CN-122004764-A - Method for estimating physiological data-driven cognitive state of plateau under hypoxia in real time

CN122004764ACN 122004764 ACN122004764 ACN 122004764ACN-122004764-A

Abstract

The invention discloses a physiological data driven cognitive state real-time estimation method under a plateau hypoxia, which belongs to the technical field of physiological signal processing and comprises the following steps of S1 obtaining physiological signals of operators under the plateau hypoxia environment, preprocessing the physiological signals to obtain physiological signal time sequence data, S2 inputting the physiological signal time sequence data into a task-aware mixed expert model framework, S3 gating network generating corresponding expert weight distribution through a self-attention mechanism according to the input physiological signal time sequence data, S4 carrying out parallel feature extraction on the input physiological signal time sequence data by a plurality of heterogeneous expert networks, S5 carrying out weighted summation on the outputs of the heterogeneous expert networks according to the expert weight distribution, and outputting estimation results of reaction time and accuracy of various cognitive dimensions. By adopting the method, the real-time accurate estimation of the multidimensional cognitive state of the operator is realized under extreme environments such as highland hypoxia.

Inventors

ZHANG WAN
CAO YONGTONG
CHEN JING
Yang Baiyan

Assignees

中日友好医院(中日友好临床医学研究所)

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (9)

1. A physiological data driven cognitive state real-time estimation method under highland hypoxia is characterized by comprising the following steps: S1, acquiring physiological signals of operators in a highland hypoxia environment, and preprocessing the physiological signals to obtain physiological signal time sequence data; S2, inputting physiological signal time sequence data into a task-aware mixed expert model framework, wherein the mixed expert model framework comprises a gate control network and a plurality of heterogeneous expert networks; S3, generating corresponding expert weight distribution through a self-attention mechanism according to the input physiological signal time sequence data by utilizing a gating network in the mixed expert model framework; S4, extracting parallel characteristics of the input physiological signal time sequence data by utilizing a plurality of heterogeneous expert networks in the mixed expert model framework; and S5, carrying out weighted summation on the outputs of the heterogeneous expert networks according to expert weight distribution, and outputting estimation results of reaction time and accuracy of various cognitive dimensions.
2. The method for real-time estimation of physiological data-driven cognitive state under altitude hypoxia according to claim 1, wherein in step S1, the preprocessing specifically comprises: The physiological signal data of different time points are aligned with the cognitive test time points by adopting a linear interpolation method, so as to obtain physiological signal time sequence data; abnormal samples with heart rate less than 30bpm or blood oxygen saturation less than 60% are removed.
3. The method for real-time estimation of physiological data driven cognitive state in highland hypoxia according to claim 2, wherein the heterogeneous expert network comprises a transducer expert network, a CNN-LSTM hybrid expert network and a residual MLP expert network, each expert network being equipped with an independent output layer.
4. The method for real-time estimation of physiological data-driven cognitive state under altitude and hypoxia according to claim 3, wherein in step S3, the process of generating expert weight distribution by the gating network comprises the following steps: Applying a self-attention mechanism to the input physiological signal time series data to obtain a global context weighted feature representation; compressing the feature representation by the multi-layer perceptron into a logits vector of dimension E, wherein E represents the number of expert networks; And processing logits vectors by using a softmax function to generate weight coefficients of each expert network.
5. The method for real-time estimation of physiological data driven cognitive state under altitude hypoxia according to claim 4, wherein the method for constructing the transducer expert network comprises the following steps: for time series of input physiological signals Obtaining a query, a key and a value matrix through linear projection: ; ; ; Wherein, the Representing the query matrix and, Representing a trainable query projection weight matrix, Representing a matrix of keys and, Representing a matrix of trainable key projection weights, A matrix of representative values is represented, Representing a trainable value projection weight matrix.
6. The method for real-time estimation of physiological data-driven cognitive state under highland hypoxia according to claim 5, wherein the construction method of the CNN-LSTM hybrid expert network comprises the following steps: First sliding in the time dimension using a one-dimensional convolution layer to detect local features; the final hidden states of the forward LSTM and the backward LSTM are then stitched using BiLSTM for feature extraction based on history and future context.
7. The method for real-time estimation of physiological data driven cognitive state under altitude and hypoxia according to claim 6, wherein the mapping function implemented by each expert network is as follows: ; Wherein, the Representing the time series of the input physiological signal, Representing the parameters of the expert network, Representing the length of the time series; The mapping function implemented by the gating network is: ; Wherein, the Representing a trainable set of parameters for the gating network, Representing the number of expert networks.
8. The method for real-time estimation of physiological data driven cognitive state under altitude hypoxia according to claim 7, wherein in step S5, the formula for weighted summation of the outputs of the heterogeneous expert networks according to the expert weight distribution is: ; Wherein, the Representing the final predicted output of the model, Corresponding first representing the output of the gating network The weight of the individual expert network, Represents the first And the output of the personal expert network.
9. The method for real-time estimation of physiological data driven cognitive state under altitude hypoxia according to claim 8, wherein the total loss function of the task perception hybrid expert model framework is: ; Wherein, the Representing the total number of training samples, Representing the index of the sample, Represents the first In the case of a true reaction of the individual samples, Represents the first In the case of a predictive reaction of a single sample, Represents the first The true accuracy of the individual samples is that, Represents the first The prediction accuracy of the individual samples is determined, Representing the regularization coefficient of L2, Representing a set of trainable parameters for all expert and gating networks.

