A real-time detection method and system for controller workload based on few-channel electroencephalogram signals

Abstract

The present invention discloses a method for real-time workload detection of air traffic controllers based on few-channel electroencephalogram (EEG) signals, comprising the following steps: (1) obtaining EEG data and NASA‑TXL scale data of air traffic controllers in working state; (2) screening out indicators that are more sensitive to workload; (3) determining channel combinations that are more sensitive to workload changes based on sensitive indicators; (4) establishing a real-time workload detection model for air traffic controllers; (5) outputting the workload of air traffic controllers in real time. The present invention establishes a real-time workload detection model for air traffic controllers, greatly shortens the time consumption of workload evaluation, and improves the practicability of using EEG equipment to evaluate the workload of air traffic controllers.

Description

Technical Field

[0001] The present invention relates to the field of civil aviation airport operation safety technology, and in particular to a controller workload real-time detection method and system based on few-channel electroencephalogram signals. Background Art

[0002] In the past decade, my country's civil aviation industry has been in a high-speed development momentum, and air traffic has become increasingly busy. In this case, the increase in traffic has directly increased the workload of controllers, and the difficulty of control has also increased day by day, resulting in a gradual increase in accidents caused by controllers' operational errors and human errors. The impact of human factors on aviation safety is mainly manifested in the mistakes and omissions of controllers in the process of aircraft command, unreasonable attention allocation, and overload work. As the core of air traffic control, the good cognitive process and appropriate workload of controllers are the basis for ensuring the safety of aircraft operations in the sector. Based on this, it is very important to detect the workload of controllers.

[0003] In the existing methods for detecting the workload of traffic controllers, electrocardiogram signals, eye and facial information, and EEG signals can reflect the real-time status of the traffic controller and provide the basis for implementing the assessment. Among these methods, EEG signals have the highest accuracy and are known as the "golden indicator" for workload assessment. However, due to the particularity of EEG signals, this method still has some key problems in detection speed and practicality. For example, the 64-channel EEG system used in the present invention is a relatively complex device that requires well-trained technicians to position the electrodes. At the same time, the process of applying EEG paste to the 64 electrodes one by one is also very time-consuming, which seriously affects the efficiency of workload detection. At present, technical researchers are exploring how to reduce the number of EEG channels while retaining accuracy to improve the applicability of EEG equipment in detecting workload. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method and system for real-time detection of controller workload based on few-channel EEG signals, which uses an improved mRMR algorithm to screen out the highest accuracy indicators among 59 EEG power spectrum indicators, and on the basis of determining the indicators, screen out the channel combination that is most sensitive to workload, and then establish a real-time detection model for air traffic controller workload, which greatly shortens the time spent on workload assessment and improves the practicability of EEG equipment in assessing controller workload.

[0005] Technical solution: The method for real-time detection of controller workload based on few-channel EEG signals of the present invention comprises the following steps:

[0006] (1) Obtain EEG data and NASA-TXL scale data of air traffic controllers during work;

[0007] (2) Screen out indicators that are more sensitive to workload;

[0008] (3) Determine the channel combination that is more sensitive to workload changes based on sensitive indicators;

[0009] (4) Establish a real-time monitoring model for controller workload;

[0010] (5) Real-time output of controller workload.

[0011] Furthermore, the step (1) is specifically as follows: obtaining EEG data and NASA-TXL scale data of air traffic controllers in working state, and preprocessing the EEG data, including removing useless electrodes, re-referencing, filtering, and ICA; secondly, using frequency domain analysis method to calculate the values ​​of different indicators of each subject in different lead channels;

[0012] In frequency domain analysis, the wavelet transform of a continuous function x(t) is defined as in, is the mother wavelet, a and b are the scale and position of the wavelet respectively; through the translation and expansion of the mother wavelet, the wavelet transform has an abnormal time-frequency positioning; the EEG index data include:

[0013] d α / β γ / (θ+β) θ+a d+b+c i (a+θ) / b γ / (θ+α) θ+γ b+i+a α (a+θ) / (a+β) γ / (θ+β+α) β+α θ+α+γ β c c / (d+b+i+a) a+c b+i+c d+b+i+a d+b+i+a+c c / (d+b+the+a+c) b+c b+a+c d / (d+b+i+a) c / d d+i d+i+a d+b+a+c θ / (δ+β+θ+α) c / i d+a d+b+i d+b+i+c a / (d+b+i+a) c / a d+b d+i+g d+i+a+c β / (δ+β+θ+α) b / w d+c d+b+a b+i+a+c w / w c / (a+b) β+θ d+a+c

