Physiological fatigue adaptive control system and method of motor imagery brain-computer interface exoskeleton

Abstract

The application discloses a kind of physiological fatigue self-adaptive control system and method of motor imagery brain-computer interface exoskeleton, system includes: multiple physiological signal acquisition module, for collecting motor imagery electroencephalogram and the physiological signal of physiological fatigue state;Physiological fatigue monitoring module is used to extract multidimensional feature calculation physiological fatigue index and fatigue grade;Electroencephalogram decoding and feature self-adaptive reconstruction module are used to adaptively adjust electroencephalogram decoding model according to physiological fatigue index;Auxiliary force compensation module is used to dynamically calculate exoskeleton auxiliary force output parameter according to preset mapping relationship of physiological fatigue index;Exoskeleton execution module is used to receive control instruction and auxiliary force output parameter drive lower limb exoskeleton and execute corresponding movement;Man-machine interaction and feedback module is used to provide real-time feedback and guide to user according to physiological fatigue index, effectively solve the decoding accuracy problem caused by user fatigue, guarantee the control stability of training.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation robots and brain-computer interface technology, and in particular to a physiological fatigue adaptive control system and method for a motor imagery brain-computer interface exoskeleton. Background Technology

[0002] Motor imagery brain-computer interface (MI-BCI) has broad application prospects in lower limb exoskeleton rehabilitation training. The MI-BCI system collects electroencephalogram (EEG) signals during the user's motor imagery process, decodes the user's motor intentions, and then controls the exoskeleton to assist the user in completing corresponding rehabilitation movements, achieving rehabilitation training based on active intention. Compared with passive rehabilitation training devices, the MI-BCI-driven exoskeleton system can fully mobilize the patient's central nervous system, which is beneficial for promoting neural remodeling and motor function recovery.

[0003] However, in actual long-term rehabilitation training, users are prone to physiological and psychological fatigue from prolonged use of the MI-BCI system. This fatigue can cause drift in the characteristics of the user's EEG signals, such as decreased signal amplitude, reduced signal-to-noise ratio, and altered frequency band energy distribution. Existing MI-BCI systems typically employ pre-trained EEG decoding models, whose recognition accuracy significantly decreases when the user is fatigued, leading to unstable control commands and even misjudgments, severely impacting the effectiveness of rehabilitation training and user compliance.

[0004] Current solutions rely on single-modal EEG signals for fatigue assessment, which fails to comprehensively and accurately reflect the user's overall physiological fatigue state. Furthermore, they only provide fatigue warnings without adaptively adjusting the EEG decoding model or exoskeleton control strategy based on fatigue levels, thus failing to address the decline in decoding accuracy caused by fatigue. Other solutions employ multi-modal training switching to adapt to different rehabilitation stages, but lack real-time monitoring and dynamic compensation for the user's physiological fatigue state, still facing issues of decreased control stability when the user is fatigued. In summary, existing technologies lack real-time fatigue monitoring and quantification methods based on multimodal physiological signal fusion, making it impossible to accurately perceive the user's fatigue state; EEG decoding models cannot adaptively adjust to compensate for feature drift when the user is fatigued, leading to a continuous decline in decoding accuracy; and exoskeleton assistive forces cannot dynamically change according to fatigue levels, making it difficult to provide effective load reduction when the user is fatigued.

[0005] Therefore, there is an urgent need for a control method that can integrate multimodal physiological signals for fatigue perception and adaptively adjust EEG decoding and exoskeleton assistance to ensure the long-term rehabilitation training effect and safety of the MI-BCI lower limb exoskeleton system. Summary of the Invention

[0006] This invention aims to address the problems of decreased decoding accuracy, poor control stability, and limited rehabilitation effects caused by physiological fatigue in existing MI-BCI exoskeleton systems during long-term user use. It provides a physiological fatigue adaptive control system for a motor imagery brain-computer interface exoskeleton, which realizes closed-loop control of real-time quantification of physiological fatigue, adaptive adjustment of EEG decoding, and dynamic compensation of assistive force, thereby improving the effectiveness of long-term rehabilitation training, user comfort, and system robustness.

