CN-122020503-A - Electroencephalogram fatigue accurate detection method and equipment based on self-adaptive time sequence modeling

CN122020503ACN 122020503 ACN122020503 ACN 122020503ACN-122020503-A

Abstract

The invention relates to an electroencephalogram fatigue accurate detection method and equipment based on self-adaptive time sequence modeling, the method comprises the steps of collecting real-time working condition data and electroencephalogram data of a person to be detected, preprocessing, extracting multidimensional features of the electroencephalogram data to obtain a feature matrix, inputting the feature matrix and the real-time working condition data into a pre-established fatigue prediction model to obtain fatigue degree scores, wherein the process of obtaining the fatigue degree scores comprises the steps of carrying out space feature modeling and time relation modeling through a double-layer transducer encoder based on the feature matrix to obtain electroencephalogram space features and fatigue time features, taking the complexity factors of operation scenes obtained based on the real-time working condition data into consideration in the space feature modeling and the time relation modeling, and outputting the fatigue degree scores based on the fatigue time features by adopting the multiscale time sequence modeling. Compared with the prior art, the method and the device improve the accuracy and the effectiveness of feature extraction and the prediction stability under different working conditions.

Inventors

LI FAN
DU ZHAOXIN
FENG QIAN
TIAN YINGJIE
SU YUN
YANG XINGANG
SHI ZHIXIONG
PAN AIQIANG
ZHANG MENGYUAN
ZHOU DESHENG
ZHAO YINGYING

Assignees

国网上海市电力公司

Dates

Publication Date: 20260512
Application Date: 20251211

Claims (10)

1. An electroencephalogram fatigue accurate detection method based on self-adaptive time sequence modeling is characterized by comprising the following steps: Acquiring real-time working condition data and brain electrical data of a person to be tested, respectively preprocessing the data, and acquiring a working scene complexity factor from the real-time working condition data; the fatigue prediction model comprises a double-layer transducer encoder, wherein the process of obtaining the fatigue degree score comprises the steps of carrying out space feature modeling and time relation modeling through the double-layer transducer encoder based on the feature matrix to obtain fatigue time features, considering operation scene complexity factors obtained based on real-time working condition data in the space feature modeling and the time relation modeling, carrying out fatigue state prediction by adopting multi-scale time sequence modeling based on the fatigue time features, and outputting the fatigue degree score.
2. The accurate detection method of brain electrical fatigue based on adaptive time series modeling of claim 1, wherein the preprocessing of brain electrical data of the person under test comprises: band-pass filtering is carried out on the original electroencephalogram signals; and removing artifacts in the electroencephalogram signals by adopting independent component analysis, wherein the artifacts comprise electrooculogram and myoelectricity.
3. The method for accurately detecting the brain electrical fatigue based on the adaptive time sequence modeling according to claim 1, wherein the real-time working condition data comprises: the fault correlation class is equipment fault alarming times and fault grades in preset time; operation related class, namely scheduling instruction issuing frequency and operation difficulty coefficient in preset time; Environmental related classes, namely noise level and illumination intensity of the operation site; And equipment related class, namely responsible for regional equipment running state.
4. The method for accurately detecting the brain electrical fatigue based on the adaptive time sequence modeling according to claim 3, wherein the preprocessing of the real-time working condition data comprises the following steps: aligning the real-time working condition data with an acquisition time stamp of brain electricity data of a tested person, cleaning abnormal values, and filling an average value for missing data; Adopting Min-Max standardization to standardize numerical data in the real-time working condition data; Adopting ordered coding and standardization to code and convert the classified data in the real-time working condition data into values of [0,1] intervals; If the normalized/encoded real-time working condition data do not fall in the [0,1] interval, the excess part is forcibly cut off; And calculating sub factors of each type of data in the real-time working condition data, and obtaining a working scene complexity factor by weighting, summing and fusing the sub factors according to a preset weight, wherein the weight is set based on the priority of the power working risk.
5. The method for accurately detecting the brain electrical fatigue based on the adaptive time sequence modeling according to claim 1, wherein the process of extracting the multidimensional feature comprises the following steps: dividing the preprocessed continuous electroencephalogram signals into time windows with fixed lengths; calculating differential entropy characteristics of different frequency bands for each time window; and constructing a feature matrix based on differential entropy features of different frequency bands.
6. The method for accurately detecting the brain electrical fatigue based on the adaptive time sequence modeling according to claim 1, wherein the process of performing spatial feature modeling based on the double-layer transducer encoder comprises the following steps: reconstructing the feature matrix into an input format suitable for spatial modeling; based on the reconstructed feature matrix, respectively generating a query matrix, a key matrix and a value matrix through three preset learnable weight matrices; calculating attention weight matrixes among different channels by taking the channels as sequence units; calculating a channel adjustment coefficient based on the complexity factor of the operation scene, and dynamically scaling the attention weight matrix based on the channel adjustment coefficient; Substituting the dynamically adjusted weight matrix into Softmax for calculation, and outputting spatial features; and inputting the spatial characteristics into a feedforward network, and obtaining the electroencephalogram spatial characteristics by combination layer normalization after linear transformation and activation function processing.
7. The method for accurately detecting the brain electrical fatigue based on the adaptive time sequence modeling according to claim 6, wherein the process of modeling the time relationship based on the double-layer transducer encoder comprises the following steps: Reconstructing the electroencephalogram spatial features obtained after the spatial feature modeling into an input format matched with time modeling; calculating a window adjusting coefficient based on the operation scene complexity factor, adjusting the coverage range of a dynamic window based on the window adjusting coefficient, and dynamically generating a window mask matrix; based on the reconstructed brain electricity space characteristics and the window mask matrix, adopting a sliding window attention mechanism, and simultaneously capturing brain electricity short-time nerve oscillation and long-time fatigue accumulation processes; The capturing result of the attention mechanism of the sliding window is input into a feedforward network, and after linear transformation and activation function processing, fatigue time characteristics are obtained by combination layer normalization.
8. The method for accurately detecting the brain electrical fatigue based on the adaptive time series modeling according to claim 1, wherein the process of predicting the fatigue state by adopting the multi-scale time series modeling and obtaining the fatigue degree score comprises the following steps: Based on a plurality of convolution layers constructed in a grading manner with different scales, based on fatigue time characteristics, respectively capturing instantaneous neural oscillation, medium-term fatigue accumulation and long-term fatigue trend, and acquiring short-term characteristics, medium-term characteristics and long-term characteristics; The short-term characteristics, the medium-term characteristics and the long-term characteristics are subjected to multi-scale characteristic fusion through splicing operation, and multi-scale characteristics are obtained; and inputting the multi-scale features into a linear layer for dimension mapping, normalizing the dimension mapping result to the [0,1] range through a Sigmoid function, and obtaining the fatigue degree score.
9. The method for accurately detecting brain electrical fatigue based on adaptive time series modeling according to claim 1, wherein the fatigue prediction model is trained based on a time series regularization loss function, the time series regularization loss function comprising regression loss and smoothing loss.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that, The steps of the method for accurately detecting the brain fatigue based on adaptive time series modeling according to any one of claims 1 to 9 are realized when the processor executes the computer program.