[0014] Furthermore, the step (2) is specifically as follows: first, determine the index set based on the EEG data, that is, which EEG power spectrum indicators can be used for real-time detection of controller workload; second, screen out indicators that are more sensitive to workload based on the improved mRMR algorithm: let S represent the set of channels. After removing 5 useless channels, 59 channels remain. First, calculate the mutual information between all channels (features) and workload (category), so that S satisfies Secondly, calculate the sum of the internal redundancy of the channel R, the calculation formula is Finally, the value of the maximum correlation-minimum redundancy Φ of all indicators is obtained and sorted to find the best indicator. The calculation formula is maxφ(D,R), φ=DR; among them, the indicator most sensitive to the controller workload is β+θ+α+γ;

[0015] Furthermore, the step (3) is specifically as follows: based on the screened sensitive indicators, the mRMR algorithm is used to screen the channels that are more sensitive to the workload; wherein, 59 channels are used as feature subsets, workload is used as category input, and the output is a channel set that has the greatest correlation with the workload but the smallest correlation between channels.

[0016] Furthermore, the step (4) comprises the following steps:

[0017] (41) Data preparation: Organize data related to controller workload, including input indicator value X and target variable Y workload;

[0018] (42) Data division: The EEG index data and workload data are divided into a training set and a test set; the training set is used to build the model, and the test set is used to evaluate the model performance;

[0019] (43) Model construction: Use the training set data to train the decision tree regression model; select one subject as the test sample each time, and the remaining subjects as the training samples, and repeat this process until all samples have been tested once;

[0020] (44) Model evaluation: Use the test set data to evaluate the performance of the model; use classification accuracy and recall ratio indicators to measure the prediction accuracy of the model;

[0021] (45) Parameter adjustment: According to the evaluation results, the parameters of the decision tree regression model are adjusted.

[0022] The present invention provides a real-time detection system for controller workload based on a few-channel electroencephalogram signal, comprising:

[0023] Data acquisition module: used to obtain EEG data and NASA-TXL scale data of air traffic controllers in working state;

[0024] Sensitive indicator module: used to filter out indicators that are more sensitive to workload;

[0025] Sensitive channel combination module: used to determine the channel combination that is more sensitive to workload changes based on sensitive indicators;

[0026] Detection module: used to establish a real-time detection model for controller workload;

[0027] Output module: used to output controller workload in real time.

[0028] Furthermore, the data acquisition module is as follows: the EEG data and NASA-TXL scale data of air traffic controllers in working state are obtained, and the EEG data are preprocessed, including the removal of useless electrodes, re-reference, filtering, and ICA; secondly, the frequency domain analysis method is adopted to calculate the values ​​of different indicators of each subject in different lead channels;

[0029] In frequency domain analysis, the wavelet transform of a continuous function x(t) is defined as in, is the mother wavelet, a and b are the scale and position of the wavelet respectively; through the translation and expansion of the mother wavelet, the wavelet transform has an abnormal time-frequency positioning; the EEG index data include:

[0030]

[0031]

[0032] Furthermore, the sensitive indicator module is specifically as follows: first, determine the indicator set based on the EEG data, that is, which EEG power spectrum indicators can be used for real-time detection of controller workload; second, screen out indicators that are more sensitive to workload based on the improved mRMR algorithm: let S represent the set of channels. After removing 5 useless channels, 59 channels remain. First, calculate the mutual information between all channels (features) and workload (category), so that S satisfies Secondly, calculate the sum of the internal redundancy of the channel R, the calculation formula is Finally, the value of the maximum correlation-minimum redundancy Φ of all indicators is obtained and sorted to find the best indicator. The calculation formula is maxφ(D,R), φ=DR; among them, the indicator most sensitive to the controller workload is β+θ+α+γ;

[0033] Furthermore, the sensitive channel combination module is specifically as follows: based on the screened sensitive indicators, the mRMR algorithm is used to screen channels that are more sensitive to workload; among them, 59 channels are used as feature subsets, workload is used as category input, and the output is the channel set with the greatest correlation with workload but the smallest correlation between channels.

