the fatigue driving risk value FR, emotional driving risk value ER, and superimposed risk value into the prediction model as feature variables after standardization, to obtain the pilot's comprehensive risk prediction value of dangerous driving; (6) Obtaining comprehensive risk values ​​of dangerous driving of multiple pilots, performing cluster analysis on the comprehensive risk values ​​of dangerous driving of pilots, classifying the comprehensive risk of dangerous driving of pilots into levels, and obtaining the threshold value of each level of risk; (7) Based on the comprehensive risk value of the pilot's dangerous driving obtained in step (5) and the threshold value of each level of risk, the risk level of the pilot is determined. When the risk level of the pilot reaches the required warning level, a graded warning signal is issued according to the pilot's risk level, and control measures are taken.

2. The method for quantitative assessment and early warning classification of dangerous driving risks considering risk superposition according to claim 1 is characterized in that: Important metrics to characterize the risk of fatigue driving include the degree of eye closure, yawning frequency, and nodding frequency; Important measurement indicators that characterize emotional driving risks include the angle of eye corners rising, the angle of mouth corners rising, the degree of eyebrow cohesion, the degree of eye openness, and the degree of mouth openness.

3. The method for quantitative assessment and early warning classification of dangerous driving risks considering risk superposition according to claim 1 is characterized in that: The construction of the single-source risk quantitative assessment model for dangerous driving of pilots includes: Step 3.1, set the judgment criteria for each indicator, and divide the threshold into four levels. Different levels correspond to different scores and four levels of risk. The default level 1 risk coefficient is the highest. Step 3.2: Establish a single-source risk model based on the scores of each indicator. The formula is: Where: i Represents the scores of various indicators, ranging from [0, 1], i = 1, 2, 3...n, and Risk represents the risk of fatigue driving or emotional driving.

4. The method for quantitative assessment and early warning classification of dangerous driving risks considering risk superposition according to claim 1 is characterized in that: Based on the principle of inclusion-exclusion and the idea of ​​time series, a multi-source risk superposition model of dangerous driving for pilots is constructed. The steps are as follows: Step 4.1, collecting facial data of pilots in group Q during flight over a continuous period of time; Step 4.2, using formula (13) to calculate the pilot's fatigue driving risk value FR and emotional driving risk value ER in group Q data respectively, and count the amount of data M at level 1 fatigue driving risk, the amount of data N at level 1 emotional driving risk, and the amount of data K at both level 1 fatigue driving risk and level 1 emotional driving risk in group Q data; Step 4.3: Calculate the superposition probability of multiple sources of risk of dangerous driving by pilots based on the principle of inclusion and exclusion. The formula is: P(FR∪ ER)=P(FR)+P(ER)-P(FR∩ ER)=M / Q+N / QK / Q (14) Thus, a quantitative assessment of the cumulative effect of multiple sources of risk in dangerous driving by pilots can be achieved.

5. The method for quantitative assessment and early warning classification of dangerous driving risks considering risk superposition according to claim 1 is characterized in that: The prediction model of the comprehensive risk of dangerous driving of pilots adopts the XGBoost machine learning model, and the prediction function expression of the comprehensive risk of dangerous driving of pilots is: Where: is the comprehensive risk prediction value of dangerous driving of the pilot in the tth iteration, It is the characteristic variable of the comprehensive risk of dangerous driving of pilots, which is composed of the fatigue driving risk value FR, the emotional driving risk value ER and the superimposed risk value after standardization. For about The prediction function of .

6. The method for quantitative assessment and early warning classification of dangerous driving risks considering risk superposition according to claim 5 is characterized in that: After multiple iterations, the objective function L t Reach the set value and get the final prediction function; The objective function is a regularized synthesis of the loss function and the suppression function model complexity, and the expression is: Where: L t is the objective function value of the tth iteration, is the loss function of a single sample, representing the error between the predicted value and the true value of the comprehensive risk of dangerous driving of the pilot. is the overall loss function for all samples, is the sum of the regularization terms of the first t iterations, representing the complexity of the model decision tree; The objective function is expanded by a second-order Taylor, and the expression of the expanded objective function is: Where: g i 、h i They are First and second order derivatives, is the loss function of a single sample in the t-1 iteration.

7. The method for quantitative assessment and early warning classification of dangerous driving risks considering risk superposition according to claim 5 is characterized in that: In step six, the RM-K-means++ algorithm is used to perform cluster analysis on the comprehensive risk value of pilots' dangerous driving. By introducing risk factors, random initialization and multiple iterative optimization, the comprehensive risk of pilots' dangerous driving is divided into primary risk, secondary risk, tertiary risk and quaternary risk: Step 6.1, initialization phase: Get the comprehensive risk values ​​of dangerous driving of multiple pilots in different states as a point set, randomly select a data point as the first cluster center, and for each remaining data point R i , calculate its distance D(R i ), and D(R i ) as the weight, and select the next cluster center by probability. The probability formula is as follows: Where: j represents the number of clusters; D(R i ) is larger, the probability of being selected as the cluster center is greater, that is, P(R i ) is larger, repeat the above steps until k initial cluster centers are selected; Step 6.2, Risk Factor Adjustment: When selecting the initial cluster center, the risk consequence severity σ is introduced as the risk factor to correct the distance. The corrected distance D σ (R i ) is calculated as: D σ (R i )=D(R i )×(1+ασ(R i )) (19) Among them, α is the adjustment parameter, σ(R i ) is the data point R i The risk value of Step 6.3, iterative optimization: Assignment stage: assign each data point to the nearest cluster center to form k clusters; Update phase: Calculate the centroid of each cluster and update the cluster center. The centroid calculation formula is: Among them, μ j is the centroid of the jth cluster, i.e., the threshold of the jth level risk, C j are all the data points of the jth cluster; Repeat the allocation and update phases, using the Euclidean distance to calculate and minimize each risk sample point R i To the center of mass μ j The sum of the Euclidean distances is calculated until the cluster center no longer changes or the preset number of iterations is reached. The Euclidean distance calculation formula is: Where d(R,μ) is the Euclidean distance, R ji is the i-th data point in the j-th cluster; Step 6.4: Finally, output cluster partition C = {C1, C2, C3, C4}.

Images