A method for designing a training set of a quality evaluation model for aircraft special situation training of a flight simulator

Abstract

The application discloses a kind of aircraft special situation training quality evaluation model training set design methods of flight simulator, including step one, data acquisition, step two, data preprocessing, step three, the calculation of SHAP value, step four, feature weight calculation and weighted adjustment, step five, the retraining of model, step six, the determination of aircraft special situation training feature parameter, step seven, the building of evaluation model training set;The application is collected and preprocessed to flight simulator training data, and the method of using SHAP value is combined with weighted random forest model to select feature parameter, to build scientific, effective training quality evaluation model training set, not only can improve the efficiency and accuracy of data processing, but also can significantly improve the objectivity and reliability of special situation training quality evaluation, provide solid guarantee for the cultivation of pilot and flight safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aviation training and flight simulation, and particularly relates to a flight simulator aircraft special situation training quality evaluation model training set design method. BACKGROUND

[0002] With the rapid development of aviation technology, the training needs of pilots are increasingly complex and diverse, especially in dealing with aircraft special situations (special situations), the adaptability and control skills of pilots are crucial to ensure flight safety; however, traditional training methods often fail to comprehensively cover various special situations and cannot effectively quantify the training effect of pilots; therefore, relying on high-fidelity flight simulators for pilot special situation training and scientifically and objectively evaluating the training quality have become one of the important research directions in the current aviation field.

[0003] As an advanced training tool, high-fidelity flight simulators can realistically simulate various flight conditions and special situations, providing a highly realistic training environment for pilots; however, in special situation training, how to effectively extract feature parameters related to training quality from a large amount of flight data and establish an accurate and reliable training quality evaluation model remains a difficult problem to be solved.

[0004] Current research typically relies on manual experience or simple statistical methods for special situation training quality evaluation, which has problems such as non-uniform evaluation standards and strong subjectivity, making it difficult to meet the high standards of modern aviation training, therefore, the present application proposes a flight simulator aircraft special situation training quality evaluation model training set design method to solve the problems in the prior art. SUMMARY

[0005] To solve the above problems, the present application proposes a flight simulator aircraft special situation training quality evaluation model training set design method, which collects and preprocesses flight simulator training data, uses SHAP values combined with a weighted random forest model to select feature parameters, and thus constructs a scientific and effective training quality evaluation model training set, which not only improves the efficiency and accuracy of data processing, but also significantly improves the objectivity and reliability of special situation training quality evaluation, providing a solid guarantee for pilot training and flight safety.

[0006] To achieve the purpose of the present application, the present application realizes the following technical scheme: a flight simulator aircraft special situation training quality evaluation model training set design method, comprising the following steps:

[0007] Step one, data collection, through the flight simulator training evaluation system to collect the aircraft emergency situation data including aircraft state data, pilot control data and fault injection data, and the training data of the flight student when the aircraft is normal;

[0008] Step two, data preprocessing, first replace the infinite and maximum values in the collected data with NAN, and fill them with the average value of the column, then use Lagrange interpolation method to complete the missing data, and finally combine the aircraft state data and pilot control data into one data and add a label column;

[0009] Step three, SHAP value calculation, define a basic random forest model through "clf=RandomForestClassifier(random_state=42)", and calculate the SHAP value after preliminary training through Tree SHAP method;

[0010] Step four, feature weight calculation and weighted adjustment, based on SHAP value, the feature importance is evaluated, the feature weight calculation value and weighted adjustment are completed;

[0011] Step five, model retraining, after calculating the SHAP value of each feature, the corresponding weight is obtained and the weight is applied to the feature matrix X to generate the weighted feature matrix Through the feature in the weighted feature matrix , the random forest model is retrained, and the grid search method is used to find the optimal model parameters to obtain the optimized weighted random forest model;

[0012] Step six, determination of aircraft emergency training feature parameters, for a certain determined aircraft emergency training subject, select n groups of training data to input the optimized weighted random forest model to select feature parameters, view the model output result, and determine the aircraft emergency training feature parameters;

[0013] Step seven, building of model training set, based on the determined aircraft emergency training feature parameters, the corresponding feature parameters are extracted from the original data to form new data, and further build the aircraft emergency training quality evaluation model training set.

