Hybrid method capable of simultaneously evaluating load state and load value

Abstract

The invention discloses a hybrid method capable of simultaneously evaluating a load state and a load value. The hybrid method comprises the following steps: (1) collecting electroencephalogram data and workload data of an air traffic controller under the conditions of resting, low load, medium load and overload; (2) extracting features and screening out the features with the best performance by using an mRMR algorithm; (3) data conversion: the discrete load labels are converted into specific numerical values; (4) simultaneously taking the load state and the load value as regression targets, and establishing a dual-target regression model; according to the method, the load state and the specific workload value of the controller are output at the same time, and a theoretical basis is provided for more accurately evaluating the load state and the load value of the controller.

Description

Technical Field

[0001] The present invention relates to the technical field of civil aviation airport operation, and in particular to a hybrid method capable of simultaneously evaluating a load state and a load value. Background Art

[0002] In recent years, due to the high load status, the operational errors of controllers caused by fatigue have increased year by year. Therefore, from the perspective of ensuring the safety and stability of civil aviation, it is of great significance to detect the load of controllers and keep them in an appropriate load status to reduce unsafe accidents.

[0003] The load detection model for air traffic controllers is usually a classification model or a regression model. The classification model can detect the load status of air traffic controllers based on the load label, while the regression model detects the load of air traffic controllers based on the specific value of the air traffic controller's workload. Among them, the classification model is too general and it is easy to make mistakes in the load status between two categories. The output value of the regression model is generally a specific value. Due to the difference in perception between individuals, it is easy to make mistakes in judging the load status of air traffic controllers based on this value. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a hybrid method that can simultaneously evaluate the load status and the load value, solve the problem of inconsistent perception of individual workload thresholds, and simultaneously evaluate the load status and the load value.

[0005] Technical solution: The hybrid method for simultaneously evaluating the load state and the load value of the present invention is characterized by comprising the following steps:

[0006] (1) Collect EEG data and workload data of air traffic controllers at rest, low load, medium load, and overload;

[0007] (2) Extract features and use the mRMR algorithm to select the best performing features;

[0008] (3) Data conversion: converting discrete load labels into specific values;

[0009] (4) Taking load status and load value as regression targets, a dual-target regression model is established.

[0010] Furthermore, in step (1), firstly, the experimental scenarios are designed according to the flow changes within the specified time: resting, low load, medium load, and overload; secondly, the EEG data of the controllers under different loads and the corresponding NASA-TXL scale data are obtained.

[0011] Furthermore, in step (2), the EEG data is preprocessed and wavelet transformed to obtain the values ​​corresponding to different EEG features of air traffic controllers in different scenarios, thereby completing the feature extraction work; using workload data as the target variable, the redundancy of features in different channels and the degree of correlation between features and targets are obtained, and the features that perform best are selected using the mRMR algorithm.

[0012] Furthermore, the specific steps of using the best features of the mRMR algorithm are as follows:

[0013] (21) Let S represent the EEG channel x i A collection of;

[0014] (22) Consider all channels as a whole and calculate the mutual information between all channels and the load c corresponding to a certain feature so that S satisfies the formula maxD(S,c), Obtain the maximum correlation D between the feature and the load;

[0015] (23) Calculate the redundant information between the feature and the channel, that is, calculate the sum of the redundancy R within the channel. The formula is as follows:

[0016] (24) Define Φ as the value of the final maximum correlation minus the minimum redundancy Φ: maxφ(D,R), φ=DR. Calculate the Φ of all features and sort them, evaluate the performance of the EEG features, and select the EEG features with the best performance.

[0017] Furthermore, in step (3), considering the discrete characteristics of the load category labels, data conversion processing is performed on them, and the threshold is perceived, so that the model can be used to simultaneously predict the discrete load state and the continuous load value. According to the load labels 0 load, low load, medium load, and overload, they are converted into values ​​0, 1, 2, and 3 respectively.

