A method and system for monitoring fatigue state

Abstract

The application relates to the technical field of fatigue state monitoring and early warning, and discloses a fatigue state monitoring method and system, which comprises the following steps: obtaining a target signal of a target user, identifying a calm anchor point in the signal, reconstructing the signal to obtain a first signal; comparing a real-time target signal of the target user with the first signal in a high-dimensional space, identifying a first area in the signal; analyzing fatigue causes in combination with real-time environmental data, constructing a fatigue cause-and-effect diagram; matching the first area in the fatigue cause-and-effect diagram, simulating the influence of changes of each node on the first area, analyzing the causes of fatigue, obtaining a fatigue evaluation result, and monitoring the fatigue state; the application provides a signal benchmark for a fatigue analysis process by identifying a calm anchor point to reconstruct a signal, reduces the influence of individual physiological differences on the fatigue analysis process, improves the accuracy of the fatigue evaluation result, accurately quantitatively analyzes the fatigue causes, and improves the individual applicability of the fatigue analysis result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fatigue state monitoring and early warning, and more particularly to a fatigue state monitoring method and system. BACKGROUND

[0002] At present, heart rate variability, as a key indicator reflecting the function of the autonomic nervous system, has been widely used in fatigue monitoring. The existing technology usually evaluates fatigue state directly based on the time domain, frequency domain or nonlinear features of heart rate variability by setting a fixed threshold or applying a machine learning model. However, these methods mostly rely on general models or population benchmarks, failing to fully consider individual physiological differences and environmental dynamic changes, resulting in inaccurate evaluation results. In addition, the existing technology focuses more on the hierarchical evaluation of fatigue degree, lacking in-depth analysis of the root cause of fatigue, leading to insufficient fatigue analysis.

[0003] The existing technology has the following problems: using a unified threshold or standard model, without considering the interference of physiological state data of users by noise and emotional fluctuations during the analysis process, resulting in analysis results that cannot reflect the actual state of users; only analyzing the fatigue level, lacking in-depth analysis of the causes of fatigue, and unable to provide effective intervention measures; based on historical data or a single model, unable to respond to environmental changes in real time, resulting in inaccurate evaluation results; to solve at least one of the above problems, the present application proposes a fatigue state monitoring method and system. SUMMARY

[0004] In view of the deficiencies of the existing technology, the purpose of the present application is to provide a fatigue state monitoring method and system that can effectively solve the problems in the background art. The specific technical solution of the present application is as follows:

[0005] A fatigue state monitoring method, comprising:

[0006] According to the pre-acquired target signal of the target user, a state recognition model is used to recognize the calm anchor point in the signal, and a first signal is obtained by reconstructing the signal;

[0007] The real-time target signal of the target user is mapped to a high-dimensional space for comparison with the first signal, and a first region in the signal is identified;

[0008] Combining real-time environmental data to analyze the causes of fatigue, taking the causes of fatigue as nodes, establishing directed edges between the nodes according to the causal relationship between the causes of fatigue, and constructing a fatigue causal graph;

[0009] Matching the first region in the fatigue causal graph, simulating the influence of changes in each node on the first region through a pre-set fatigue analysis model, analyzing the causes of fatigue, and obtaining a fatigue evaluation result to monitor the fatigue state.

[0010] Specifically, the pre-acquired target signal of the target user is reconstructed into the first signal according to the calm anchor points in the signal by a preset state recognition model, including:

[0011] According to the pre-acquired target signal of the target user, the target signal is segmented in time sequence to obtain a plurality of signal segments;

[0012] For each signal segment, the corresponding calm anchor points in the signal are identified by a preset state recognition model;

[0013] According to the calm anchor points, the reference signal is decoupled from each signal segment, and the reference signal is spliced and reconstructed to obtain the first signal.

[0014] Specifically, for each signal segment, the corresponding calm anchor points in the signal are identified by a preset state recognition model, including:

[0015] For each signal segment, the signal features are extracted by a preset state recognition model, the similarity between each signal feature vector is calculated, and a feature similarity matrix is constructed;

[0016] The feature map is generated based on the feature similarity matrix, the feature distribution in the feature map is analyzed, and the connected subgraph is identified;

[0017] The point with the most stable feature distribution in each connected subgraph is identified as the corresponding calm anchor point.

[0018] Specifically, according to the calm anchor points, the reference signal is decoupled from each signal segment, and the reference signal is spliced and reconstructed to obtain the first signal, including:

[0019] According to the calm anchor points, the reference signal is decoupled from each signal segment by a preset signal decoupling model;

[0020] The features of the reference signal are extracted, the features are clustered, and the reference signal is spliced and reconstructed according to the cluster center to obtain the first signal.

[0021] Specifically, the real-time target signal of the target user and the first signal are mapped to a high-dimensional space for comparison to identify a first region in the signal, including:

[0022] The real-time target signal of the target user and the first signal are mapped to a high-dimensional space, the attractor trajectory is reconstructed, and the real-time attractor and the reference attractor are obtained;

[0023] The real-time attractor and the reference attractor are respectively subjected to modal decomposition, the corresponding dynamic modes are extracted, and the real-time mode set and the reference mode set are obtained;

[0024] Identify a first region in the signal based on the real-time modal set and the benchmark modal set.

[0025] Specifically, the first region in the signal is identified based on the real-time modal set and the benchmark modal set, including:

[0026] Based on the benchmark modal set, a benchmark basis is constructed.

[0027] Project the real-time modal set onto the benchmark basis, calculate the projection residual, analyze the feature difference between the real-time modal set and the benchmark modal set, and construct a difference feature vector.

[0028] According to the difference feature vector, a first region in the signal is identified.

[0029] Specifically, the fatigue reasons are analyzed in combination with real-time environmental data, the fatigue reasons are taken as nodes, directed edges are established between nodes according to the causal relationship between fatigue reasons, a fatigue causal graph is constructed, including:

[0030] The fatigue reasons are analyzed in combination with real-time environmental data, and the causal relationship between the fatigue reasons is analyzed to construct a causal feature matrix.

[0031] Take the fatigue reasons as nodes, and establish directed edges between nodes according to the causal feature matrix to construct a fatigue causal graph.

[0032] Specifically, the first region is matched in the fatigue causal graph, the influence of changes in each node in the fatigue analysis model on the first region is simulated, the reasons for fatigue are analyzed, and a fatigue evaluation result is obtained, including:

[0033] Match the first region in the fatigue causal graph to identify the corresponding reason nodes and obtain a candidate reason node set.