[0014] Further improvement lies in that the Lagrange interpolation method in step two is calculated by the following formula ;

[0015] Where x represents the position of the independent variable, y represents the value of the corresponding function, represents the Lagrange polynomial.

[0016] Further improvement lies in that the label column adding method in the step two is that the label of the aircraft state data of the flight student during training is assigned as 1 when the aircraft appears a special situation, and the label of the aircraft state data of the flight student during training is assigned as 0 when the aircraft is normal; the label of the flight student control data is assigned as 1 when the aircraft appears a special situation, and the label of the flight student control data is assigned as 0 when the aircraft is normal.

[0017] Further improvement lies in that the random forest model defined in the step three includes the number of decision trees n_estimators, the maximum number of features max_features considered at each split, the maximum depth of the decision tree max_depth and the split criterion criterion; and the SHAP contribution value of each feature is obtained through “shap.TreeExplainer” after the SHAP value is calculated.

[0018] Further improvement lies in that the SHAP value calculation process in the step three is as follows:

[0019] S1, the target of the SHAP value is to calculate the marginal contribution of each feature to the model prediction value, which is calculated by the following formula ;

[0020] wherein, represents the SHAP value of feature i, N is the set of all features, S is the feature subset, is the prediction output of the model under the feature subset S;

[0021] S2, for the decision tree model in the random forest model, Tree SHAP is used to calculate the marginal contribution of the feature on each node of the model according to the split condition of the feature on the path;

[0022] S3, for each path in each decision tree model, Tree SHAP calculates the SHAP value of each feature through a recursive algorithm, and for each node, the probability of the decision path from the root node to the leaf node is assumed as the contribution of each feature to the prediction value is the recursive distribution of the SHAP value is carried out using the following formula wherein, represents the SHAP value of feature i in the sample x in a single tree, is the probability of the sample x passing through feature i in the path j, is the size of the feature subset that has passed in the path, is the prediction value of feature i after the feature is added to the feature subset z;

[0023] S4, for each path, Tree SHAP needs to calculate the contribution of each feature to the output of the leaf node from the root node to the leaf node, and the prediction value of the tree is The values ​​of the leaf nodes are used, and the contribution of each path is calculated by the following formula. in, The summation of contributions from all paths within a single tree represents the SHAP contribution of feature i to sample x, where T represents the number of trees. is the weight of the feature subset z, representing the probability of that subset appearing on the path. It is the contribution increment after feature i is added;

[0024] S5. In the entire random forest model, Tree SHAP is obtained by summing the SHAP values ​​of each decision tree using the following formula to obtain the final SHAP value. in, This represents the final SHAP value of feature i for sample x throughout the entire random forest. It is the SHAP value of feature i in the t-th decision tree.

[0025] The further improvement lies in: step four specifically is as follows:

[0026] Step 1: Perform absolute value processing on the SHAP value calculated in Step 3, using... This represents the absolute value of the contribution of feature i to sample x;

[0027] The second step is to calculate the absolute average of the SHAP values ​​of all samples using the following formula. Where n is the number of samples in the dataset. It is the absolute value of the SHAP value of the j-th sample on feature i;

[0028] The third step is to normalize the average SHAP value of each feature, expressed by the following formula: in, It is the weight of feature i. is the absolute value of the average SHAP of feature i, d is the total number of features, and the sum of the weights of all features after normalization is equal to 1;

[0029] Step 4: First, assume a feature matrix X with shape n×d, where n is the number of samples and d is the number of features. Each column of this matrix represents a feature, and each row represents a sample. X = [x1, x2, ..., xd]. Use X... i This represents the value of the i-th feature. For each feature X... i Use the calculated weight W i After weighted adjustment, the weighted feature matrix is: Where W is a diagonal matrix, W = diag Weighted features It is achieved by multiplying each feature column by its corresponding weight W. i The obtained, and weighted, features are represented as follows:

[0030] A further improvement is that, in step six, determining the characteristic parameters for aircraft emergency training specifically means that if a certain parameter appears more than or equal to n / 2 times in the output result, then that parameter is determined as a characteristic parameter related to this type of aircraft emergency training subject.