[0018] Furthermore, in step (4), after completing the data conversion, the leave-one-out-crossover method (LOOCV) is used to establish a multi-objective regression model. The specific steps are as follows:

[0019] (41) Input the characteristic value X and the target variable Y (specific value of workload), and the target variable Z (workload status value);

[0020] (42) Data partitioning: Divide the data sets corresponding to X, Y, and Z into training sets and test sets. The training set is used to build the random forest regression model, and the test set is used to evaluate the model performance. Among them, one sample in the original data set is used as the test set, and the remaining samples are used as the training set;

[0021] (43) Model construction: Use the training set data to train the random forest regression model. Each time, select one subject as the test sample and the remaining subjects as the training samples, and repeat this process until all samples have been tested once;

[0022] (44) Model performance evaluation: Use the test set data to evaluate the performance of the model. Use mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ) and other indicators are used to measure the detection accuracy of the model, and the model parameters are adjusted according to the indicator results to achieve model optimization.

[0023] After completing the model optimization, save the model and use it to detect the controller's load status and specific load values. By inputting relevant EEG feature data, the controller's load status and specific load values ​​can be output.

[0024] The hybrid system of the present invention capable of simultaneously evaluating load status and load value comprises:

[0025] Collection module: used to collect EEG data and workload data of air traffic controllers at rest, low load, medium load, and overload;

[0026] Extraction module: used to extract features and select the best performing features using the mRMR algorithm;

[0027] Data conversion module: used to convert discrete load labels into specific values;

[0028] Regression module: used to establish a dual-objective regression model with load status and load value as regression targets at the same time.

[0029] An electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, it implements any one of the hybrid methods described in one item for simultaneously evaluating load status and load value.

[0030] A storage medium described in the present invention stores a computer program, and when the computer program is executed by a processor, it implements any one of the hybrid methods for simultaneously evaluating the load state and the load value.

[0031] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (In view of the problem that the load status and specific values ​​of the controller's workload cannot be output simultaneously due to individual differences in the existing controller's cognitive workload assessment model, the mRMR algorithm is used to find the best performing features. On this basis, according to the discrete characteristics of the load category labels, data conversion processing is performed on them and a multi-objective regression model is established to achieve the simultaneous prediction of discrete load states and continuous load values ​​using the same model. This hybrid model not only solves the problem of inconsistent perception of individual load thresholds, but also can simultaneously output the controller's load status and specific values ​​of workload, providing a theoretical basis for more accurate assessment of the controller's load status and load values. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of the present invention;

[0033] Figure 2 It is the experimental site map of the present invention;

[0034] Figure 3 This is a diagram of the model results of the present invention. DETAILED DESCRIPTION

[0035] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0036] like Figure 1 As shown, an embodiment of the present invention provides a hybrid model that can simultaneously evaluate load status and load value, including the design of experimental scenarios, the collection of EEG data and NASA-TXL scale data, feature screening, data conversion, etc. The established model can simultaneously predict discrete load states and continuous load values, which can avoid the technical problem of inconsistent load threshold perception among individuals.

[0037] A hybrid model that can simultaneously evaluate the load state and load value. The empirical experiment was conducted in the Civil Aviation Student Control Simulation Laboratory of Nanjing University of Aeronautics and Astronautics. The mRMR algorithm was used to screen features and establish a hybrid model for evaluating the controller's load state and load value, including the following steps:

[0038] Step 1: First, design the experimental scenarios that meet the actual working conditions of the controllers according to the traffic changes: rest, low load, medium load, and overload. Secondly, obtain the EEG data of the controllers in each scenario and the corresponding NASA-TXL scale data.

[0039] Step 2: Preprocess and wavelet transform the EEG data to complete feature extraction. The specific feature selection is shown in the following table.