[0034] Simulate the influence of changes in each node in the candidate reason node set on the first region through the preset fatigue analysis model, analyze the reasons for fatigue, and obtain a fatigue evaluation result.

[0035] Specifically, the influence of changes in each node in the candidate reason node set on the first region is simulated through the preset fatigue analysis model, the reasons for fatigue are analyzed, and a fatigue evaluation result is obtained, including:

[0036] Simulate the influence of changes in each node in the candidate reason node set on the first region through the preset fatigue analysis model to obtain a simulation result.

[0037] According to the simulation result, the influence of a single node and node combination is analyzed respectively, and the fatigue contribution value of each node is calculated.

[0038] According to the fatigue contribution value, the reasons for fatigue are analyzed to obtain a fatigue evaluation result.

[0039] A fatigue state monitoring system for implementing the fatigue state monitoring method, comprising:

[0040] A signal analysis module, according to the pre-acquired target signal of the target user, identifies the calm anchor point in the signal through a preset state recognition model, and reconstructs the first signal according to the signal;

[0041] A first region identification module, maps the real-time target signal of the target user to the high-dimensional space for comparison with the first signal, and identifies the first region in the signal;

[0042] A fatigue causal diagram construction module, analyzes fatigue causes in combination with real-time environmental data, takes the fatigue causes as nodes, establishes directed edges between the nodes according to the causal relationship between the fatigue causes, and constructs a fatigue causal diagram;

[0043] A fatigue analysis module, matches the first region in the fatigue causal diagram, simulates the influence of the change of each node on the first region through a preset fatigue analysis model, analyzes the causes of fatigue, obtains a fatigue evaluation result, and monitors the fatigue state.

[0044] The application has the following beneficial effects: the physiological signal is processed in segments through the state recognition model, the calm anchor point is identified, the reference signal is decoupled for signal reconstruction, the real-time target signal is mapped to the high-dimensional space with the first signal, the feature difference is extracted by using the attractor trajectory and dynamic modal decomposition, the first region is located, the fatigue causal diagram is constructed in combination with the real-time environmental data, the influence of the change of the node on the first region is simulated, the fatigue contribution value of each node is quantified, the corresponding fatigue causes are analyzed, and the fatigue state evaluation is realized; the signal reference is provided for the fatigue analysis process by identifying the calm anchor point to reconstruct the signal, the influence of individual physiological differences on the fatigue analysis process is reduced, the accuracy of the fatigue evaluation result is improved, the fatigue causes can be accurately quantitatively analyzed by constructing the fatigue causal diagram, a reference is provided for fatigue intervention, and the environmental applicability and individual applicability of the fatigue analysis result are improved. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 A workflow diagram of the fatigue state monitoring method in embodiment 1 of the application;

[0046] Figure 2 A schematic diagram of the feature map in embodiment 1 of the application;

[0047] Figure 3 A schematic diagram of the fatigue causal diagram in embodiment 1 of the application;

[0048] Figure 4 A structural schematic diagram of the fatigue state monitoring system in embodiment 1 of the application. DETAILED DESCRIPTION

[0049] The application will be further described below in conjunction with the accompanying drawings and embodiments.

[0050] In the embodiments of the present application, the word "exemplary" or "for example" is used to mean serving as an example, instance, or illustration. Any embodiment or design presented as "exemplary" or "for example" in the embodiments of the present application is not necessarily to be construed as preferred or advantageous over other embodiments or design. In fact, a variety of implementations of the embodiments of the present application are possible, and each of the embodiments or design presented as "exemplary" or "for example" can be implemented in a wide variety of contexts. Therefore, the wording "exemplary" or "for example" is used merely for the purpose of presentation of relevant concepts.

[0051] Hereinafter, the terms "first", "second", and the like are used generically and are only intended for the purpose of description, and should not be construed as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the embodiments of the present application, the meaning of "a plurality of" is two or more, unless otherwise specified.

[0052] Embodiment 1:

[0053] Reference Figure 1 As shown in the specific implementation of the fatigue state monitoring method of the present application, comprising:

[0054] S101, according to the pre-acquired target signal of the target user, the calm anchor point in the signal is recognized by the pre-set state recognition model, and the first signal is obtained by reconstructing the signal;

[0055] S102, mapping the real-time target signal of the target user to the first signal to the high-dimensional space for comparison, and identifying the first area in the signal;

[0056] S103, combining real-time environmental data to analyze the fatigue cause, taking the fatigue cause as a node, establishing a directed edge between the nodes according to the causal relationship between the fatigue causes, and constructing a fatigue causal diagram;

[0057] S104, matching the first area in the fatigue causal diagram, simulating the influence of the change of each node on the first area through the pre-set fatigue analysis model, analyzing the cause of fatigue, obtaining the fatigue evaluation result, and monitoring the fatigue state.

[0058] In the workplace, fatigue is an important factor leading to decreased work efficiency and frequent accidents; in the sports field, excessive fatigue can cause sports injuries; in the medical and health care field, fatigue is a common symptom of chronic diseases. The embodiment can help users to understand their own fatigue state in real time and identify the cause of fatigue, so as to take targeted intervention measures and prevent health risks.

[0059] In the embodiment, the target signal is the physiological signal corresponding to the target user. According to the pre-acquired physiological signal of the target user, an individualized reference signal representing the health state is extracted. The heart rate variability signal presents a specific regularity pattern when the user is in a calm state, which can be used as a reference for the health state. Through a pre-set state recognition model, the moment when the physiological indicators in the signal are most stable and the interference is least is recognized as a calm anchor point, and a first signal that is not affected by temporary factors is reconstructed. By recognizing the calm anchor point and reconstructing the first signal, a corresponding first signal reference can be developed for each user, which is suitable for individuals with large differences in physiological characteristics and avoids the influence of individual physiological differences on the fatigue evaluation process, thereby improving the accuracy of the fatigue evaluation result. By recognizing the calm anchor point, noise and abnormal fluctuations in the signal can be filtered out, ensuring the stability and representativeness of the reference signal. The reconstructed first signal can reflect the real-time calm state of the user.

[0060] Specifically, the real-time physiological signal of the target user is mapped to a high-dimensional space for comparison with the first signal. The non-linear characteristics of the heart rate variability signal are more easily captured in the high-dimensional space. The high-dimensional space mapping amplifies the subtle differences between signals, and the first region in the signal is identified. Through high-dimensional space comparison and modal decomposition, subtle physiological changes caused by fatigue can be amplified, early or mild fatigue can be detected, the timeliness of fatigue evaluation can be improved, and the accuracy of the first region identification result can be improved.

