Electric power practitioner brain-computer interaction device channel screening method based on unsupervised clustering

Abstract

The invention relates to the field of electroencephalogram data acquisition and processing, in particular to an electric power practitioner brain-computer interaction device channel screening method based on unsupervised clustering, which comprises the following steps: multi-channel electroencephalogram information acquisition in an electric power working scene, multi-channel electroencephalogram signal noise elimination and multi-band signal extraction. And performing multi-channel clustering to obtain a clustered multi-channel multi-band signal, calculating the distance between the multi-detected multi-channel multi-band signal and the category cluster, formulating a selection standard, and outputting channel data and positions. According to the channel screening method based on unsupervised clustering, effective electroencephalogram channels can be accurately screened in an electric power working scene: on one hand, the number of electroencephalogram device channels is reduced, and the electroencephalogram signal redundancy is reduced; on the other hand, technical and theoretical support is provided for designing portable electroencephalogram information acquisition equipment, so that the objectivity of an electric power system on practitioner psychological safety literacy evaluation is improved, and finally the working risk of the electric power industry is reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of electroencephalogram data acquisition and processing, and particularly relates to a power practitioner brain-computer interaction device channel screening method based on unsupervised clustering. BACKGROUND

[0002] The power industry is a high-risk industry, and the psychological safety quality of practitioners directly affects operation safety and work efficiency, so it is necessary to monitor the psychological state at all times to reduce accidents and improve safety. However, there are obvious limitations at present: first, traditional practitioner psychological safety quality evaluation relies on questionnaires or observation, which is difficult to objectively reflect the psychological state; second, ordinary electroencephalogram acquisition devices contain multiple channel electrodes, and the data volume is huge, and the reaction to the characteristics of electroencephalogram information under the special working conditions of the power industry is not strong, so it cannot be directly used for actual work in the power industry, and therefore it is urgent to design a power practitioner electroencephalogram acquisition device channel screening method and system. SUMMARY

[0003] The purpose of the present application is to provide a power practitioner brain-computer interaction device channel screening method based on unsupervised clustering to solve the problems raised in the background art.

[0004] In order to solve the above technical problems, the present application provides the following technical solutions:

[0005] The power practitioner brain-computer interaction device channel screening method based on unsupervised clustering comprises the following steps:

[0006] In the working scene of power practitioners, a multi-channel electroencephalogram acquisition device is used to acquire the electroencephalogram information of the practitioners;

[0007] For the acquired multi-channel electroencephalogram signals, noise is removed by filtering processing to obtain multi-channel electroencephalogram information;

[0008] The multi-channel electroencephalogram information after noise removal is processed by a band-pass filter to extract five waveforms that can reflect human psychological emotions;

[0009] The multi-channel multi-band electroencephalogram information is clustered by an unsupervised clustering method to obtain the clustering centers of the same frequency band electroencephalogram signals under different channels;

[0010] The distances between the multi-channel multi-band electroencephalogram information and the corresponding clustering centers are calculated, and the distances of different bands of the same channel are accumulated and summed up;

[0011] A channel selection threshold T is set according to the distance accumulation result;

[0012] When the sum of the distances of a channel multi-band signal and the clustering centers is less than the threshold T, it is marked as a selected channel; otherwise, it is an unselected channel.

[0013] Further, the use of multi-channel EEG acquisition device to collect the brain electrical information of the practitioner, specifically, selecting electric power practitioners with different service length and different posts as the test objects, collecting the brain electrical information of the test objects in different working states of routine work and sudden task processing, the collection time covering the complete working process cycle, including work preparation, core operation, intermittent rest and task end stage, and recording the real-time working behavior and task type of the test objects synchronously during the collection process.

[0014] Further, the collected multi-channel EEG signal is filtered to remove noise, specifically, the collected multi-channel EEG signal is first baseline corrected, and the direct current drift component is removed by subtracting the mean value of each channel signal; then a notch filter is applied to suppress power frequency interference and its harmonics, and a high-pass filter is used to filter out low-frequency drift noise and retain effective EEG signals; then an independent component analysis algorithm is used to decompose the preprocessed multi-channel signal, and the physiological noise component is separated and identified by calculating the kurtosis and spectral characteristics of each independent component; after removing the noise component, the denoised EEG signal is reconstructed; finally, wavelet threshold denoising processing is performed on the reconstructed signal to suppress residual random noise.

[0015] Further, the denoised multi-channel EEG information is processed using a band-pass filter to extract five waveforms reflecting human psychological emotions; specifically, according to the typical waveband characteristics related to psychological emotions in the EEG signal, a Butterworth band-pass filter is used to extract the denoised multi-channel EEG information by waveband, extract five waveforms reflecting psychological emotions, including Delta wave, Theta wave, Alpha wave, Beta wave and Gamma wave, and for each waveband, the corresponding low-pass and high-pass cutoff frequencies are set to configure the second-order Butterworth band-pass filter corresponding to each waveband, and during filtering, zero-phase offset processing is used, and the multi-channel EEG signal is sequentially filtered, and five waveband signal sequences are output for each channel.