[0031] The beneficial effects of this invention are as follows: By preprocessing the data collected by the flight simulator, this invention greatly improves the quality and processing efficiency of the data, making the training data more accurate and reliable, reducing the impact of noise and data bias, and using the SHAP value combined with the weighted random forest model to screen out the key feature parameters related to the special situation training subjects, making the feature selection process more objective and accurate.

[0032] Furthermore, this method can be flexibly adjusted according to different aircraft emergency training subjects. By inputting and filtering multiple sets of data, it ensures that the selected feature parameters are highly correlated with the specific training subjects, thereby improving the applicability and universality of the evaluation model. Attached Figure Description

[0033] Figure 1 This is a flowchart of the aircraft special situation training characteristic parameter selection method in Embodiment 1 of the present invention.

[0034] Figure 2 This is a training data diagram of a single-engine failure of an aircraft after preprocessing in Embodiment 2 of the present invention.

[0035] Figure 3 This is a schematic diagram of the SHAP values ​​of some parameters in the single-shot failure training data of Embodiment 2 of the present invention.

[0036] Figure 4 This is a diagram illustrating the parameter weights of the training data for a single-shot failure in Embodiment 2 of the present invention.

[0037] Figure 5 This is a diagram illustrating the output of the single-shot failure training subject model in Embodiment 2 of the present invention.

[0038] Figure 6 The diagram illustrates the training characteristic parameters of a single-engine failure of an aircraft selected for Embodiment 2 of the present invention.

[0039] Figure 7 This is a diagram of the training set file of the aircraft single-engine failure training quality assessment model that has been built in Embodiment 2 of the present invention. Detailed Implementation

[0040] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0041] Example 1

[0042] according to Figure 1 As shown, this embodiment provides a method for designing a training set for an aircraft emergency training quality assessment model for a flight simulator, including the following steps:

[0043] Step 1: During flight training using a high-fidelity flight simulator, various systems and subsystems of the aircraft will generate a large amount of diverse data, including but not limited to flight status data, environmental parameter data, pilot operation data, and fault injection data. The flight simulator training debriefing system (BDS) collects flight status data, pilot operation data, and fault injection data, as well as flight student training data when the aircraft is in special condition or when the aircraft is in normal condition.

[0044] Step 2: Preprocess the collected aircraft status data, pilot control data, and fault injection data. These data are a time-series data sequence arranged in chronological order. Due to the inherent limitations of the flight simulator training and debriefing system (BDS), the collected aircraft system data, pilot control data, and fault injection data will inevitably have significant deviations, garbled characters, and data loss issues. Abnormal data needs to be deleted or corrected.

[0045] Replace infinity and maximum values ​​in the collected data with NAN, and fill the gaps with the average value of the corresponding column. For missing data, use Lagrange interpolation to complete the missing data. The formula for Lagrange interpolation to complete missing data is as follows: Where x represents the position of the independent variable, and y represents the value of the corresponding function. Represents a Lagrange polynomial;

[0046] The processed aircraft status data during emergency situations and during normal flight training are combined into a single dataset. A label column is added to this combined dataset, with the aircraft status data during emergency situations assigned a label of 1, and the aircraft status data during normal flight training assigned a label of 0. Similarly, the control data during emergency situations and during normal flight training are also combined into a single dataset, and a label column is added to this combined dataset, with the control data during emergency situations assigned a label of 1, and the control data during normal flight training assigned a label of 0.

[0047] Step 3: Define a basic random forest model using "clf=RandomForestClassifier(random_state=42)", including the number of decision trees n_estimators, the maximum number of features to consider during each split max_features, the maximum depth of the decision trees max_depth, and the splitting criterion. Calculate the SHAP value after initial training using the Tree SHAP method, and obtain the SHAP contribution value of each feature using "shap.TreeExplainer".