[0061] At the same time, fatigue is often caused by the interaction of multiple factors such as environment and behavior. By analyzing the causes of fatigue in combination with real-time environmental data, the causes of fatigue are used as nodes, and directed edges are established between the nodes according to the causal relationship between the causes of fatigue, and a fatigue causal diagram is constructed. By constructing the fatigue causal diagram, the causes of fatigue can be analyzed in multiple dimensions, reflecting the indirect effects and feedback loops between nodes, and helping to identify the causes of fatigue. By constructing the fatigue causal diagram in combination with real-time environmental data, the fatigue causal diagram can adapt to real-time environmental changes, improving the practicality and real-time performance of the fatigue causal diagram.

[0062] Specifically, the real-time environmental data includes, but is not limited to, target user personal physiological and behavioral state data and physical environment parameter data. The personal physiological and behavioral state data includes, but is not limited to, total sleep duration, deep sleep proportion, wake-up times, physical activity intensity and duration, and continuous working time. Sleep deprivation and physical overload can directly consume physiological resources and reduce parasympathetic nervous tension, which corresponds to a decrease in heart rate variability high-frequency power. Long-term mental load can lead to sustained stress response. Real-time collection of target user personal physiological and behavioral state data can analyze the internal causes of physiological and psychological fatigue. The physical environment parameter data includes, but is not limited to, environmental temperature and humidity, environmental noise level, and environmental lighting conditions. A hot environment with high temperature and humidity can increase cardiovascular and thermal regulation load. Noise is a recognized stressor that can cause sympathetic nervous excitation. Abnormal lighting can interfere with circadian rhythms. Real-time collection of physical environment parameter data provides rich data support and cause support for analyzing fatigue causes.

[0063] Specifically, the first region is matched in the fatigue causal graph, and the physiological features and associated environmental features of the first region are extracted. The physiological features include, but are not limited to, residual distribution and signal fluctuation amplitude of the real-time modal set in the first region. The associated environmental features include, but are not limited to, average environmental temperature and noise peak value collected synchronously in the first region period. The extracted features are arranged in order to obtain a first region feature vector. The cosine similarity of the first region feature vector and each node feature label vector is calculated, a fatigue similarity threshold is set by statistical history fatigue cases, and reason nodes with a cosine similarity greater than the fatigue similarity threshold are screened to obtain a candidate reason node set. The influence of each node in the candidate reason node set on the first region is simulated through a pre-set fatigue analysis model, and the contribution value of each node to fatigue is quantified. A greater contribution value indicates that the node is a core reason for inducing current fatigue. The main fatigue reason and fatigue level are determined in combination with the contribution value to analyze the reason for fatigue, and a fatigue evaluation result is obtained to evaluate the fatigue state. Through node simulation and contribution value calculation, quantitative analysis of fatigue causes is realized, high-contribution-value fatigue causes can be improved preferentially, subjective judgment can be avoided, and the accuracy of fatigue cause analysis results can be improved. Multi-factor interaction can be analyzed, fatigue causes can be analyzed in combination with fatigue evaluation, the practicality of fatigue analysis results can be improved, and thus fatigue conditions can be quickly intervened.

[0064] The application processes physiological signals in sections through a state recognition model, recognizes calm anchor points, and decouples reference signals for signal reconstruction, maps real-time target signals and first signals to a high-dimensional space, extracts feature differences using attractor trajectories and dynamic mode decomposition, locates to the first area, constructs a fatigue causal diagram combining real-time environmental data, simulates the influence of node changes on the first area, quantifies the fatigue contribution value of each node, analyzes the corresponding fatigue causes, and realizes fatigue state evaluation; the signal reference provided by reconstructing the signal through the recognition of calm anchor points for the fatigue analysis process reduces the influence of individual physiological differences on the fatigue analysis process, improves the accuracy of the fatigue evaluation result, and through the construction of the fatigue causal diagram, the fatigue causes can be accurately quantified and analyzed, providing a reference for fatigue intervention and improving the environmental applicability and individual applicability of the fatigue analysis result.

[0065] Further, according to the pre-acquired target signal of the target user, the calm anchor points in the signal are recognized through a preset state recognition model, and the first signal is obtained by reconstructing the signal, including:

[0066] S201, according to the pre-acquired target signal of the target user, the target signal is processed in sections according to time sequence, and a plurality of signal segments are obtained;

[0067] S202, for each signal segment, the corresponding calm anchor points in the signal are recognized through a preset state recognition model;

[0068] S203, according to the calm anchor points, decouple the reference signal from each signal segment, and splice and reconstruct the reference signal to obtain the first signal.

[0069] In this embodiment, the long-time continuous physiological signal includes both a smooth section reflecting a stable baseline and a fluctuating section affected by instantaneous interference. According to the pre-acquired physiological signal of the target user, the physiological signal is processed in sections according to time sequence, and the continuous signal is disassembled into a plurality of independent signal segments; the time length of the sliding window is set according to the calm anchor point recognition accuracy, the sliding window step length is set to 1 / 2 of the window length, and a plurality of signal segments are obtained according to the sliding window. By disassembling the long-time continuous physiological signal into a plurality of independent signal segments through the sliding window, the risk of local fluctuations being masked or misjudged due to too long signals can be reduced, the signal processing efficiency can be improved, and the accuracy of the first area recognition result can be improved.

[0070] Specifically, in each signal segment, the user's short-term slight physiological fluctuations, including but not limited to short-term breath holding and involuntary muscle tension, will be included, which will cause the heart rate variability characteristics of some areas within the segment to deviate from the stable healthy state. Through a preset state recognition model, the signal region with the smallest feature fluctuation and the highest consistency in the signal is identified as the calm anchor point that best reflects the user's stable healthy physiological state. By identifying the calm anchor point, accurate positioning basis is provided for decoupling the reference signal, avoiding the reference signal being contaminated by interference components due to local fluctuations within the segment, and improving the accuracy of the reconstructed first signal, which accurately reflects the user's actual calm state.

[0071] Specifically, the calm anchor point corresponds to the stable and healthy feature region within the signal segment. According to the calm anchor point, the reference signal is decoupled from each signal segment, and the reference signal that reflects the user's stable and healthy state without short-term fluctuation interference is separated. A single reference signal only corresponds to a local stable region of the signal segment. By clustering, reference signals with high feature consistency are selected, and then spliced in time sequence to obtain a continuous and complete first signal. The signal decoupling model based on the calm anchor point can accurately and quickly separate short-term fluctuation interference, avoid the influence of interference components on the fatigue judgment process, and construct the first signal based on the user's stable and healthy features, which can match the individual physiological differences of users and improve the accuracy of fatigue analysis results for each user.