[0034] Furthermore, the detection module includes the following steps:

[0035] (41) Data preparation: Organize data related to controller workload, including input indicator value X and target variable Y workload;

[0036] (42) Data division: The EEG index data and workload data are divided into a training set and a test set; the training set is used to build the model, and the test set is used to evaluate the model performance;

[0037] (43) Model construction: Use the training set data to train the decision tree regression model; select one subject as the test sample each time, and the remaining subjects as the training samples, and repeat this process until all samples have been tested once;

[0038] (44) Model evaluation: Use the test set data to evaluate the performance of the model; use classification accuracy and recall ratio indicators to measure the prediction accuracy of the model;

[0039] (45) Parameter adjustment: According to the evaluation results, the parameters of the decision tree regression model are adjusted.

[0040] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: The present invention evaluates the EEG power spectrum index to obtain the highest accuracy index, and on this basis, uses the mRMR algorithm to find the optimal channel combination corresponding to the index, and establishes a fast real-time detection model for the controller's workload. It improves the practicality and speed of evaluating workload with a few-channel EEG signals, and provides a theoretical basis for scientifically evaluating the workload of air traffic controllers. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of the present invention;

[0042] Figure 2 It is the key flow chart of the present invention;

[0043] Figure 3 This is a scene diagram of the simulated control experiment of the present invention. DETAILED DESCRIPTION

[0044] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0045] like Figure 1-2 As shown, an embodiment of the present invention provides a method for real-time detection of controller workload based on a few-channel electroencephalogram signal, comprising the following steps:

[0046] (1) Obtaining EEG data and NASA-TXL scale data of air traffic controllers under working conditions; specifically, obtaining EEG data and NASA-TXL scale data of air traffic controllers under working conditions, and preprocessing the EEG data, including removing useless electrodes, re-referencing, filtering, and ICA; secondly, using frequency domain analysis to calculate the values ​​of different indicators of each subject in different lead channels;

[0047] In frequency domain analysis, the wavelet transform of a continuous function x(t) is defined as in, is the mother wavelet, a and b are the scale and position of the wavelet respectively; through the translation and expansion of the mother wavelet, the wavelet transform has an abnormal time-frequency positioning; the EEG index data include:

[0048] d α / β γ / (θ+β) θ+a d+b+c i (a+θ) / b γ / (θ+α) θ+γ b+i+a α (a+θ) / (a+β) γ / (θ+β+α) β+α θ+α+γ β c c / (d+b+i+a) a+c b+i+c d+b+i+a d+b+i+a+c c / (d+b+the+a+c) b+c b+a+c d / (d+b+i+a) c / d d+i d+i+a d+b+a+c θ / (δ+β+θ+α) c / i d+a d+b+i d+b+i+c a / (d+b+i+a) c / a d+b d+i+g d+i+a+c β / (δ+β+θ+α) b / w d+c d+b+a b+i+a+c w / w c / (a+b) β+θ d+a+c

[0049] (2) Screen out indicators that are more sensitive to workload; specifically as follows: Traditional workload detection for traffic controllers generally includes characteristic analysis of four wave bands: α wave, β wave, θ wave, and δ wave, while ignoring the role of γ wave in mental state analysis. γ wave is a manifestation of neuronal excitement and should be used for workload detection for traffic controllers. Therefore, in order to better analyze the mental state of traffic controllers in different scenarios, the present invention first considers common traditional indicators: α absolute energy, β absolute energy, β / (δ+β+θ+α), (α+θ) / β, θ / β, α / β, etc. On this basis, new indicators related to γ ​​wave are proposed, such as γ absolute energy, γ / (δ+β+θ+α), etc. The specific indicators are selected as shown in the following table.