[0040] Feature 1 δ Feature 10 δ+β+θ+α+γ Feature 2 θ Feature 11 γ / θ Feature 3 α Feature 12 γ / α Feature 4 β Feature 14 γ / β Feature 5 δ+β+θ+α Feature 15 γ / (α+β) Feature 6 θ / β Feature 16 γ / (θ+β) Feature 7 α / β Feature 17 γ / (θ+α) Feature 8 (α+θ) / β Feature 18 γ / (δ+β+θ+α) Feature 9 (α+θ) / (α+β) Feature 19 γ / (δ+β+θ+α+γ) Feature 10 γ Feature 20 β+θ+α+γ

[0041] Take workload data as the target variable, find the redundancy of features in different channels and the correlation between features and targets, and use the best features of mRMR algorithm. The specific steps are as follows:

[0042] (1) Let S represent the EEG channel x i Collection

[0043] (2) Consider all channels as a whole and calculate the mutual information between all channels and the load c corresponding to a certain feature so that S satisfies the formula maxD(S,c), Obtain the maximum correlation D between the feature and the load;

[0044] (3) Calculate the redundant information between features and channels, that is, calculate the sum of the redundancy R within the channel. The formula is as follows:

[0045] The evaluation results are shown in the following table:

[0046]

[0047]

[0048] It can be seen from the above table that the feature β+θ+α+γ performs best. Therefore, the following hybrid model is established based on the channel data of the feature β+θ+α+γ.

[0049] Step three, perform data conversion processing on the load category, convert the category data into continuous data, and according to the load labels of the present invention, 0 load, low load, medium load, and overload are converted into numerical values ​​0, 1, 2, and 3 respectively. The larger the predicted numerical value, the higher the load level perceived by the threshold.

[0050] Step 4: Establishment of multi-target regression model. After completing data conversion, use the leave-one-out-crossover method (LOOCV) to establish a multi-target regression model. The specific steps are as follows:

[0051] (1) Input the characteristic value X and the target variable Y (specific value of workload) and the target variable Z (workload status value)

[0052] (2) Data partitioning: Divide the data sets corresponding to X, Y, and Z into training sets and test sets. The training set is used to build the random forest regression model, and the test set is used to evaluate the model performance. Among them, one sample in the original data set is used as the test set, and the remaining samples are used as the training set.

[0053] (3) Model construction: Use the training set data to train the random forest regression model. Each time, select one subject as the test sample and the remaining subjects as the training samples, and repeat this process until all samples have been tested once.

[0054] (4) Model performance evaluation: Use the test set data to evaluate the performance of the model. Use mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ) and other indicators are used to measure the detection accuracy of the model, and the model parameters are adjusted according to the indicator results to achieve model optimization.

[0055] (5) After the model optimization is completed, the model is saved and can be used to detect the load status and specific load value of the controller. By inputting the relevant EEG characteristic data, the load status and specific load value of the controller can be output.

[0056] Finally, the saved model can be used to evaluate the load status and specific load values. In this example, for our 41 subjects, the dimension of the specific value of the objective function Y workload is 100, the average MAE is 3.69, the average RMSE is 6.83, and the R 2 The mean is 0.85, the dimension of the objective function Z working state is 1, its MAE mean is 0.13, the RMSE mean is 0.24, R 2 The mean value is 0.91.

[0057] The embodiment of the present invention further provides a hybrid system capable of simultaneously evaluating a load state and a load value, comprising:

[0058] Collection module: used to collect EEG data and workload data of air traffic controllers at rest, low load, medium load, and overload;

[0059] Extraction module: used to extract features and select the best performing features using the mRMR algorithm;

[0060] Data conversion module: used to convert discrete load labels into specific values;

[0061] Regression module: used to establish a dual-objective regression model with load status and load value as regression targets at the same time.

[0062] An embodiment of the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the hybrid method for simultaneously evaluating the load state and the load value as described in any one of the items is implemented.

[0063] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements any one of the hybrid methods for simultaneously evaluating a load state and a load value.