[0072] Further, for each signal segment, the corresponding calm anchor point in the signal is identified by a preset state recognition model, including:

[0073] S301, for each signal segment, extract signal features by a preset state recognition model, calculate the similarity between each signal feature vector, and construct a feature similarity matrix;

[0074] S302, generate a feature map based on the feature similarity matrix, analyze the feature distribution in the feature map, and identify a connected subgraph;

[0075] S303, identify the point with the most stable feature distribution in each connected subgraph as the corresponding calm anchor point.

[0076] In the embodiment, for each signal segment, the signal segment is converted into a quantifiable feature vector by a preset state recognition model, the similarity between each signal feature vector is calculated, the higher the similarity, the closer the physiological state of the corresponding period to the stable and healthy state, and a feature similarity matrix is constructed. The state recognition model includes but is not limited to a neural network model, a large amount of historical signal segment data is used to train the neural network model, a pre-trained neural network model is obtained, each signal segment is input into the pre-trained neural network model, and the model extracts time domain features and frequency domain features of the signal respectively. The time domain features include but are not limited to the standard deviation of the signal interval sequence and the root mean square of the adjacent signal interval difference, and the frequency domain features include but are not limited to the low frequency band power and the high frequency band power. The extracted features are sequentially constructed into a signal feature vector, the similarity between any two signal feature vectors is obtained by calculating the cosine similarity, and a feature similarity matrix is constructed according to the similarity. Through multi-dimensional feature extraction, the key indicators of the heart rate variability signal reflecting the physiological stable state can be obtained, the feature loss caused by single-dimensional features is avoided, the comprehensiveness of the similarity calculation is improved, and a data basis is provided for signal state evaluation by constructing the feature similarity matrix.

[0077] As shown in Figure 2 , the similarity threshold is set to 0.3, the feature vector nodes with a similarity value greater than 0.3 between two feature vectors are connected, and the similarity value between the nodes is used as the weight of the edge. Based on the feature similarity matrix, a feature graph is generated, each feature vector in the feature similarity matrix is used as a node of the feature graph, the similarity threshold is set by statistical analysis of the feature similarity distribution of a large number of healthy people's signal segments, if the similarity value between two feature vectors in the matrix is greater than the similarity threshold, a connection is established between the corresponding two nodes, the weight of the edge is the corresponding similarity value, and the feature graph is generated. Analyze the connection density of the nodes in the feature graph and the feature distribution in the feature graph, start from any node of the feature graph, traverse all nodes connected to the node through edges by a depth-first search algorithm, and identify a connected subgraph. Repeat the process until all nodes in the feature graph are traversed, and a plurality of non-overlapping connected subgraphs are obtained. By constructing the feature graph and identifying the connected subgraph, a candidate area is provided for the identification and positioning of the calm anchor point, the identification range of the calm anchor point is narrowed, the calm anchor point is ensured to be selected from the feature stable area, and the accuracy and identification efficiency of the calm anchor point identification process are improved.

[0078] The feature vectors in the same connected subgraph have high similarity, but there are still slight feature fluctuations, and not all nodes can represent the stable and healthy state of the signal segment. The node with the smallest feature fluctuation and the most stable distribution in each connected subgraph is identified as a calm anchor point, and the signal area corresponding to the node can best reflect the stable physiological state of the user. For each connected subgraph, the feature vector corresponding to the node is extracted, and the standard deviation of the feature vector in the connected subgraph and the mean value of the feature vector in the connected subgraph are calculated to obtain a feature fluctuation coefficient. The node with the smallest feature fluctuation coefficient is selected as the calm anchor point. By selecting the calm anchor point in each connected subgraph, it is ensured that the calm anchor point is not only in a stable state stage, but also is the most typical moment in the stage that is least affected by slight fluctuations. The representativeness of the individual health state baseline signal is improved, the quality of the calm anchor point is improved, and thus the accuracy of the baseline signal decoupling and the reconstruction of the first signal process is improved, and the accuracy and reliability of the fatigue analysis result are improved.

[0079] Further, according to the calm anchor point, the baseline signal is decoupled from each signal segment, and the baseline signal is spliced and reconstructed to obtain a first signal, comprising:

[0080] S401, according to the calm anchor point, a predetermined signal decoupling model is used to decouple the baseline signal from each signal segment;

[0081] S402, feature extraction is performed on the baseline signal, and the features are clustered, and the baseline signal is spliced and reconstructed according to the cluster center to obtain a first signal.

[0082] In this embodiment, the signal segment of a certain time length before and after the calm anchor point is taken as the decoupling target segment, and the time length can be set according to the calculation accuracy requirement. By taking the decoupling target segment, the influence of other fluctuation regions in the segment on decoupling can be reduced. A predetermined signal decoupling model is used to decouple the baseline signal from each signal segment. The signal decoupling model includes but is not limited to an independent component analysis model pre-trained by a large amount of historical decoupling target segment data. The model separates stable and healthy components and short-term interference components from the decoupling target segment, and the stable and healthy components are used as the baseline signal. By decoupling the signal based on the calm anchor point, the baseline signal reflecting the inherent and stable physiological characteristics of the user can be effectively extracted from the signal segment, the influence of transient interference on the baseline signal estimation is reduced, accurate baseline component separation is realized, and a data basis for constructing a high-quality first signal is provided.

[0083] Specifically, feature extraction is performed on the reference signals to extract corresponding time domain features and frequency domain features, a feature vector is constructed, the feature vector is clustered by an AP clustering algorithm, a plurality of categories are obtained, and each clustering center is taken as an effective reference signal; the effective reference signals are arranged in sequence according to the time sequence of the original signal segments corresponding to the effective reference signals, data filling and reconstruction are performed by using a linear interpolation algorithm, and a continuous and complete target user first signal is obtained. By screening effective reference signals through feature clustering and reconstructing the signal, the information of reference signal segments at different time points can be fused, the first signal obtained by reconstruction provides a data basis for real-time fatigue analysis, improves the stability and representativeness of the first signal, and thus improves the accuracy and reliability of detecting the physiological state deviation process caused by real fatigue based on the first signal.