[0016] Further, the multi-channel and multi-waveband EEG information is clustered using an unsupervised clustering method to obtain the clustering center of the EEG signal in the same frequency band under different channels; specifically, the denoised multi-waveband EEG information is grouped according to the channel and frequency band, and five waveband signals associated with psychological emotions are extracted for each channel to form independent data subsets corresponding to single channels and single frequency bands; time domain and frequency domain features are extracted for each subset to construct feature vectors and perform Z-score standardization; the number of clusters is initially set according to the power demand, optimized by the elbow rule, and the similarity is calculated in units of information granules to generate the clustering center of each subset, and abnormal data is removed during clustering using the 3σ criterion, and the effect is verified by the silhouette coefficient after clustering, and the clustering center of each single channel and single frequency band is output when the result is reliable.

[0017] Further, the distance of the calculated multi-channel multi-band electroencephalogram information and the corresponding cluster center is calculated, and the distances of different bands of the same channel are accumulated and summed up; specifically, first, confirm the calculation object as a single-channel, single-band data subset and the corresponding cluster center, and confirm the feature vector of the effective electroencephalogram signal segment in the data subset; then, calculate the distance of each signal segment and the cluster center by using the Euclidean distance, take the average as the representative distance of this group, and re-calculate after verifying and removing the abnormal segments; finally, collect the representative distances of the five frequency bands according to the channel and accumulate the multi-band distance sum, and record the channel number synchronously.

[0018] Further, the channel selection threshold T is set according to the distance accumulation result, specifically, first, systemically arrange the multi-band distance sum data of all electroencephalogram channels, and form a complete data set by summarizing the distance accumulation results of different post subjects under the typical power scene, then remove the extreme distance sum value by using the 3σ criterion, and then build a threshold calculation model, first calculate the mean μ and the standard deviation σ of the distance sum of all channels after removing the abnormal values, then set the adjustment coefficient k, and preliminarily set the threshold T, the calculation formula of T is: T = μ-k×σ; take the preliminarily set T as the reference, filter out the channels with distance sum less than T, and count the number and corresponding brain region distribution thereof; then, under the same power scene, collect the electroencephalogram data of the same batch of subjects by using the filtered channels and all channels respectively, and compare the accuracy rates in the psychological state evaluation to determine the rationality of the threshold T, if the accuracy rate decreases beyond the standard or the stability is insufficient, then readjust the k value, repeat the calculation and verification steps, until the threshold T meets the dual goals of reducing the number of channels and controllable loss of evaluation accuracy.

[0019] Further, when the sum of the multi-band signals of a certain channel and the distance of the cluster center is less than the threshold T, it is marked as a selected channel, otherwise it is an unselected channel, specifically, first confirm the complete and accurate multi-band distance sum of the channel to be determined, and remove the abnormal values of the power scene; then compare with the preset threshold T, if less than the threshold T, preliminarily mark as a candidate selected channel, otherwise as an unselected channel; then verify according to the brain region function and the power scene, preferentially select the candidate channels of the key brain regions and confirm the cross-subject stability, and backtrack to check the collection quality and rule out accidental factors for a few of them; then count the number and position of the selected channels, and finally verify in practice, compare the evaluation accuracy and redundancy of the simplified and original equipment, and adjust and verify again if it does not meet the standard.

[0020] The application also discloses an electronic device, which comprises:

[0021] at least one processor; and

[0022] a memory connected with the at least one processor in communication; wherein

[0023] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the power practitioner brain-computer interaction device channel screening method based on unsupervised clustering described above.

[0024] Beneficial effects: The power practitioner brain-computer interaction device channel screening method based on unsupervised clustering of the present application can accurately screen effective electroencephalogram channels after multi-channel electroencephalogram information acquisition, denoising, multi-band extraction, clustering, distance calculation and threshold screening in the power working scene based on the unsupervised clustering channel screening method: on the one hand, it reduces the number of electroencephalogram device channels and reduces the redundancy of electroencephalogram signals; on the other hand, it provides technical and theoretical support for the design of portable electroencephalogram information acquisition devices, thereby improving the objectivity of the power system in assessing the psychological safety literacy of practitioners, and ultimately reducing the work risk in the power industry. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The flowchart of the power practitioner brain-computer interaction device channel screening method based on unsupervised clustering of the present application is shown in the figure. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0027] The present application provides a power practitioner brain-computer interaction device channel screening method based on unsupervised clustering, as shown in the figure, comprising the steps of: Figure 1

[0028] In the working scene of power practitioners, a multi-channel electroencephalogram acquisition device is used to acquire the electroencephalogram information of the practitioners.

[0029] For the acquired multi-channel electroencephalogram signals, filter processing is used to remove noise to obtain multi-channel electroencephalogram information.