Description

Method for estimating physiological data-driven cognitive state of plateau under hypoxia in real time Technical Field The invention relates to the technical field of physiological signal processing, in particular to a method for estimating a cognitive state driven by physiological data under highland hypoxia in real time. Background The highland hypoxia environment forms a serious threat to the cognitive function of a human body, and as the altitude increases, the atmospheric pressure and the oxygen partial pressure decrease, so that the blood oxygen saturation of the human body decreases, and a series of negative changes of the physiological and cognitive functions are initiated. Proved by demonstration researches, the core cognitive functions of the individual such as situational memory, working memory, inhibition control and the like are obviously damaged at the altitude of more than 2500 meters. The central nervous system dysfunction caused by hypoxia directly or indirectly is particularly characterized by distraction, slow information processing speed and reduced judging and decision making capability, is extremely easy to cause human errors in safety key fields such as altitude operation and the like, and forms a serious challenge for operation safety and efficiency. In recent years, with the development of wearable sensing technology, real-time dynamic assessment of cognitive states by using continuously and noninvasively acquired physiological signals (such as heart rate and blood oxygen) has become a very potential technical direction. In early researches, classical machine learning algorithms such as linear regression, support vector regression and random forest are mostly adopted, however, the limitation is that the feature engineering is highly dependent on expert knowledge, and complex interaction modes with high dimension, nonlinearity and time sequence dynamics in physiological signals are difficult to fully capture. For automatic learning feature representation, researchers turn to deep learning, cyclic neural networks and their variant long-short term memory networks are widely used because of their time-series modeling capabilities, and single architecture models exhibit performance beyond that of traditional methods, but their capability boundaries are obvious, and when facing the high heterogeneity of physiological-cognitive relationships (different tasks, different individuals, different stress phases) in extreme environments, a ubiquitous single network structure often has difficulty in achieving optimal performance. In addition, existing studies focus on estimating the state of a single cognitive task (e.g., fatigue, attention), lacking a multi-task learning framework for simultaneous prediction and modeling for multiple cognitive dimensions (e.g., attention, working memory, processing speed, etc.), which limits the ability of the model to comprehensively evaluate the cognitive state of a person in a complex work environment. Disclosure of Invention The invention aims to provide a physiological data driven cognitive state real-time estimation method under plateau hypoxia, which realizes real-time accurate estimation of a multidimensional cognitive state of an operator under extreme environments such as plateau hypoxia. In order to achieve the above purpose, the invention provides a physiological data driven cognitive state real-time estimation method under highland hypoxia, which comprises the following steps: S1, acquiring physiological signals of operators in a highland hypoxia environment, and preprocessing the physiological signals to obtain physiological signal time sequence data; S2, inputting physiological signal time sequence data into a task-aware mixed expert model framework, wherein the mixed expert model framework comprises a gate control network and a plurality of heterogeneous expert networks; S3, generating corresponding expert weight distribution through a self-attention mechanism according to the input physiological signal time sequence data by utilizing a gating network in the mixed expert model framework; S4, extracting parallel characteristics of the input physiological signal time sequence data by utilizing a plurality of heterogeneous expert networks in the mixed expert model framework; and S5, carrying out weighted summation on the outputs of the heterogeneous expert networks according to expert weight distribution, and outputting estimation results of reaction time and accuracy of various cognitive dimensions. Preferably, in step S1, the preprocessing specifically includes: The physiological signal data of different time points are aligned with the cognitive test time points by adopting a linear interpolation method, so as to obtain physiological signal time sequence data; abnormal samples with heart rate less than 30bpm or blood oxygen saturation less than 60% are removed. Preferably, the heterogeneous expert networks include a transform expert network, a CNN-