[0084] Further, the real-time target signal of the target user and the first signal are mapped to a high-dimensional space for comparison, and a first region in the signal is identified, including:

[0085] S501, mapping the real-time target signal of the target user and the first signal to a high-dimensional space, reconstructing an attractor trajectory, obtaining a real-time attractor and a reference attractor;

[0086] S502, modal decomposition is performed on the real-time attractor and the reference attractor respectively, corresponding dynamic modes are extracted, and a real-time modal set and a reference modal set are obtained;

[0087] S503, based on the real-time modal set and the reference modal set, a first region in the signal is identified.

[0088] In this embodiment, the real-time physiological signal of the target user and the first signal are mapped to a high-dimensional space, the linearly inseparable signal features in the low-dimensional space are converted into linearly separable feature vectors in the high-dimensional space by a pre-trained kernel principal component analysis model, the kernel principal component analysis model is an analysis method for mapping data in the prior art, which will not be described here, and the signals are mapped to the high-dimensional space and the adjacent points in different dimensions are analyzed. The signal feature vectors in the high-dimensional space are rearranged to form continuous trajectory points, the trajectory corresponding to the real-time physiological signal is taken as a real-time attractor, and the trajectory corresponding to the first signal is taken as a reference attractor. By mapping the real-time physiological signal and the first signal to the high-dimensional phase space to reconstruct the attractor, the nonlinear dynamic characteristics hidden in the signal and difficult to directly capture in the time domain or the frequency domain can be analyzed. The geometric characteristics of the reconstructed attractor trajectory are more sensitive to the state change of the system. Even if the signal amplitude changes little, if the dynamic mode changes, the attractor morphology difference can also be detected, thereby improving the early and subtle fatigue state detection precision and the accuracy of the fatigue analysis result.

[0089] Specifically, the real-time attractor and the reference attractor are respectively subjected to mode decomposition by a pre-trained empirical mode decomposition model, different scale modes are gradually separated, and local maximum points and minimum points in the attractor trajectory are identified. The empirical mode decomposition model is a signal analysis method for mode decomposition of data in the prior art, and will not be described here. A cubic spline interpolation fitting is used to form an upper envelope line and a lower envelope line, and a mean line of the envelope line is calculated. The attractor trajectory is subtracted from the mean line to obtain an initial component. If the initial component satisfies that the mean values of the upper and lower envelope lines at any time are close to 0, and the number of extreme points and the number of zero-crossing points differ by no more than 1, it is determined as a dynamic mode. The dynamic mode is analyzed and determined to obtain a real-time mode set and a reference mode set. By extracting the dynamic mode set, the dynamic characteristics are dimensionally reduced and characterized, and the continuous trajectory comparison is converted into discrete mode set comparison, which can simplify the analysis process and improve the analysis efficiency. Through mode decomposition, the oscillation mode changed under the fatigue state can be quantitatively analyzed, thereby providing a target and data basis for accurate positioning of the first region.

[0090] Specifically, based on the real-time mode set and the reference mode set, the reference mode set is the dynamic characteristic of the user in a healthy state, and the real-time mode set reflects the dynamic characteristic of the current physiological state. The real-time mode is projected onto the reference base constructed by the reference mode to quantify the difference between the two types of modes. The greater the difference, the farther the real-time mode deviates from the healthy standard, and the signal period corresponding to the signal is in a fatigue state. The first region in the signal is identified. By comparing the modes and analyzing the differences, the start time and duration of the first region can be identified, improving the accuracy of the positioning result of the first region. The first region provides an accurate time-related anchor point for fatigue cause analysis, improving the accuracy and practicality of the fatigue analysis result.

[0091] Further, based on the real-time mode set and the reference mode set, the first region in the signal is identified, including:

[0092] S601, constructing a reference base based on a reference mode set;

[0093] S602, projecting a real-time mode set onto the reference base, calculating a projection residual, analyzing the feature difference between the real-time mode set and the reference mode set, and constructing a difference feature vector;

[0094] S603, identifying a first region in a signal according to the difference feature vector.

[0095] In the embodiment, based on the reference modal set, the reference modal set is converted into a modal matrix, the rows of the matrix correspond to different modes, and the columns correspond to time sequence sampling points of each mode; singular value decomposition is performed on the modal matrix to obtain a left singular vector matrix, a singular value matrix and a right singular vector matrix, the left singular vector matrix reflects the characteristic correlation between modes, the singular value matrix reflects the importance of each characteristic dimension, and the right singular vector matrix reflects the characteristic distribution of the time dimension; the cumulative proportion of the singular values from large to small is calculated, a cumulative threshold is set according to the calculation accuracy requirement, the left singular vectors corresponding to the singular values with the cumulative proportion greater than the cumulative threshold are screened out, and a reference basis is constructed; by constructing the reference basis, the correlation interference between different modes can be eliminated, the influence of different dynamic modes can be distinguished, and an accurate data basis for the identification of the first area is provided.

[0096] Specifically, the real-time modal set is projected onto the reference basis, each real-time modal vector is projected onto the corresponding characteristic dimension of the reference basis by using the least square method, the projection vector of the real-time modal on the reference basis is calculated, the difference between the real-time modal vector and the projection vector is quantified by using the Euclidean two norm, the projection residual is calculated, and the deviation degree of the corresponding modal is reflected; the residual values of all real-time modes are arranged in order to construct a difference characteristic vector; through the projection and residual calculation, the dynamic mode difference is quantified, the first area identification process is simplified, and the first area identification efficiency and accuracy are improved.

[0097] According to a plurality of physiological signals of the target user in a known health state, a difference characteristic vector of each group of signals is calculated, and a residual judgment threshold is set according to the distribution of the corresponding difference characteristic vector; the real-time difference characteristic vector is compared with the residual judgment threshold, and a time segment with a difference characteristic vector greater than the residual judgment threshold is selected as a fatigue period; and the continuous fatigue periods are combined to obtain a complete first area. By analyzing the time sequence of the difference characteristic vector to identify the first area, the fatigue state can be time-positioned, the starting time and duration of fatigue occurrence can be identified, and the true and continuous fatigue state can be distinguished from the temporary and random physiological fluctuation or measurement noise, so that the fatigue false alarm rate is reduced and the accuracy of the first area identification result is improved.

[0098] Further, fatigue reasons are analyzed in combination with real-time environmental data, fatigue reasons are taken as nodes, directed edges are established between the nodes according to the causal relationship between the fatigue reasons, a fatigue causal graph is constructed, including:

[0099] S701, fatigue reasons are analyzed in combination with real-time environmental data, a causal relationship between the fatigue reasons is analyzed, and a causal characteristic matrix is constructed;

[0100] S702, fatigue reasons are taken as nodes, directed edges are established between the nodes according to the causal characteristic matrix, and a fatigue causal graph is constructed.