[0030] A band-pass filter is used to process the denoised multi-channel electroencephalogram information to extract five waveforms that can reflect human psychological emotions.

[0031] An unsupervised clustering method is used to cluster the multi-channel multi-band electroencephalogram information to obtain the clustering centers of the same frequency band electroencephalogram signals under different channels.

[0032] The distances between the multi-channel multi-band electroencephalogram information and the corresponding clustering centers are calculated, and the distances of different wavebands of the same channel are accumulated and summed.

[0033] ​Set the channel selection threshold T based on the distance accumulation result;

[0034] When the sum of the distances between a channel's multi-band signal and the cluster center is less than a threshold T, it is marked as a selected channel; otherwise, it is marked as an unselected channel.

[0035] The proposed method involves using multi-channel EEG acquisition equipment to collect EEG information from power industry workers. Specifically, typical work scenarios for power industry workers, such as substation inspection, high-voltage equipment operation, and power dispatch monitoring, are selected to ensure coverage of different work states, including routine operations and handling emergencies. A 32-channel or 64-channel EEG acquisition device conforming to the international 10-20 system electrode layout is used, and the device must be resistant to electromagnetic interference to adapt to the power frequency interference environment in power scenarios. Power industry workers with different lengths of service and different positions (such as maintenance personnel, dispatchers, and repair workers) are selected as subjects. During the acquisition, subjects are required to wear the device in a natural working state, avoiding deliberate adjustments to posture or emotions. The acquisition duration must cover the entire workflow cycle, including the work preparation stage, core operation stage, intermittent rest stage, and task completion stage. The sampling frequency for each channel is set to 250Hz-1000Hz to ensure the capture of detailed features of the EEG signal. During the acquisition process, the subjects' real-time work behavior and task type are recorded simultaneously to provide scenario annotation basis for subsequent correlation analysis between EEG information and psychological state.

[0036] The acquired multi-channel EEG signals are filtered to remove noise and obtain multi-channel EEG information. Specifically, considering the susceptibility of EEG signals in power scenarios to 50Hz power frequency interference, equipment electromagnetic radiation, muscle activity (electromyography), and eye movement (electroophthalmos) noise, a hierarchical filtering strategy is employed to process the multi-channel EEG signals: First, baseline correction is performed by subtracting the mean of each channel signal to remove DC drift components and avoid baseline offset interference with subsequent analysis; second, a 50Hz notch filter (bandwidth set to ±1Hz) is applied to suppress power frequency interference and its harmonics, while a 0.5Hz high-pass filter is used simultaneously. Low-frequency drift noise was filtered out, retaining effective EEG signals above 0.5Hz. Then, Independent Component Analysis (ICA) was used to decompose the preprocessed multi-channel signal. By calculating the kurtosis value and spectral characteristics of each independent component, electrooculography (EOG) and electromyography (EMG) physiological noise components were identified and separated. After removing the independent components corresponding to the noise, the remaining components were reconstructed to obtain the denoised EEG signal. Finally, wavelet thresholding denoising was performed on the reconstructed signal. A 3-5 level decomposition was performed using the db4 wavelet basis, and a soft thresholding function was applied to the high-frequency wavelet coefficients to further suppress residual random noise. During processing, the time synchronization of the signals in each channel must be maintained to avoid phase differences between channels. This ensures that the denoised multi-channel EEG information effectively retains characteristic waveforms related to psychological emotions while reducing the influence of interference signals specific to the electrical scene, providing a high-quality data foundation for subsequent multi-band extraction.

[0037] The method involves using a bandpass filter to process the denoised multichannel EEG information and extracting five waveforms that reflect human psychological emotions.

[0038] Specifically, based on the typical waveband characteristics of EEG signals related to psychological emotions, a Butterworth bandpass filter is used to extract wavebands from the denoised multichannel EEG information, identifying five target waveforms and their corresponding frequency ranges: Delta waves (1-3Hz), mainly reflecting deep relaxation or fatigue; Theta waves (4-7Hz), related to inattention and mild anxiety; Alpha waves (8-13Hz), corresponding to a calm and relaxed state of wakefulness; Beta waves (14-30Hz), reflecting alertness, focus, or tension; and Gamma waves (31-50Hz), related to higher cognitive activities and emotional processing. A second-order Butterworth bandpass filter was designed for each band. The cutoff frequencies for Delta waves were set to 1Hz (low-pass) and 3Hz (high-pass), for Theta waves to 4Hz and 7Hz, for Alpha waves to 8Hz and 13Hz, for Beta waves to 14Hz and 30Hz, and for Gamma waves to 31Hz and 50Hz. Zero-phase shift processing was used during the filtering process to avoid signal distortion. The multi-channel EEG signals were filtered sequentially according to the above frequency ranges, with each channel outputting signal sequences for the five bands. The filtering effect was verified by calculating the power spectral density (PSD) of each band signal, ensuring that the energy of each band signal was concentrated within the target frequency range and that the signal-to-noise ratio (SNR) was not less than 10dB. The time axis consistency of the five band signals in each channel was maintained during the extraction process to provide a synchronous data basis for subsequent cluster analysis and distance calculation of different bands in the same channel, ensuring that the extracted multi-band information accurately reflects the psychological and emotional changes of power industry workers in their work environment.