[0050] d α / β γ / (θ+β) θ+a d+b+c i (a+θ) / b γ / (θ+α) θ+γ b+i+a α (a+θ) / (a+β) γ / (θ+β+α) β+α θ+α+γ β c c / (d+b+i+a) a+c b+i+c d+b+i+a d+b+i+a+c c / (d+b+the+a+c) b+c b+a+c d / (d+b+i+a) c / d d+i d+i+a d+b+a+c θ / (δ+β+θ+α) c / i d+a d+b+i d+b+i+c a / (d+b+i+a) c / a d+b d+i+g d+i+a+c β / (δ+β+θ+α) b / w d+c d+b+a b+i+a+c w / w c / (a+b) β+θ d+a+c

[0051] Based on the improved mRMR algorithm, we can select indicators that are more sensitive to workload. Find the correlation between each indicator and workload. Further, according to the formula Calculate the redundancy between the 59 channels of each indicator, and finally calculate the value of the maximum correlation - minimum redundancy Φ of all indicators according to the formula maxφ(D,R),φ=DR, and sort them to find the best indicator. The best indicator found in this example is β+θ+α+γ

[0052] (3) Determine the channel combination that is more sensitive to workload changes based on sensitive indicators; specifically, use the mRMR algorithm to screen channels that are more sensitive to workload. The algorithm is written using the Python pymrmr library, takes 59 channels as feature subsets, and workload as category input, and outputs the channel set with the greatest correlation with workload but the smallest correlation between channels. According to the algorithm output results, the optimal channel combination is channels 36, 38, and 41.

[0053] (4) Establishing a real-time detection model for controller workload; including the following steps:

[0054] (41) Data preparation: Organize data related to controller workload, including input indicator value X and target variable Y workload;

[0055] (42) Data division: The EEG index data and workload data are divided into a training set and a test set; the training set is used to build the model, and the test set is used to evaluate the model performance;

[0056] (43) Model construction: Use the training set data to train the decision tree regression model; select one subject as the test sample each time, and the remaining subjects as the training samples, and repeat this process until all samples have been tested once;

[0057] (44) Model evaluation: Use the test set data to evaluate the performance of the model; use classification accuracy and recall ratio indicators to measure the prediction accuracy of the model;

[0058] (45) Parameter adjustment: According to the evaluation results, the parameters of the decision tree regression model are adjusted.

[0059] (5) Real-time output of controller workload.

[0060] like Figure 3 As shown in the figure, in this example, the Recall, Precision, F1-score, and ROC AUC of the full channel model are 0.959183673, 0.98, 0.98989899, ​​and 0.95, respectively, while the Recall, Precision, F1-score, and ROC AUC of the best channel combination model are 0.971236559, 0.986515576, 0.99277481, and 0.976752114, respectively. By comparison, the channel screening and modeling method in this paper is effective, which retains accuracy while reducing channels, and achieves high-precision and high-efficiency evaluation of workloads.

[0061] The embodiment of the present invention also provides a real-time detection system for controller workload based on a few-channel electroencephalogram signal, comprising:

[0062] Data acquisition module: used to obtain EEG data and NASA-TXL scale data of air traffic controllers in working state; specifically as follows: obtain EEG data and NASA-TXL scale data of air traffic controllers in working state, pre-process the EEG data, including removing useless electrodes, re-referencing, filtering, and ICA; secondly, use frequency domain analysis method to calculate the values ​​of different indicators of each subject in different lead channels;

[0063] In frequency domain analysis, the wavelet transform of a continuous function x(t) is defined as in, is the mother wavelet, a and b are the scale and position of the wavelet respectively; through the translation and expansion of the mother wavelet, the wavelet transform has an abnormal time-frequency positioning; the EEG index data include:

[0064] d α / β γ / (θ+β) θ+a d+b+c i (a+θ) / b γ / (θ+α) θ+γ b+i+a α (a+θ) / (a+β) γ / (θ+β+α) β+α θ+α+γ β c c / (d+b+i+a) a+c b+i+c d+b+i+a d+b+i+a+c c / (d+b+the+a+c) b+c b+a+c d / (d+b+i+a) c / d d+i d+i+a d+b+a+c θ / (δ+β+θ+α) c / i d+a d+b+i d+b+i+c a / (d+b+i+a) c / a d+b d+i+g d+i+a+c β / (δ+β+θ+α) b / w d+c d+b+a b+i+a+c w / w c / (a+b) β+θ d+a+c