[0101] In the embodiment, the fatigue causes are analyzed in combination with real-time environmental data, potential fatigue causes are screened out, a cause set is obtained, a pre-trained causal inference model is used to analyze the causal relationship between the fatigue causes, the corresponding causal strength is calculated, the causal inference model is an analysis method for variable causal analysis of data in the prior art, and details are not repeated here. A causal feature matrix is constructed according to the causal strength. The causal relationship between fatigue and causes can be analyzed by constructing the causal feature matrix, so that the fatigue causes can be analyzed and identified. The fatigue cause analysis provides a data basis.

[0102] As shown in Figure 3 , the fatigue causes are taken as nodes, each node contains a cause name and a cause type; a directed edge is established between the nodes according to the causal strength in the causal feature matrix, the direction of the edge is from the cause to the result, and a fatigue causal graph is constructed. By constructing the fatigue causal graph, the main causes of fatigue and the interaction paths between the causes can be clearly seen, data support is provided for analyzing the fatigue causes, the corresponding cause nodes can be quickly identified, and a basis is provided for formulating accurate intervention strategies.

[0103] Further, the first area is matched in the fatigue causal graph, the influence of the change of each node in the fatigue analysis model on the first area is simulated, the causes of fatigue are analyzed, and fatigue evaluation results are obtained, including:

[0104] S801, matching the first area in the fatigue causal graph, identifying the corresponding cause nodes, and obtaining a candidate cause node set;

[0105] S802, simulating the influence of the change of each node in the candidate cause node set on the first area by using the pre-set fatigue analysis model, analyzing the causes of fatigue, and obtaining fatigue evaluation results.

[0106] In the embodiment, the first region is matched in the fatigue causal diagram, physiological features and associated environmental features of the first region are extracted, the physiological features include but are not limited to residual distribution of a real-time modal set in the first region, signal fluctuation amplitude, the associated environmental features include but are not limited to average temperature of an environment collected synchronously in a time period of the first region, noise peak value, the extracted features are arranged in order to obtain a first region feature vector; a cosine similarity of the first region feature vector and each node feature label vector is calculated, a fatigue similarity threshold is set by statistics of historical fatigue cases, a reason node with a cosine similarity greater than the fatigue similarity threshold is screened to obtain a candidate reason node set. By matching the first region with the fatigue causal diagram, the monitored physiological abnormalities are associated and screened with corresponding fatigue reasons, the fatigue reason analysis range is narrowed, and the fatigue reason analysis efficiency is improved; the fatigue feature matching improves the pertinence and accuracy of the candidate reason node set, provides a data basis for fatigue reason analysis, and concentrates the reason analysis process on a few key factors causing the current fatigue, thereby improving the accuracy and practical value of the fatigue evaluation result.

[0107] Specifically, the influence of each node in the candidate reason node set on the first region is simulated through a preset fatigue analysis model, the contribution value of each node to fatigue is quantified, and a greater contribution value indicates that the node is a core reason for inducing the current fatigue. The main fatigue reason is determined in combination with the contribution value, the fatigue level is analyzed, and a fatigue evaluation result is obtained. Through node simulation and contribution value calculation, the causal effect can be quantified, the influence effect of the node can be simulated and analyzed, and the accuracy of the fatigue reason analysis result can be improved.

[0108] Further, the influence of each node in the candidate reason node set on the first region is simulated through a preset fatigue analysis model, the fatigue reason is analyzed, and a fatigue evaluation result is obtained, including:

[0109] S901, simulating the influence of each node in the candidate reason node set on the first region through a preset fatigue analysis model to obtain a simulation result;

[0110] S902, analyzing the influence of a single node and a node combination respectively according to the simulation result, and calculating a fatigue contribution value of each node;

[0111] S903, analyzing the fatigue reason according to the fatigue contribution value to obtain a fatigue evaluation result.

[0112] In the embodiment, the influence of each node change in the candidate cause node set on the first area is simulated by a preset fatigue analysis model, the fatigue analysis model includes but is not limited to a convolutional neural network model, the convolutional neural network model is trained by using a large amount of cause node set data to obtain a pre-trained convolutional neural network model, the candidate cause node set is input into the pre-trained convolutional neural network model, the model sets a reasonable change range according to the node type to ensure that the change range conforms to the actual scene and can cause observable first area feature change, for each node in the candidate cause node set, an input pair of baseline parameters and changed parameters is generated, and other node parameters are fixed as current measured values to avoid cross interference; the model outputs the change amount of the first area core feature after the corresponding node parameter change to obtain a simulation result.

[0113] It should be noted that the multi-factor influence can be separated by node change simulation, the independent contribution of each candidate cause node to the current first area can be analyzed, the misleading caused by only relying on the correlation or time sequence between variables for attribution can be avoided, the simulation result is used to provide data support for calculating the contribution value of each node, and the scientificity and accuracy of the fatigue analysis result are improved.

[0114] Specifically, the influence of a single node and a node combination is analyzed according to the simulation result, the change rate of a single node feature and the change rate of a feature after node random combination are calculated by a pre-trained random forest model, the random forest model is an existing technology for classifying and analyzing data, and details are not described herein, and the feature change rate reflects the influence intensity of the first area under the action of the node. The fatigue contribution value of each node is obtained by averaging the calculated node feature change rate. By calculating the fatigue contribution value of the node, the total contribution of each cause node to fatigue in the case of multiple factors and interaction can be quantified; the synergistic effect between factors can be considered by combining node analysis, the importance of a single factor is avoided to be overestimated or underestimated, the accuracy of the calculated fatigue contribution value is improved, and thus the accurate fatigue cause is analyzed.

[0115] Specifically, the causes of fatigue are analyzed according to the fatigue contribution value, the fatigue contribution values are sorted from high to low, the top N nodes are selected as the core fatigue causes, and N is set according to the fatigue contribution value proportion; for each core fatigue cause, the corresponding influence feature is analyzed, the influence degree of the fatigue feature is analyzed through the pre-trained deep neural network model, the corresponding fatigue grade is determined, and the deep neural network model is an analysis method for deep analysis of data in the prior art, which will not be described here; the fatigue grade and the fatigue cause are integrated to obtain the fatigue evaluation result. Through the contribution value analysis and the fatigue grade evaluation, the core fatigue cause can be identified, the corresponding influence feature can be associated, the explainability of the fatigue analysis result can be enhanced, the one-sidedness of single feature determination can be avoided, the evaluation result can be obtained through analysis, and accurate fatigue analysis result and grade and fatigue cause are obtained, thereby providing an accurate direction for fatigue intervention.