[0039] The unsupervised clustering method is used to cluster multi-channel and multi-band EEG information separately to obtain the cluster centers of EEG signals in the same frequency band under different channels;

[0040] In practice, the EEG information after denoising and extraction of multiple bands is first grouped and processed according to channel and frequency band. Specifically, for each EEG acquisition channel (such as a single channel in a 32-channel or 64-channel device conforming to the international 10-20 system), the signals of the five bands of Delta wave (1-3Hz), Theta wave (4-7Hz), Alpha wave (8-13Hz), Beta wave (14-30Hz), and Gamma wave (31-50Hz) acquired in that channel are extracted to form an independent data subset of "single channel - single frequency band". This ensures that each data subset corresponds only to a specific psychological and emotional associated band of a specific channel, avoiding cross-interference between data from different channels or bands.

[0041] Subsequently, an information granular clustering method was selected to conduct unsupervised clustering. This method is suitable for the nonlinear and dynamic characteristics of EEG signals and can better capture the subtle features of EEG signals that change with task status (such as substation inspection and high-voltage equipment operation) during the work of power practitioners. Before clustering, feature extraction was performed on each "single channel - single frequency band" data subset. Time-domain features (signal mean, variance, kurtosis) and frequency-domain features (power spectral density peak value, center frequency) were selected to construct feature vectors. Then, Z-score standardization was used to eliminate the differences in the dimensions of different features, so that the mean of the feature vector is 0 and the standard deviation is 1, ensuring that the weights of each feature are balanced during the clustering process.

[0042] During the clustering process, the number of clusters was determined based on the actual needs of the power industry. Referring to common psychological states among power industry workers (calm and relaxed, focused and alert, mild anxiety, fatigue and burnout), the initial number of clusters was set to 4-6. This was then validated and optimized using the elbow rule—using the sum of squared errors within clusters (SSE) as an indicator. The optimal number of clusters was determined when the rate of decrease in SSE with increasing cluster number significantly slowed down. During the iteration process, information particles were used as the basic unit, and the similarity between each information particle was calculated (using Euclidean distance). Information particles with high similarity were grouped into the same class. This process was iterated until the intra-class similarity was maximized and the inter-class similarity was minimized. Finally, a cluster center was generated for each "single-channel - single-band" data subset. This cluster center was presented as a feature vector, accurately representing the typical characteristics of EEG signals for different psychological states under that channel and frequency band.

[0043] Simultaneously, abnormal data interference needs to be removed during the clustering process. For abnormal EEG signals caused by sudden equipment interference or instantaneous actions of subjects (such as raising a hand to adjust equipment) that may occur in the power scenario, the 3σ criterion is used to identify and remove them. The mean and standard deviation of the feature vectors of each data subset are calculated, and feature vectors that exceed the range of "mean ± 3 times standard deviation" are judged as outliers and removed to avoid abnormal data affecting the accuracy of cluster centers. After the clustering is completed, the clustering effect is verified by the silhouette coefficient. When the silhouette coefficient is in the range of 0.6-0.8, it indicates that the clustering result is highly reliable, and the output "single channel - single frequency band" cluster centers can be used for subsequent distance calculation.

[0044] The calculation involves determining the distance between multi-channel, multi-band EEG information and the corresponding cluster center, and then summing the distances of different bands within the same channel.

[0045] In practice, the calculation objects are first clearly defined as each "single channel - single frequency band" data subset and its corresponding cluster center: For each group of data previously grouped by "single channel - single frequency band" (such as Delta wave under channel C3 and Alpha wave under channel Fz) and obtained by information granular clustering, the feature vectors of all valid EEG signal segments in the group are first confirmed. That is, the vectors constructed by continuing the time-domain features (signal mean, variance, kurtosis) and frequency-domain features (power spectral density peak value, center frequency) extracted in the clustering stage, and the feature vectors have been standardized by Z-score (mean is 0, standard deviation is 1) to ensure that the feature vectors of the corresponding cluster centers are consistent in dimension and scale.