[0048] Tree SHAP is a specialization of the SHAP (SHapley Additive exPlanations) method. It is used to accelerate the calculation of SHAP values ​​for tree models (such as random forests, gradient boosting trees, etc.). Tree SHAP utilizes the structure of decision trees to avoid traversing all feature subsets, thereby significantly speeding up the calculation. Its computational complexity is linear and depends on the depth of the tree.

[0049] The SHAP value is calculated as follows:

[0050] S1, Definition of SHAP value

[0051] The SHAP value is based on the concept of Shapley value and aims to calculate the marginal contribution of each feature to the model's predicted value. The Shapley value is the average marginal contribution based on feature combinations. The formula for calculating the SHAP value is as follows: in, Let S represent the SHAP value of feature i, where N is the set of all features and S is a subset of features. The model's predicted output under feature subset S;

[0052] S2, Decision Tree Structure and Path

[0053] In decision tree models, Tree SHAP utilizes the structure of a tree to optimize the calculation process of Shapley values. Each tree has multiple paths, and samples are predicted along the tree from the root node to the leaf node. For each node in the tree, the feature value determines which branch the sample should follow. The idea behind Tree SHAP is to calculate the marginal contribution of features at each node based on the splitting conditions of features on the path, without having to traverse all feature subsets.

[0054] S3. Allocation of Path Dependency Weights

[0055] For each path in each decision tree model, Tree SHAP calculates the SHAP value of each feature using a recursive algorithm. For each node, the probability of a decision path leading from the root node to a leaf node is assumed to be... The contribution of each feature to the predicted value is: Tree SHAP uses the following formula to recursively assign SHAP values. in, This represents the SHAP value of feature i in sample x within a single tree. It is the probability that sample x passes through feature i in path j. It is the size of the feature subset that has already been traversed in the path. It is the predicted value of feature i after it is added to feature subset z;

[0056] S4. Cumulative Expectations and Contributions

[0057] For each path, from the root node to the leaf node, Tree SHAP needs to calculate the contribution of each feature to the output of the leaf node, and the predicted value of the tree. The values ​​of the leaf nodes are used, and the contribution of each path is calculated by the following formula. in, The summation of contributions from all paths within a single tree represents the SHAP contribution of feature i to sample x, where T represents the number of trees. is the weight of the feature subset z, representing the probability of that subset appearing on the path. It is the contribution increment after feature i is added;

[0058] The final sum of S5 and SHAP values

[0059] In the entire random forest model, Tree SHAP is obtained by summing the SHAP values ​​of each decision tree using the following formula: in, is the SHAP value of feature i in the t-th decision tree. The final SHAP value is the sum of the contributions of all trees.

[0060] Step 4, Feature Weight Calculation and Weighting Process: This process mainly assesses the importance of features based on SHAP values, and completes the feature weight calculation and weighting adjustment.

[0061] The process is as follows:

[0062] The first step is to use the SHAP value calculated in the preliminary model training in step three. Since SHAP values ​​can be positive or negative (reflecting the positive or negative contribution of features to the result), the absolute value of the SHAP value is taken here to calculate the importance weight of the features. SHAP represents the absolute value of feature i's contribution to sample x; the purpose of taking the absolute value of the SHAP value calculated for each sample is to measure the relative influence of the feature, regardless of whether it is positive or negative.

[0063] The second step is to calculate the average contribution of each feature across the entire dataset, using the following formula to calculate the absolute average of the SHAP values ​​of all samples. Where n is the number of samples in the dataset. It is the absolute value of the SHAP value of the j-th sample on feature i, calculated by... The average contribution of feature i can be obtained, representing the magnitude of its influence on the entire dataset;

[0064] Third, to convert feature importance into weights, the average SHAP value of each feature is normalized, as expressed by the following formula. Among them, W i It is the weight of feature i. is the average absolute value of the SHAP of feature i, and d is the total number of features. After normalization, the sum of all feature weights equals 1, which makes the weights comparable and facilitates subsequent feature matrix weighting.