Claims

1. A hybrid method for simultaneously evaluating load status and load value, characterized in that: The following steps are involved: (1) Collect EEG data and workload data of air traffic controllers at rest, low load, medium load, and overload; (2) Extract features and use the mRMR algorithm to select the best performing features; (3) Data conversion: converting discrete load labels into specific values; (4) Taking load status and load value as regression targets, a dual-target regression model is established.

2. A hybrid model capable of simultaneously evaluating load status and load value according to claim 1, characterized in that: In step (1), firstly, the experimental scenarios are designed according to the flow changes within the specified time: rest, low load, medium load, and overload; secondly, the EEG data of the controllers under different loads and the corresponding NASA-TXL scale data are obtained.

3. A hybrid model capable of simultaneously evaluating load status and load value according to claim 1, characterized in that: In step (2), the EEG data is preprocessed and wavelet transformed to obtain the values ​​corresponding to different EEG features of air traffic controllers in different scenarios, thus completing the feature extraction work. The workload data is used as the target variable to obtain the redundancy of features in different channels and the correlation between features and targets, and the best feature is obtained using the mRMR algorithm.

4. A hybrid model capable of simultaneously evaluating load status and load value according to claim 3, characterized in that: The specific steps for using the best features of the mRMR algorithm are as follows: (21) Let S represent the EEG channel x i A collection of; (22) Consider all channels as a whole and calculate the mutual information between all channels and the load c corresponding to a certain feature so that S satisfies the formula maxD(S,c), Obtain the maximum correlation D between the feature and the load; (23) Calculate the redundant information between the feature and the channel, that is, calculate the sum of the redundancy R within the channel. The formula is as follows: (24) Define Φ as the value of the final maximum correlation minus the minimum redundancy Φ: maxφ(D,R), φ=DR. Calculate the Φ of all features and sort them, evaluate the performance of the EEG features, and select the EEG features with the best performance.

5. A hybrid model capable of simultaneously evaluating load status and load value according to claim 1, characterized in that: In step (3), considering the discrete characteristics of the load category labels, data conversion processing is performed on them, the threshold is perceived, and the model is used to simultaneously predict the discrete load state and the continuous load value. According to the load labels 0 load, low load, medium load, and overload, they are converted into values ​​0, 1, 2, and 3 respectively.

6. A hybrid model capable of simultaneously evaluating load status and load value according to claim 1, characterized in that: In step (4), after completing the data conversion, the leave-one-out-crossover method (LOOCV) is used to establish a multi-objective regression model. The specific steps are as follows: (41) Input the characteristic value X and the target variable Y (specific value of workload), and the target variable Z (workload status value); (42) Data partitioning: Divide the data sets corresponding to X, Y, and Z into training sets and test sets. The training set is used to build the random forest regression model, and the test set is used to evaluate the model performance. Among them, one sample in the original data set is used as the test set, and the remaining samples are used as the training set; (43) Model construction: Use the training set data to train the random forest regression model. Each time, select one subject as the test sample and the remaining subjects as the training samples, and repeat this process until all samples have been tested once; (44) Model performance evaluation: Use the test set data to evaluate the performance of the model. Use mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ) and other indicators are used to measure the detection accuracy of the model, and the model parameters are adjusted according to the indicator results to achieve model optimization. After completing the model optimization, save the model and use it to detect the controller's load status and specific load values. By inputting relevant EEG feature data, the controller's load status and specific load values ​​can be output.

7. A hybrid system capable of simultaneously evaluating load status and load value, characterized in that: include: Collection module: used to collect EEG data and workload data of air traffic controllers at rest, low load, medium load, and overload; Extraction module: used to extract features and select the best performing features using the mRMR algorithm; Data conversion module: used to convert discrete load labels into specific values; Regression module: used to establish a dual-objective regression model with load status and load value as regression targets at the same time.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, a hybrid method capable of simultaneously evaluating a load state and a load value according to any one of claims 1 to 6 is implemented.

9. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a hybrid method capable of simultaneously evaluating a load state and a load value according to any one of claims 1 to 6 is implemented.

Images