[0046] Subsequently, Euclidean distance was used to calculate the distance between the feature vectors of each signal segment within a single "single-channel-single-band" data subset and the feature vectors of the cluster centers of that group. For any EEG signal segment feature vector (denoted as X, containing 5 dimensions including time and frequency domains) under a certain "single-channel-single-band" data subset, and the feature vector of the cluster center of that group (denoted as C), the Euclidean distance between them was calculated. Specifically, the difference between the corresponding dimensions of the two vectors was calculated one by one, the squares of the differences in each dimension were summed, and the arithmetic square root of the sum was taken to obtain the distance value between the signal segment and the cluster center. At the same time, the average distance value of all valid signal segments under the "single-channel-single-band" data was taken as the representative distance corresponding to the "single-channel-single-band" data subset (to avoid the influence of random fluctuations of a single signal segment on the results). If there are abnormal signal segments in a certain "single-channel-single-band" data subset that were not completely eliminated by the previous 3σ criterion (such as distance values ​​far exceeding the average distance of the group 2), the distance between the signal segment and the cluster center is considered. If the distance is more than twice the normal value, the original EEG data of the segment needs to be re-verified to see if it was affected by sudden power interference (such as instantaneous electromagnetic pulses from the device). If an anomaly is confirmed, the segment should be removed and the average distance of the group should be recalculated to ensure the validity of the distance value.

[0047] After completing all "single channel - single frequency band" distance calculations, the distances are accumulated according to the "same channel" dimension. Taking a single EEG channel (such as Fp2, T7, Pz conforming to the international 10-20 system) as the unit, representative distances of the five frequency bands of Delta wave (1-3Hz), Theta wave (4-7Hz), Alpha wave (8-13Hz), Beta wave (14-30Hz), and Gamma wave (31-50Hz) that have been calculated under that channel are collected. These five distance values ​​are directly accumulated and summed (the distance weight of each frequency band is the same, as they are all key bands reflecting psychological emotions, and there is no priority difference), to obtain the "multi-band distance sum" of that channel. During the accumulation process, the channel number and corresponding sum value need to be recorded synchronously to ensure that each channel corresponds to only one sum result, and the calculation standards for the distance between each frequency band within the same channel (such as feature vector dimension, standardization method, and distance algorithm) are completely consistent. This avoids distortion of the sum value due to differences in calculation logic and provides accurate data support for setting the channel selection threshold T based on the distance sum and filtering effective channels.

[0048] The channel selection threshold T is set based on the distance accumulation result;

[0049] In practice, the process involves first systematically organizing the "multi-band distance summation" data from all EEG channels. This involves summarizing the distance accumulation results of each channel for subjects in different positions (maintenance personnel, dispatchers, and maintenance workers) under typical power scenarios (substation inspection, high-voltage equipment operation, and power dispatch monitoring) to form a complete dataset. Simultaneously, the 3σ criterion from previous data processing is used to calculate the mean and standard deviation of this dataset. Extreme distance summation values ​​exceeding the range of "mean ± 3 times standard deviation" are removed to avoid the impact of transient abnormal EEG signals (such as those caused by sudden interference) on the objectivity of the threshold setting.

[0050] Subsequently, based on the core needs of the power industry, the threshold setting guidelines were determined. In collaboration with power safety operation and maintenance experts and EEG technicians, a balance was established between equipment portability and the accuracy of psychological assessment. If priority is given to meeting the assessment accuracy requirements of high-risk scenarios (such as high-voltage equipment operation), it must be ensured that the channels retained after screening cover key brain regions related to emotion and attention (such as channels corresponding to the prefrontal and parietal lobes). If the focus is on equipment portability to adapt to mobile inspection scenarios, the screening criteria can be appropriately relaxed to reduce the number of channels. Based on this, a statistical method was used to construct a threshold calculation model: first, the mean μ and standard deviation σ of the sum of distances of all channels after outlier removal were calculated; then, an adjustment coefficient k (range 0.5-1.0) was set according to requirements, and the initial threshold T = μ - k × σ was set. The selection of the value of k needs to refer to the target number of channels (e.g., when a 32-channel device needs to be screened down to 8-12 channels, k = 0.8 can be determined through trial calculation), ensuring that the number of channels after screening meets the design requirements of lightweight equipment while retaining core information.

[0051] Next, the validity of the threshold was verified and iteratively optimized. Using the initially set T as a benchmark, channels with a total distance less than T were selected, and their number and corresponding brain region distribution were counted. Subsequently, under the same electrical scenario, EEG data were collected from the same group of subjects using both the selected channels and the full channels. The accuracy of the two in assessing psychological states (calm, focused, mild anxiety, fatigue) was compared. If the accuracy of the selected channels decreased by no more than 5% compared to the full channels, and the channel stability across subjects and scenarios (i.e., the probability of the same channel being selected in different tests) was not less than 80%, then the threshold T was deemed reasonable. If the accuracy decreased excessively or the stability was insufficient, the k value needed to be readjusted (if the accuracy was too low, the k value was reduced to lower T and retain more channels; if the stability was insufficient, the outlier removal was checked to see if it was thorough, or the weight of the scenario data was adjusted). The above calculation and verification steps were repeated until the threshold met the dual objectives of "reducing the number of channels (60%-70% less than the original number of channels) + controllable loss of assessment accuracy (≤5%)".