[0065] Step 4: First, assume a feature matrix X with shape n×d, where n is the number of samples and d is the number of features. Each column of this matrix represents a feature, and each row represents a sample. X = [x1, x2, ..., xd]. Use X... i This represents the value of the i-th feature. For each feature X... i Use the calculated weight W i After weighted adjustment, the weighted feature matrix is: Where W is a diagonal matrix, W = diag Weighted features It is achieved by multiplying each feature column by its corresponding weight W. i The obtained, and weighted, features are represented as follows:

[0066] After the above four steps, the value of each feature in the feature matrix will be adjusted according to the importance of its SHAP value; features with higher weights will have a greater impact on the model, reflecting their importance in the model, while features with lower weights will have a smaller impact on the model.

[0067] Step 5: Retrain the model, calculate the SHAP value of each feature, and obtain the corresponding weight W. i The weights are then applied to the feature matrix X to generate a weighted feature matrix. For each sample x, the weighted feature values

[0068] Using the weighted feature matrix To replace the original feature matrix X,

[0069] The random forest model is retrained using weighted features, and the optimal model parameters are found using a grid search method to obtain the optimized weighted random forest model. By subjectively setting a threshold (such as the top n in importance or the importance exceeding a certain threshold), parameters higher than the threshold are selected as the output of the random forest model.

[0070] Step Six: For a specific aircraft emergency training subject (such as single-engine failure training), the feature parameters selected using a set of training data by combining SHAP values ​​with a weighted random forest model are not accurate enough. For a specific aircraft emergency training subject, select n sets of training data and use the SHAP value combined with a weighted random forest model to select feature parameters. Observe the model output results. When a certain parameter appears more than or equal to n / 2 times in the output results, then that parameter is determined as a feature parameter related to this type of aircraft emergency training subject. For a specific aircraft emergency training subject, the total number of feature parameters selected that are related to this aircraft emergency training can be determined subjectively.

[0071] Step 7: Based on the determined aircraft emergency training feature parameters, extract the corresponding feature parameters from the original data and combine them into new data, and further build a training set for the aircraft emergency training quality assessment model.

[0072] Example 2

[0073] according to Figures 2-7 As shown in the figure, this embodiment takes flight trainees conducting single-engine failure training as an example to introduce in detail the process of building the training set of the single-engine failure training quality assessment model.

[0074] 1. Single-shot failure training data acquisition and preprocessing

[0075] First, training data is collected from flight trainees during single-engine failure training and training data during normal aircraft conditions using the flight simulator training and review system (BDS). The collected training data includes aircraft status data, pilot control data, and fault injection data.

[0076] For infinity and maximum values ​​in the data, first replace them with NAN, and then fill them with the average value of the column; for missing data, use Lagrange interpolation to complete the data.

[0077] The processed aircraft status data during single-engine failure training and the aircraft status data during normal training are combined into a single dataset. A label column is added to this combined dataset, assigning a label of 1 to the aircraft status data during single-engine failure training and a label of 0 to the aircraft status data during normal training. Similarly, the control data during single-engine failure training and the aircraft status data during normal training are also combined into a single dataset, and a label column is added to this combined dataset. The control data during emergency situations is assigned a label of 1, and the control data during normal training is assigned a label of 0. The preprocessed data is shown in the appendix to the instruction manual. Figure 2 As shown.

[0078] 2. Selection of training characteristic parameters for single-engine failure of aircraft

[0079] First, a basic random forest model is defined using "clf=RandomForestClassifier(random_state=42)". The parameters are set as follows: "n_estimators=100" (number of decision trees), "max_features=sqrt" (maximum number of features considered during each split), "max_depth=None" (maximum depth of the decision trees), and "criterion=squard-error" (split criterion). Then, the preprocessed single-failure training data is input into the random forest model for initial training, and the SHAP contribution value of each parameter is calculated. The SHAP values ​​of some parameters in the single-failure training data are shown in the attached documentation. Figure 3 As shown.

[0080] Secondly, the absolute values ​​of the SHAP values ​​of each parameter in the single-shot failure training data are summed and normalized. Then, the weight of each parameter is calculated. The weights of some parameters in the single-shot failure training data are shown in the appendix of the instruction manual. Figure 4 As shown, the feature matrix is ​​weighted and adjusted using these weights.