[0116] As shown in Figure 4 A fatigue state monitoring system for implementing a fatigue state monitoring method, comprising:

[0117] A signal analysis module identifies a calm anchor point in the signal through a preset state recognition model according to a pre-acquired target signal of a target user, and reconstructs a first signal from the signal;

[0118] A first region identification module maps the real-time target signal of the target user to a high-dimensional space for comparison with the first signal, and identifies a first region in the signal;

[0119] A fatigue causal diagram construction module analyzes fatigue causes in combination with real-time environmental data, takes the fatigue causes as nodes, establishes directed edges between the nodes according to the causal relationship between the fatigue causes, and constructs a fatigue causal diagram;

[0120] A fatigue analysis module matches the first region in the fatigue causal diagram, simulates the influence of changes in each node on the first region through a preset fatigue analysis model, analyzes the causes of fatigue, and obtains a fatigue evaluation result to monitor the fatigue state.

[0121] In this embodiment, the signal analysis module receives the physiological signal pre-acquired by the target user, identifies the signal segment in which the user is in a calm and non-fatigue state through the state recognition model, and reconstructs the first signal reflecting the health state of the user according to the calm anchor point; by identifying the calm anchor point and reconstructing the signal, the signal noise introduced by temporary emotional fluctuations, environmental interference or accidental activities in the physiological signal can be removed, and it is ensured that the constructed first signal can reflect the physiological state of the user, thereby providing a comparison benchmark for comparing and analyzing the fatigue state of the user.

[0122] The first region identification module maps the real-time physiological signal and the first signal to a high-dimensional phase space, respectively reconstructs corresponding attractor trajectories, extracts dynamic modes through dynamic mode decomposition, and identifies the first region by comparing the differences between the dynamic modes. By comparing the dynamic characteristics of the system in the high-dimensional space, subtle changes in the fatigue state can be identified. By analyzing the internal oscillation patterns of the physiological rhythm through dynamic mode decomposition, the anti-interference ability of the detection results can be enhanced, the fatigue state can be time-positioned, and accurate first region identification results can be obtained.

[0123] The fatigue causal diagram construction module analyzes the causal relationship and strength between each environmental factor and fatigue in combination with real-time environmental data, and constructs a fatigue causal diagram. By constructing the fatigue causal diagram, direct and indirect causal relationships between fatigue causes can be analyzed, thereby analyzing the causes of fatigue and providing accurate correlation reference for the fatigue analysis process. The fatigue analysis module matches the first region in the fatigue causal diagram, simulates and analyzes the influence degree of each reason node on the first region through a pre-set fatigue analysis model, analyzes the causes of fatigue, and obtains a fatigue evaluation result to evaluate the fatigue state. Through fatigue simulation analysis, the influence of each reason on the fatigue state can be calculated, the most critical influencing factor can be solved in priority according to the influence degree, and the efficiency and accuracy of the fatigue analysis process can be improved.

[0124] Embodiment 2

[0125] This embodiment combines a specific application scenario to illustrate the entire workflow of the technical solution from data input to result output. For example, the target user is an office worker. The system collects data through the smart watch and the paired smart phone worn by the user, including but not limited to physiological signals: 24-hour continuous monitoring of electrocardiogram signals, from which the RR interval sequence, i.e. the original signal of heart rate variability (HRV), is extracted; personal physiological and behavioral state data: sleep data through the watch accelerometer and heart rate monitor; work load data: through the phone usage record and schedule synchronization, it is confirmed that the user has been continuously conducting video conference for 3 hours in the morning; physical environment parameter data: through the phone microphone, intermittent sampling is performed during the office period, and it is detected that the average value of environmental noise is 65 decibels; all data streams are synchronized and aligned through a unified time stamp.

[0126] The system selects the night HRV signal of the user on the rest day (sufficient sleep, no work pressure) in the past week as the pre-acquired physiological signal, segments the night HRV signal into multiple signal segments with a length of 5 minutes and an overlap rate of 50%; for each signal segment, the state recognition model extracts frequency domain features to construct a feature vector, the frequency domain features include but are not limited to low frequency power LF, high frequency power HF, and LF / HF ratio, calculates the cosine similarity between all feature vectors in the segment, and constructs a feature similarity matrix; by analyzing the graph structure generated by the feature similarity matrix, multiple connected subgraphs are recognized, in a connected subgraph, the system locates to a point with the most gentle change of the feature vector, and the corresponding time point is 3 o'clock in the morning, at this time, the LF / HF ratio is stable at about 1.2, and the HF power is high, and the point is marked as the calm anchor point of the segment; taking the calm anchor point as the center, the HRV signal of 2.5 minutes before and after the anchor point is taken as the reference template of the segment; through an adaptive filter, the reference HRV signal component of the 5-minute signal segment is decoupled by taking the template as the expected response; repeat the above operation on all night signal segments to obtain a series of reference signal segments; extract features by principal component analysis on the reference signal segments, and find the corresponding core mode through K-means clustering; splice the reference signal representing the core mode to obtain a smooth and continuous healthy state HRV signal as the personal reference of the user.

[0127] On the target day, the system analyzes the real-time HRV signal of the user; the real-time HRV signal and the reconstructed healthy state HRV signal are respectively mapped to a three-dimensional phase space by using a time delay method (time delay τ=1 beat, embedding dimension m=3); in the phase space, the reconstructed trajectory of the healthy state HRV signal presents a regular ellipsoid shape, while the reconstructed trajectory of the real-time HRV signal appears diffusion and trajectory divergence in a local period. The reconstructed trajectory of the healthy state HRV signal is taken as a reference attractor, and the reconstructed trajectory of the real-time HRV signal is taken as a real-time attractor, and the two attractors are respectively subjected to dynamic mode decomposition to obtain the main mode frequency of the reference attractor and the mode frequency of the real-time attractor. The mode of the real-time attractor is projected into the subspace spanned by the modes of the reference attractor, and the projection residual is calculated. If the norm of the projection residual continuously exceeds 2 times the standard deviation of the healthy state data statistical threshold, the corresponding period is recognized as the fatigue domain, i.e. the first region, and the feature vector of the HRV dynamic mode anomaly in the period is recorded.