[0052] Furthermore, considering the differences in EEG characteristics among different power positions, it is necessary to verify the universality of the threshold. The optimal threshold for single job datasets for dispatchers and maintenance workers should be calculated separately, and the deviation between the threshold for each job and the universal threshold should be compared. If the deviation is less than 10%, the universal threshold T should be adopted to simplify equipment application. If the optimal threshold for a certain job (such as dispatchers, who are in a state of high concentration for a long time and have more significant EEG alpha wave characteristics) deviates significantly from the universal threshold, a job correction coefficient should be added on the basis of the universal threshold to ensure that the threshold can adapt to the EEG signal characteristics of major power positions, and finally form a channel selection threshold T that takes into account both universality and job adaptability.

[0053] When the sum of the distances between a channel's multi-band signal and the cluster center is less than a threshold T, it is marked as a selected channel; otherwise, it is marked as an unselected channel.

[0054] In practice, the first step is to verify the completeness and accuracy of the "multi-band distance summation" calculation result for the channel to be judged. This ensures that the summation is a valid accumulation of representative distances from the five frequency bands under that channel: Delta (1-3Hz), Theta (4-7Hz), Alpha (8-13Hz), Beta (14-30Hz), and Gamma (31-50Hz), with no missing bands or calculation logic deviations, and that abnormal distance values ​​caused by sudden power-related interference (such as instantaneous electromagnetic pulses to equipment) have been removed. Then, the total distance of the channel is compared with a preset threshold T. If the total distance is less than T, it is initially marked as a candidate channel; if it is greater than or equal to T, it is initially marked as an unselected channel.

[0055] After initial labeling, a secondary verification is required, taking into account the physiological functions of brain regions and the demands of electrical work scenarios. Priority should be given to retaining candidate selection channels located in key brain regions (such as the prefrontal Fp1 / Fp2 region, associated with emotion regulation; the parietal Pz / P3 region, associated with attention; and the temporal T3 / T4 region, associated with cognitive processing). Even if the total distance of some key brain region channels is close to T but still less than T, it is necessary to confirm their stability in cross-subject tests (different years of service and positions). If the channel meets the condition of a total distance less than T in the same work scenario (such as substation inspection and high-voltage equipment operation) in more than 80% of the subjects, it is officially labeled as a selection channel. If the condition is met only in a few subjects, the acquisition quality of the corresponding EEG signal of the channel needs to be retrospectively checked (such as whether the electrode contact is good and whether it is affected by electromyography / electroometry interference). Individual accidental factors should be excluded before determining the labeling results.

[0056] Meanwhile, after all channels are marked, the number and location distribution of selected channels are statistically analyzed to ensure that the number meets the design requirements of portable devices (e.g., 8-12 channels are retained after screening for 32-channel devices, and 15-20 channels are retained for 64-channel devices), and that the location covers the core brain regions related to emotion, attention, and cognition, so as to avoid situations where key functional brain regions have no selected channels.

[0057] Finally, the selected channels were applied to actual EEG acquisition verification. In the same power scenario, a simplified device with only selected channels was used to simultaneously collect EEG data from practitioners with the original multi-channel device. The accuracy of the two devices in assessing psychological states (calm, focused, mild anxiety, fatigue) was compared. If the accuracy of the simplified device decreased by no more than 5% compared to the original device, and the data redundancy was reduced by more than 60%, the labeling results were confirmed to be valid. If the accuracy decreased by more than 5%, the threshold T setting or the distance summation calculation process needed to be re-verified. After adjustment, the channel labeling and verification were carried out again until the selected channels could meet the objective requirements of assessing the psychological safety qualities of power practitioners and provide a clear and feasible technical basis for the electrode layout and hardware integration of the portable EEG acquisition device, adapting to the usage requirements of different work scenarios in the power industry (such as portability in inspection scenarios and assessment accuracy in high-voltage operation scenarios).

[0058] This application also provides an embodiment of an electronic device. The electronic device is manifested in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: one or more processors or processing units, memory, and buses connecting different components (including memory and processing units).

[0059] A bus refers to one or more of several bus architectures, including memory buses or memory controllers, peripheral buses, graphics acceleration ports, processors, or local buses using any of the various bus architectures. Examples of these architectures include, but are not limited to, Industry Standard Architecture (ISA) buses, Micro Channel Architecture (MCA) buses, Enhanced ISA buses, Video Electronics Standards Association (VESA) local buses, and Peripheral Component Interconnect (PCI) buses.

[0060] Electronic devices typically include a variety of computer-readable media. These media can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, and removable and non-removable media.

[0061] The memory may include computer-readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. Electronic devices may further include other removable / non-removable, volatile / non-volatile computer device storage media. By way of example only, the storage system may be used to read and write non-removable, non-volatile magnetic media.