[0081] Finally, the random forest model was retrained based on the weighted feature matrix. The hyperparameters were optimized using a grid search method to find the optimal "n_estimators", "max_features", "max_depth", and "criterion" parameters. The model output was then determined by a subjectively set threshold—the top 20 most important parameter names were used as the model's output. The model output for single-failure training is shown in the appendix to the instruction manual. Figure 5 As shown.

[0082] For a specific aircraft emergency training subject, the feature parameters selected using a set of training data by combining SHAP values ​​with a weighted random forest model are not accurate enough. For a specific aircraft emergency training subject, n sets of training data can be selected, and the feature parameters can be selected by combining SHAP values ​​with a weighted random forest model. The output results of the random forest model can be viewed. If a certain parameter appears more than or equal to n / 2 times in the output results, then that parameter is determined as a feature parameter related to this type of aircraft emergency training subject.

[0083] This example uses five sets of training data to select training feature parameters for single-engine failure of an aircraft. If a parameter appears more than or equal to three times in the output of the random forest model, it is considered a training feature parameter for single-engine failure of the aircraft. Through experiments, the selected training feature parameters for single-engine failure of the aircraft are shown in the appendix of the instruction manual. Figure 6 As shown.

[0084] 3. Build a training set for the single-engine failure training quality assessment model for aircraft.

[0085] Based on the experimental results, the selected single-engine failure training feature parameters were extracted from the original data and combined to form new data, which was then used to build a training set for the single-engine failure training quality assessment model. The completed training set for the single-engine failure training quality assessment model is shown in the appendix to the instruction manual. Figure 7 As shown.

[0086] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for designing a training set for an aircraft emergency training quality assessment model for a flight simulator, characterized in that, Includes the following steps: Step 1: Data Acquisition. The flight simulator training and debriefing system is used to collect data including aircraft status data, pilot operation data, and fault injection data. The data includes data on special situations that occur when the aircraft is in operation and training data of flight trainees when the aircraft is in normal operation. Step 2: Data preprocessing. First, replace the infinite and maximum values ​​in the collected data with NAN and fill them with the average value of the column. Then, use Lagrange interpolation to fill in the missing data. Finally, combine the aircraft status data and pilot operation data into one data set and add a label column. Step 3: Calculate the SHAP value. Define a basic random forest model using "clf=RandomForestClassifier" and calculate the SHAP value after initial training using the Tree SHAP method. Step 4: Feature weight calculation and weighted adjustment. The importance of features is evaluated based on the SHAP value, and the feature weights are calculated and adjusted. Step 5: Retrain the model, calculate the SHAP value of each feature, and obtain the corresponding weight w. i The weights are then applied to the feature matrix X to generate a weighted feature matrix X. weighted Through the weighted feature matrix X weighted The random forest model was retrained using the features in the model, and the grid search method was used to find the optimal model parameters to obtain the optimized weighted random forest model. Step 6: Determining the characteristic parameters of aircraft emergency training. For a specific aircraft emergency training subject, select n sets of training data and input them into the optimized weighted random forest model to select characteristic parameters. Check the model output results and determine the characteristic parameters of aircraft emergency training. Step 7: Evaluate the construction of the model training set. Based on the determined aircraft emergency training feature parameters, extract the corresponding feature parameters from the original data and combine them into new data, and further construct the training set of the aircraft emergency training quality evaluation model.

2. The method for designing a training set for an aircraft emergency training quality assessment model for a flight simulator according to claim 1, characterized in that: In step two, the Lagrange interpolation method is used to complete the missing data, which is calculated by the following formula. Where x represents the position of the independent variable, and y represents the value of the corresponding function. It represents a Lagrange polynomial.

3. The method for designing a training set for an aircraft special situation training quality assessment model for a flight simulator according to claim 1, characterized in that: In step two, the label column is added as follows: when the aircraft is in an emergency, the aircraft status data during the flight trainee's training is labeled as 1, and when the aircraft is in normal condition, the aircraft status data during the flight trainee's training is labeled as 0; when the aircraft is in an emergency, the flight trainee's control data is labeled as 1, and when the aircraft is in normal condition, the flight trainee's control data is labeled as 0.