[0007] The first aspect of this invention provides a physiological fatigue adaptive control system for a motor imagery brain-computer interface exoskeleton, comprising: The multi-physiological signal acquisition module is used to simultaneously acquire the user's motor imagery EEG signals and multimodal physiological signals of physiological fatigue state. The multimodal physiological signals include at least heart rate variability (HRV) related signals, skin conductance (GSR) signals, and EEG frequency band signals. The physiological fatigue monitoring module is connected to the multi-physiological signal acquisition module and is used to extract features from the multi-modal physiological signals and calculate the physiological fatigue index and corresponding fatigue level of the target user based on the extracted multi-dimensional features. The EEG decoding and feature adaptive reconstruction module is connected to the multi-physiological signal acquisition module and the physiological fatigue monitoring module, respectively. It is used to receive the motor imagery EEG signal and adaptively adjust the EEG decoding model according to the physiological fatigue index. When the model completes initialization and passes verification and the decoding confidence reaches the preset condition, it outputs a motor control command; otherwise, it outputs a safety control command to maintain, decelerate, or pause execution. An auxiliary force compensation module, connected to the physiological fatigue monitoring module, is used to dynamically calculate the exoskeleton auxiliary force output parameters according to the physiological fatigue index and a preset mapping relationship. The exoskeleton execution module is connected to the EEG decoding and feature adaptive reconstruction module and the auxiliary force compensation module, respectively, and is used to receive the control commands and the auxiliary force output parameters, and drive the lower limb exoskeleton to perform corresponding movements according to the control commands and the auxiliary force output parameters; The human-computer interaction and feedback module is connected to the physiological fatigue monitoring module and the auxiliary force compensation module, and is used to provide real-time feedback and guidance to the user based on the physiological fatigue index and / or the fatigue level.

[0008] Preferably, the physiological fatigue monitoring module includes: The feature extraction unit is used to extract fatigue-related features from the multimodal physiological signals. The fatigue-related features include at least two of the following: HRV time-domain features, HRV frequency-domain features, GSR features, and EEG frequency band power features. The fatigue index calculation unit is used to fuse the fatigue-related features and calculate the physiological fatigue index, which represents the user's current physiological load level, based on the fused features using a pre-trained machine learning model. The fatigue level determination unit is used to compare the physiological fatigue index with a preset multi-level fatigue threshold to determine the user's current fatigue level. The threshold determination branch unit is used to generate corresponding control signals according to the fatigue level and output them to the corresponding modules respectively: when the fatigue level is a non-fatigue level, the EEG decoding and feature adaptive reconstruction module is controlled to call the initial motor imagery decoding model established based on the user's non-fatigue state EEG data to decode the motor imagery EEG signal. When the fatigue level is at the fatigue level, the EEG decoding and feature adaptive reconstruction module is controlled to adaptively adjust the initial motor imagery decoding model according to the current fatigue level and the current EEG features to obtain a fatigue state motor imagery decoding model, and the module is controlled to generate motor control commands using the fatigue state motor imagery decoding model.

[0009] Preferably, the feature extraction unit includes: The EEG frequency band feature extraction subunit is used to perform bandpass filtering, notch filtering and artifact removal on the EEG frequency band signal, extract EEG segments through a preset first sliding window, perform time-frequency transformation on each segment to calculate the power spectral density of Delta band, Theta band, Alpha band and Beta band, and calculate the (Theta+Alpha) / Beta ratio and / or Alpha / Beta ratio. The heart rate variability feature extraction subunit is used to detect R wave or PPG peak from the heart rate variability related signal to calculate RR interval sequence, perform artifact correction and interpolation processing on the RR interval sequence, and calculate time domain indices SDNN and / or RMSSD, as well as frequency domain indices LF, HF and / or LF / HF ratio within a preset second sliding window. The skin conductance feature extraction subunit is used to perform low-pass filtering on the skin conductance signal to remove high-frequency noise, extract GSR segments through a preset third sliding window, and calculate the average skin conductance level and the amplitude and frequency of the skin conductance response for each segment.

[0010] Preferably, the EEG decoding and feature adaptive reconstruction module includes: The initial decoding unit is used to store the initial motor imagery decoding model trained based on the user's non-fatigue EEG data; The feature adaptive reconstruction unit is used to adaptively reconstruct the current EEG features when the physiological fatigue index reaches or exceeds the first threshold T1, so as to compensate for the feature distribution drift caused by fatigue. The model update unit is used to update the parameters of the initial motion imagery decoding model based on the adaptively reconstructed features, and generate a fatigue state motion imagery decoding model. A decoding strategy switching unit is used to smoothly switch decoding strategies between the initial motor imagery decoding model and the fatigued state motor imagery decoding model based on the physiological fatigue index. When the physiological fatigue index is less than T1, the output of the initial motor imagery decoding model is directly used as the control command. When the physiological fatigue index is greater than or equal to T1, a transition phase begins. The decoding results of the initial motion imagery decoding model and the decoding results of the fatigued state motion imagery decoding model are weighted and fused to generate the final control command. The weight of the decoding result of the initial motion imagery decoding model is negatively correlated with the physiological fatigue index, while the weight of the decoding result of the fatigued state motion imagery decoding model is positively correlated with the physiological fatigue index, so as to achieve a smooth transition from the initial model to the adaptive model in the decoding process.