[0062] The electronic device can also communicate with one or more external devices (e.g., keyboard, pointing device, camera, etc.), may include a display, and may communicate with one or more devices that enable a user to interact with the electronic device, and / or with any device that enables the electronic device to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via an input / output (I / O) interface. Furthermore, the electronic device can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN) and / or public networks, such as the Internet) via a network adapter. The network adapter communicates with other modules of the electronic device via a bus. The processor executes various functional applications and data processing by running programs stored in memory, such as implementing the channel screening method for brain-computer interface devices for power industry practitioners based on unsupervised clustering provided in the above embodiments of the present invention.

[0063] The method of this application will be further described below with reference to a specific embodiment:

[0064] In one specific embodiment, the method for screening brain-computer interface devices for power industry practitioners based on unsupervised clustering provided by the present invention includes the following steps:

[0065] Step 1: Multi-channel EEG information acquisition in power work scenarios;

[0066] In real-world power industry scenarios, practitioners wear multi-channel EEG acquisition devices to collect EEG information, which is expressed as... , This indicates the number of channels in the EEG acquisition device. This indicates the ID number of the practitioner being tested.

[0067] Step 2: Noise removal from multi-channel EEG signals;

[0068] Independent component analysis (ICA) is used to filter noise from multi-channel computer data. The expression is as follows: ,in This indicates that the number is Multichannel EEG data of the subjects after noise filtering.

[0069] Step 3: Multi-band signal extraction from EEG data;

[0070] The noise-removed multichannel EEG information was processed using a Butterworth bandpass filter to extract five waveforms from the multichannel EEG information of power industry workers during their work state. The expression is as follows: ,in, This indicates that the number is The subjects' multi-channel EEG information was processed through a Butterworth filter to obtain five types of EEG waveform signals. .

[0071] Step 4: Clustering of multi-channel, multi-band EEG information;

[0072] The information granular clustering method is used to cluster the numbers as follows: Five types of EEG waveforms from the subjects were clustered, and the cluster centers for several waveforms were obtained. The expression is as follows: ,in Indicates the number is Cluster centers for different waveforms of the subjects This refers to the information granular clustering method.

[0073] Step 5: Calculate the distance between the multi-subject, multi-channel, multi-band EEG information and the cluster center;

[0074] The expression for calculating and summing the Euclidean distances between multi-channel, multi-band EEG data from multiple subjects and their corresponding band cluster centers is as follows: .

[0075] Step Six: Establishing Channel Selection Criteria;

[0076] Based on the actual needs of the power industry, a selection threshold is set according to the cumulative distance results between the above-mentioned multi-channel different bands and the cluster center. .

[0077] Step 7: Determining the number and location of channels;

[0078] When the distance between a multi-band signal in a certain channel and the cluster center is less than a threshold The expression is A selected channel is marked as a time marker; otherwise, it is considered an unselected channel. This method reduces the number of channels in the EEG device while decreasing EEG signal redundancy. It lays the theoretical foundation and provides technical support for designing lightweight EEG information acquisition equipment for the power industry.

[0079] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for screening brain-computer interface devices for power industry practitioners based on unsupervised clustering, characterized in that, Includes the following steps: In the workplace of power industry workers, multi-channel EEG acquisition devices are used to collect the EEG information of the workers. The acquired multi-channel EEG signals were filtered to remove noise and obtain multi-channel EEG information. The noise-reduced multichannel EEG information was processed using a bandpass filter to extract five waveforms that reflect human psychological emotions. Unsupervised clustering method was used to cluster multi-channel and multi-band EEG information separately to obtain the cluster centers of EEG signals of the same frequency band under different channels; Calculate the distance between multi-channel, multi-band EEG information and the corresponding cluster centers, and sum the distances of different bands in the same channel. Set the channel selection threshold T based on the distance accumulation result; When the sum of the distances between a channel's multi-band signal and the cluster center is less than a threshold T, it is marked as a selected channel; otherwise, it is marked as an unselected channel.

2. The method for screening brain-computer interface devices for power industry practitioners based on unsupervised clustering according to claim 1, characterized in that, The method of using multi-channel EEG acquisition equipment to collect EEG information from practitioners involves selecting power industry practitioners with different lengths of service and different positions as subjects, collecting EEG information from subjects in different work states such as routine work and handling of sudden tasks, and collecting data for a duration covering the entire workflow cycle, including work preparation, core operations, intermittent rest and task completion stages. During the collection process, the subjects' real-time work behavior and task type are recorded simultaneously.

3. The method for screening brain-computer interface devices for power industry practitioners based on unsupervised clustering according to claim 1, characterized in that, The acquired multi-channel EEG signals are filtered to remove noise. Specifically, baseline correction is first performed on the acquired multi-channel EEG signals, and DC drift components are removed by subtracting the mean of each channel signal. Then, a notch filter is applied to suppress power frequency interference and its harmonics, while a high-pass filter is used to filter out low-frequency drift noise, retaining the effective EEG signal. Subsequently, the preprocessed multi-channel signal is decomposed using an independent component analysis algorithm. Physiological noise components are identified and separated by calculating the kurtosis and spectral characteristics of each independent component. After removing the noise components, the denoised EEG signal is reconstructed. Finally, wavelet threshold denoising is performed on the reconstructed signal to suppress residual random noise.