[0065] Sensitive indicator module: used to screen out indicators that are more sensitive to workload; the details are as follows: first, determine the indicator set based on the EEG data, that is, which EEG power spectrum indicators can be used for real-time detection of controller workload; second, screen out indicators that are more sensitive to workload based on the improved mRMR algorithm: let S represent the set of channels. After removing 5 useless channels, there are 59 channels left. First, calculate the mutual information between all channels (features) and workload (category), so that S satisfies Secondly, calculate the sum of the internal redundancy of the channel R, the calculation formula is Finally, the value of the maximum correlation-minimum redundancy Φ of all indicators is obtained and sorted to find the best indicator. The calculation formula is maxφ(D,R), φ=DR; among them, the indicator most sensitive to the controller workload is β+θ+α+γ;

[0066] Sensitive channel combination module: used to determine the channel combination that is more sensitive to workload changes based on sensitive indicators; specifically as follows: based on the screened sensitive indicators, the mRMR algorithm is used to screen the channels that are more sensitive to workload; among them, 59 channels are used as feature subsets, workload is used as category input, and the output is the channel set with the greatest correlation with workload but the smallest correlation between channels.

[0067] Detection module: used to establish a real-time detection model for controller workload; including the following steps:

[0068] (41) Data preparation: Organize data related to controller workload, including input indicator value X and target variable Y workload;

[0069] (42) Data division: The EEG index data and workload data are divided into a training set and a test set; the training set is used to build the model, and the test set is used to evaluate the model performance;

[0070] (43) Model construction: Use the training set data to train the decision tree regression model; select one subject as the test sample each time, and the remaining subjects as the training samples, and repeat this process until all samples have been tested once;

[0071] (44) Model evaluation: Use the test set data to evaluate the performance of the model; use classification accuracy and recall ratio indicators to measure the prediction accuracy of the model;

[0072] (45) Parameter adjustment: According to the evaluation results, the parameters of the decision tree regression model are adjusted.

[0073] Output module: used to output controller workload in real time.

Claims

1. A real-time detection method for controller workload based on few-channel EEG signals, characterized in that: The following steps are involved: (1) Obtaining EEG data and NASA-TXL scale data of air traffic controllers in working state; specifically as follows: Obtaining EEG data and NASA-TXL scale data of air traffic controllers in working state, and preprocessing the EEG data, including removing useless electrodes, re-referencing, filtering, and ICA; Secondly, the frequency domain analysis method is used to calculate the values ​​of different indicators of each subject in different lead channels; In frequency domain analysis, the wavelet transform of a continuous function x(t) is defined as in, is the mother wavelet, a and b are the scale and position of the wavelet respectively; through the translation and expansion of the mother wavelet, the wavelet transform has an abnormal time-frequency positioning; the EEG index data include: (2) Screen out indicators that are more sensitive to workload; the details are as follows: first, determine the indicator set based on the EEG data, that is, which EEG power spectrum indicators are used for real-time detection of controller workload; second, screen out indicators that are more sensitive to workload based on the improved mRMR algorithm: let S represent the set of channels. After removing 5 useless channels, 59 channels remain. First, calculate the mutual information between all channels and workload, so that S satisfies Secondly, calculate the sum of the internal redundancy of the channel R, the calculation formula is Finally, the value of the maximum correlation-minimum redundancy Φ of all indicators is calculated and sorted to find the best indicator. The calculation formula is maxφ(D,R), φ=DR; among them, the indicator most sensitive to the controller workload is β+θ+α+γ; (3) Determine the channel combination that is more sensitive to workload changes based on the sensitive indicators; specifically, based on the selected sensitive indicators, use the mRMR algorithm to select the channels that are more sensitive to workload; wherein, 59 channels are used as feature subsets, workload is used as category input, and the output is the channel set that has the greatest correlation with workload but the smallest correlation between channels; (4) Establishing a real-time detection model for controller workload; including the following steps: (41) Data preparation: Organize data related to controller workload, including input indicator value X and target variable Y workload; (42) Data division: The EEG index data and workload data are divided into a training set and a test set; the training set is used to build the model, and the test set is used to evaluate the model performance; (43) Model construction: Use the training set data to train the decision tree regression model; select one subject as the test sample each time, and the remaining subjects as the training samples, and repeat this process until all samples have been tested once; (44) Model evaluation: Use the test set data to evaluate the performance of the model; use classification accuracy and recall ratio indicators to measure the prediction accuracy of the model; (45) Parameter adjustment: According to the evaluation results, the parameters of the decision tree regression model are adjusted; (5) Real-time output of controller workload.

Images