4. The method for designing a training set for an aircraft emergency training quality assessment model for a flight simulator according to claim 1, characterized in that: The random forest model defined in step three includes the number of decision trees n_estimators, the maximum number of features considered in each split max_features, the maximum depth of the decision trees max_depth, and the splitting criterion; after calculating the SHAP value, the SHAP contribution value of each feature is obtained through "shap.TreeExplainer".

5. The method for designing a training set for an aircraft special situation training quality assessment model for a flight simulator according to claim 1, characterized in that: The SHAP value calculation process in step three is as follows: The goal of S1 and SHAP values ​​is to calculate the marginal contribution of each feature to the model's predicted value, calculated by the following formula: in, Let f(S) represent the SHAP value of feature i, N be the set of all features, S be a subset of features, and f(S) be the model's prediction output on the feature subset S. S2. For the decision tree model in the random forest model, Tree SHAP is used to calculate the marginal contribution of the feature at each node of the model based on the splitting conditions of the features on the path. S3. For each path in each decision tree model, Tree SHAP calculates the SHAP value of each feature using a recursive algorithm. For each node, assuming the probability of a decision path from the root node to a leaf node is p(z), and the contribution of each feature to the predicted value is v(z), Tree SHAP recursively assigns the SHAP value using the following formula. in, p represents the SHAP value of feature i in sample x within a single tree. ij It is the probability that sample x passes through feature i in path j. It is the size of the feature subset that has already been traversed in the path. It is the predicted value of feature i after it is added to feature subset z; S4. For each path, Tree SHAP from the root node to the leaf node needs to calculate the contribution of each feature to the output of the leaf node, and the predicted value of the tree. The values ​​of the leaf nodes are used, and the contribution of each path is calculated by the following formula. in, The summation of contributions from all paths within a single tree represents the SHAP contribution of feature i to sample x, where T represents the number of trees. is the weight of the feature subset z, representing the probability of that subset appearing on the path. It is the contribution increment after feature i is added; S5. In the entire random forest model, Tree SHAP is obtained by summing the SHAP values ​​of each decision tree using the following formula to obtain the final SHAP value. in, This represents the final SHAP value of feature i for sample x throughout the entire random forest. The SHAP value of feature i in the t-th decision tree.

6. The method for designing a training set for an aircraft special situation training quality assessment model for a flight simulator according to claim 1, characterized in that: Step four specifically involves: Step 1: Perform absolute value processing on the SHAP value calculated in Step 3, using... This represents the absolute value of the contribution of feature i to sample x; The second step is to calculate the absolute average of the SHAP values ​​of all samples using the following formula. Where n is the number of samples in the dataset. It is the absolute value of the SHAP value of the j-th sample on feature i; The third step is to normalize the average SHAP value of each feature, expressed by the following formula: Among them, w i It is the weight of feature i. is the absolute value of the average SHAP of feature i, d is the total number of features, and the sum of the weights of all features after normalization is equal to 1; Step 4: First, assume a feature matrix X of shape n×d, where n is the number of samples and d is the number of features. Each column of this matrix represents a feature, and each row represents a sample. X = [x1, x2, ..., xd]. Use x... i This represents the value of the i-th feature. For each feature x... i Use the calculated weight w i After weighted adjustment, the weighted feature matrix is ​​X. weighted =X×W, where W is a diagonal matrix, W=diag(w1, w2, ..., wd), and the weighted feature X is... weighted It is achieved by multiplying each feature column by its corresponding weight w. i The obtained, and weighted, feature is represented as X. weighted [:,i]=X[:,i]×w i .

7. The method for designing a training set for an aircraft special situation training quality assessment model for a flight simulator according to claim 1, characterized in that: In step six, determining the characteristic parameters for aircraft emergency training specifically means that if a certain parameter appears more than or equal to n / 2 times in the output results, then that parameter is determined as a characteristic parameter related to this type of aircraft emergency training subject.

Images