4. The method for screening brain-computer interface devices for power industry practitioners based on unsupervised clustering according to claim 1, characterized in that, The process involves using a bandpass filter to process the denoised multi-channel EEG information and extract five waveforms that reflect human psychological emotions. Specifically, based on the typical waveband characteristics related to psychological emotions in the EEG signal, a Butterworth bandpass filter is used to extract five waveforms reflecting psychological emotions: Delta wave, Theta wave, Alpha wave, Beta wave, and Gamma wave. For each waveband, corresponding low-pass and high-pass cutoff frequencies are set to configure a second-order Butterworth bandpass filter for each waveband. Zero-phase offset processing is used during filtering, and the multi-channel EEG signals are filtered sequentially, with each channel outputting a sequence of five waveband signals.

5. The method for screening brain-computer interface devices for power industry practitioners based on unsupervised clustering according to claim 1, characterized in that, The method employs unsupervised clustering to cluster multi-channel, multi-band EEG information separately, obtaining cluster centers for EEG signals in the same frequency band under different channels. Specifically, the denoised multi-band EEG information is grouped according to channel and frequency band, and five band signals associated with psychological emotions are extracted from each channel to form independent data subsets corresponding to single channels and single frequency bands. Time-domain and frequency-domain features are extracted from each subset, feature vectors are constructed, and Z-score standardization is performed. The number of clusters is initially set according to power demand, optimized using the elbow rule, and similarity clustering is calculated using information granules as units to generate cluster centers for each subset. During the clustering process, outlier data is removed using the 3σ criterion. After clustering, the effect is verified by the silhouette coefficient, and the cluster centers for each single channel and single frequency band are output when the results are reliable.

6. The method for screening brain-computer interface devices for power industry practitioners based on unsupervised clustering according to claim 1, characterized in that, The calculation of the distance between multi-channel, multi-band EEG information and the corresponding cluster center, and the summation of the distances of different bands in the same channel, involves: first, confirming that the calculation object is a single-channel, single-band data subset and its corresponding cluster center, and confirming the feature vector of the valid EEG signal segment within the data subset; then, calculating the distance between each signal segment and the cluster center using Euclidean distance, taking the average as the representative distance of the group, and recalculating after verifying and removing abnormal segments; finally, collecting the representative distances of the five frequency bands by channel and summing the multi-band distances, while simultaneously recording the channel number.

7. The method for screening brain-computer interface devices for power industry practitioners based on unsupervised clustering according to claim 1, characterized in that, The method for setting a channel selection threshold T based on the distance accumulation result involves first systematically organizing the multi-band distance summation data of all EEG channels, summarizing the distance accumulation results of subjects in different positions under typical power scenarios to form a complete dataset, using the 3σ criterion to remove extreme distance summation values, and then constructing a threshold calculation model. First, the mean μ and standard deviation σ of the total distance summation of all channels after outlier removal are calculated, and then an adjustment coefficient k is set to initially set the threshold T. The formula for calculating T is: T = μ - k × σ. Using the initially set T as a benchmark, channels with a total distance summation less than T are selected, and their number and corresponding brain region distribution are counted. Subsequently, under the same power scenario, EEG data are collected from the same group of subjects using both the selected channels and all channels. The accuracy of the two in psychological state assessment is compared to determine the rationality of the threshold T. If the accuracy drops excessively or the stability is insufficient, the value of k is readjusted, and the calculation and verification steps are repeated until the threshold T meets the dual objectives of reducing the number of channels and controlling the loss of assessment accuracy.

8. The method for screening brain-computer interface devices for power industry practitioners based on unsupervised clustering according to claim 1, characterized in that, When the sum of the distances between a channel's multi-band signals and the cluster center is less than a threshold T, it is marked as a selected channel; otherwise, it is marked as an unselected channel. Specifically, first, the completeness and accuracy of the sum of the multi-band distances of the channel to be judged are confirmed, and abnormal values ​​caused by sudden interference in the power scenario are eliminated; then, it is compared with a preset threshold T. If it is less than the threshold T, it is initially marked as a candidate selected channel, otherwise, it is marked as an unselected channel; then, based on brain region function and power scenario, a second verification is performed, prioritizing candidate channels in key brain regions and confirming cross-subject stability. If a few meet the requirements, the acquisition quality is checked back to eliminate accidental factors; then, the number and location of selected channels are counted, and finally, actual verification is performed. The accuracy and redundancy of the simplified and original equipment are compared, and if the standards are not met, adjustments are made and the verification is repeated.

9. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; characterized in that, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the channel screening method for brain-computer interface devices for power industry practitioners based on unsupervised clustering as described in claim 1.

Images