[0128] The system learns and constructs the personalized fatigue causal graph of the user through PC causal discovery algorithm according to long-term collected user data; the nodes in the fatigue causal graph include but are not limited to sleep duration, deep sleep proportion, continuous cognitive working time, environmental noise, environmental temperature, subjective stress and fatigue degree; the directed edges include but are not limited to sleep duration deep sleep proportion, deep sleep proportion fatigue degree, according to the fatigue domain time, the data of each node corresponding to the fatigue domain time is viewed, the data abnormal nodes are screened out, the time sequence related node set such as {sleep duration, continuous cognitive working time, environmental noise} is obtained; the causal relationship between nodes is analyzed by starting from the fatigue degree node and tracing back in the fatigue causal graph, the causal reachable node set {sleep duration, deep sleep proportion, continuous cognitive working time, environmental noise, subjective stress} is obtained; the intersection is taken to obtain the candidate reason node set {sleep duration, continuous cognitive working time, environmental noise}.

[0129] The system loads the trained convolutional neural network as the fatigue analysis model, the structure of which is the same as the fatigue causal graph Figure 1 The nodes in the candidate reason node set and other nodes are set to the actual observation values before the fatigue domain occurs, the model is run, the fatigue contribution value is calculated and calculated, the influence of the change of environmental factors on the fatigue contribution value is simulated, all node combinations are analyzed and the fatigue relief total amount after all nodes are intervened to the ideal value under each combination is calculated, and the corresponding evaluation report is generated and output, including but not limited to fatigue time, main reason analysis and fatigue relief suggestion.

[0130] The above is only the preferred embodiment of the present application, the protection scope of the present application is not limited to the above-mentioned embodiments, all the technical solutions belonging to the idea of the present application are within the protection scope of the present application. It should be noted that for ordinary skilled in the art, some improvements and decorations without departing from the principles of the present application can also be considered as the protection scope of the present application.

Claims

1. A method for monitoring fatigue status, characterized in that, include: Based on the target user's pre-acquired target signal, the target signal is segmented according to time sequence to obtain multiple signal segments; For each signal segment, signal features are extracted using a pre-defined state recognition model, the similarity between each signal feature vector is calculated, and a feature similarity matrix is ​​constructed. A feature map is generated based on the feature similarity matrix, the feature distribution in the feature map is analyzed, and connected subgraphs are identified. Identify the point with the most stable feature distribution in each connected subgraph and use it as the corresponding calm anchor point; Based on the calm anchor point, the reference signal is decoupled from each signal segment using a preset signal decoupling model; Feature extraction is performed on the baseline signal, and the features are clustered. The baseline signal is then spliced ​​and reconstructed according to the cluster centers to obtain the first signal. The target signal is the physiological signal corresponding to the target user, including the heart rate variability signal. The calm anchor point is used to reflect the user's stable and healthy physiological state. The first signal is used to reflect the user's real-time calm state. The real-time target signal of the target user is mapped to a high-dimensional space along with the first signal to reconstruct the attractor trajectory, thereby obtaining the real-time attractor and the reference attractor. Modal decomposition is performed on the real-time attractor and the reference attractor respectively to extract the corresponding dynamic modes, thereby obtaining the real-time mode set and the reference mode set; Based on the real-time mode set and the reference mode set, the first region in the signal is identified; By combining real-time environmental data analysis to identify fatigue causes, fatigue causes are used as nodes. Directed edges are established between nodes based on the causal relationships between fatigue causes to construct a fatigue causal graph. The first region is matched in the fatigue cause-effect graph. The impact of each node change on the first region is simulated by a preset fatigue analysis model. The causes of fatigue are analyzed, and fatigue assessment results are obtained to monitor the fatigue state.

2. The fatigue state monitoring method according to claim 1, characterized in that, The process of identifying the first region in the signal based on the real-time mode set and the reference mode set includes: A baseline is constructed based on the baseline mode set; The real-time mode set is projected onto the reference base, the projection residual is calculated, the feature differences between the real-time mode set and the reference mode set are analyzed, and the difference feature vector is constructed. Based on the difference feature vector, the first region in the signal is identified.

3. The fatigue state monitoring method according to claim 1, characterized in that, The process of analyzing fatigue causes using real-time environmental data involves treating fatigue causes as nodes, establishing directed edges between nodes based on the causal relationships between these causes, and constructing a fatigue causal graph, including: By combining real-time environmental data analysis, the causes of fatigue are analyzed, the causal relationships between the causes of fatigue are analyzed, and a causal feature matrix is ​​constructed. By treating fatigue causes as nodes, and establishing directed edges between nodes based on the causal feature matrix, a fatigue causal graph is constructed.

4. The fatigue state monitoring method according to claim 1, characterized in that, The process involves matching the first region to the fatigue causal graph, simulating the impact of changes at each node on the first region using a preset fatigue analysis model, analyzing the causes of fatigue, and obtaining fatigue assessment results, including: Match the first region in the fatigue cause-effect graph, identify the corresponding cause nodes, and obtain a set of candidate cause nodes; By simulating the impact of changes in each node in the candidate cause node set on the first region using a pre-set fatigue analysis model, the causes of fatigue are analyzed, and fatigue assessment results are obtained.

5. The fatigue state monitoring method according to claim 4, characterized in that, The process involves simulating the impact of changes in each node in the candidate cause node set on the first region using a preset fatigue analysis model, analyzing the causes of fatigue, and obtaining fatigue assessment results, including: The simulation results are obtained by simulating the impact of changes in each node in the candidate cause node set on the first region using a preset fatigue analysis model. Based on the simulation results, the impact of individual nodes and node combinations is analyzed separately, and the fatigue contribution value of each node is calculated. The causes of fatigue are analyzed based on the fatigue contribution value, and fatigue assessment results are obtained.

6. A fatigue condition monitoring system, characterized in that, A fatigue state monitoring method as described in any one of claims 1 to 5, comprising: The signal analysis module identifies calm anchor points in the signal based on the target signal pre-acquired by the target user and reconstructs the signal to obtain the first signal. The first region identification module maps the real-time target signal of the target user to the first signal in a high-dimensional space for comparison and identifies the first region in the signal. The fatigue cause-effect graph construction module combines real-time environmental data analysis to identify fatigue causes, uses fatigue causes as nodes, and establishes directed edges between nodes based on the causal relationships between fatigue causes to construct a fatigue cause-effect graph. The fatigue analysis module matches the first region to the fatigue cause-effect graph, simulates the impact of changes in each node on the first region through a preset fatigue analysis model, analyzes the causes of fatigue, and obtains fatigue assessment results to monitor fatigue status.

Images