Description

Electroencephalogram fatigue accurate detection method and equipment based on self-adaptive time sequence modeling Technical Field The invention relates to the technical field of fatigue detection, in particular to an electroencephalogram fatigue accurate detection method and equipment based on self-adaptive time sequence modeling. Background The safe and stable operation of the power system is highly dependent on continuous concentration and accurate decision-making of post personnel such as scheduling, operation and maintenance. However, the high-load and long-term working characteristics are extremely easy to induce cognitive fatigue, so that the risk of misoperation is remarkably increased, and a great potential safety hazard is formed. The objective monitoring of fatigue state by using physiological signals such as electroencephalogram (EEG) has become a research hotspot in this field. The conventional fatigue monitoring method mostly belongs to the category of abnormality detection, and aims to identify the fatigue state which has occurred. This mode is essentially a hysteresis response in that the cognitive ability of the operator may have been reduced to dangerous levels when the system is alerted, missing the best opportunity for intervention. The neurophysiologic process of fatigue is extremely complex, with EEG signals exhibiting closely coupled dynamic changes in the spatial dimension (functional synergy of different brain regions), the temporal dimension (dynamic change of fatigue state), and the frequency dimension (change of different brain electrical rhythms). The traditional method often separates the dimensions, which affects the recognition accuracy. In a real-world scenario, fatigue is a continuous gradual process from awake to drowsy. However, conventional methods reduce fatigue to a two-classification problem of "normal" and "fatigue. This coarse-grained binary judgment cannot reflect which specific fatigue development stage the operator is in. Chinese patent CN118177835A discloses a fatigue degree prediction method and a system based on an electroencephalogram signal, and aims at solving the problems that the existing EEG-based fatigue monitoring technology has large individual difference, so that a new individual needs long-time calibration and has poor real-time performance. However, the fatigue state judgment is not actually performed in a continuous gradual change process from waking to drowsiness, and meanwhile, the fatigue degree prediction method is not specific, and the detection precision is not high under the complex power system working scene of high load and long time. Disclosure of Invention The invention aims to overcome the defects of the prior art and provide an electroencephalogram fatigue accurate detection method and equipment based on self-adaptive time sequence modeling, which can accurately predict the fatigue state according to electroencephalogram signals of electric operators and give feedback in real time, thereby solving the defects of the traditional information processing network model in predictability, modeling completeness and assessment fineness, improving the accuracy and effectiveness of feature extraction and making up the defects of traditional deep learning in fatigue detection. The aim of the invention can be achieved by the following technical scheme: An electroencephalogram fatigue accurate detection method based on adaptive time sequence modeling, the method comprising: Acquiring real-time working condition data and brain electrical data of a person to be tested, respectively preprocessing the data, and acquiring a working scene complexity factor from the real-time working condition data; the fatigue prediction model comprises a double-layer transducer encoder, wherein the process of obtaining the fatigue degree score comprises the steps of carrying out space feature modeling and time relation modeling through the double-layer transducer encoder based on the feature matrix to obtain fatigue time features, considering operation scene complexity factors obtained based on real-time working condition data in the space feature modeling and the time relation modeling, carrying out fatigue state prediction by adopting multi-scale time sequence modeling based on the fatigue time features, and outputting the fatigue degree score. Further, preprocessing of the electroencephalogram data of the person under test includes: band-pass filtering is carried out on the original electroencephalogram signals; and removing artifacts in the electroencephalogram signals by adopting independent component analysis, wherein the artifacts comprise electrooculogram and myoelectricity. Further, the real-time operating condition data includes: the fault correlation class is equipment fault alarming times and fault grades in preset time; operation related class, namely scheduling instruction issuing frequency and operation difficulty coefficient in preset time; Environmenta