[0011] Preferably, the model update unit is configured as follows: Online acquisition of users' current motor imagery EEG data and corresponding motor intention tags; Based on the current physiological fatigue index and EEG characteristics, the most matching fatigue pattern is retrieved from the pre-stored fatigue feature database, and the pre-fine-tuned model parameters corresponding to that pattern are called as the initial parameters for model updates. Using EEG samples with motor imagery category labels in a non-fatigue state as source domain data, and EEG samples in a fatigue state determined by task prompts during calibration tasks as labeled target domain data, or EEG samples in a fatigue state that are not labeled during normal operation as unlabeled target domain data, the model parameters are optimized and updated through a domain adaptation algorithm to obtain a feature representation robust to fatigue states. Using newly collected batch sample data, the fatigue state motion imagery decoding model is incrementally updated using the gradient descent algorithm; The updated model is evaluated using validation set data, and it is officially activated only when the performance metrics meet preset conditions.

[0012] Preferably, the EEG decoding and feature adaptive reconstruction module further includes a real-time monitoring and calibration unit, which is configured as follows: Continuously monitor the online classification accuracy and / or decoding confidence of the fatigue state motion imagery decoding model; the online classification accuracy is determined based on the consistency between the control commands output by the fatigue state motion imagery decoding model and the user's actual motion intention, and the decoding confidence is determined based on the entropy value or maximum probability value of the probability distribution output by the fatigue state motion imagery decoding model. When it is detected that the online classification accuracy is continuously lower than the preset accuracy threshold within a preset time period, or the decoding confidence is continuously lower than the preset confidence threshold, a recalibration trigger signal is generated. Based on the recalibration trigger signal, the human-computer interaction and feedback module is controlled to guide the user to perform a preset calibration task, and the fatigue state motion imagery decoding model is retrained or fine-tuned based on the data collected by the calibration task to restore decoding performance.

[0013] Preferably, the auxiliary force compensation module includes: The input interface unit is used to receive the physiological fatigue index output by the physiological fatigue monitoring module in real time. The mapping calculation unit is used to map the physiological fatigue index to the corresponding auxiliary force output ratio according to a preset nonlinear mapping function. The instruction conversion unit is used to convert the auxiliary force output ratio into a motor torque command or impedance parameter adjustment amount for the exoskeleton execution module to execute; The safety limiting unit is used to perform a safety limiting check on the motor torque command or impedance parameter adjustment amount before the motor torque command or impedance parameter adjustment amount is output to the exoskeleton execution module, so as to ensure that it does not exceed a preset safety threshold.

[0014] Preferably, the nonlinear mapping function is a piecewise nonlinear function: When the physiological fatigue index is less than the first threshold T1, the auxiliary force output ratio is the minimum auxiliary force A. min ; When the physiological fatigue index is greater than or equal to T1 and less than the second threshold T2, the auxiliary force output ratio is at the minimum auxiliary force A. min It monotonically increases with respect to the first auxiliary force threshold. When the physiological fatigue index is greater than or equal to T2 and less than the third threshold T3, the auxiliary force output ratio increases monotonically between the first auxiliary force threshold and the second auxiliary force threshold at a rate greater than that of the first mapping interval. When the physiological fatigue index is greater than or equal to T3, the auxiliary force output ratio reaches the maximum auxiliary force A. max .

[0015] A second aspect of the present invention also provides a control method employing the system described above, comprising: The system simultaneously acquires the user's motor imagery EEG signals and multimodal physiological signals of physiological fatigue state. The acquired raw signals are preprocessed by denoising, filtering, and baseline drift correction. The multimodal physiological signals include at least heart rate variability-related signals, skin conductance signals, and EEG frequency band signals. Feature extraction is performed on the preprocessed multimodal physiological signals. The extracted multidimensional features are then fused and the physiological fatigue index is calculated in real time using a pre-trained machine learning model. The user's current fatigue level is determined based on the physiological fatigue index and preset multi-level fatigue thresholds. The system receives the motor imagery EEG signal and adaptively adjusts the EEG decoding model based on the physiological fatigue index to output control commands that conform to the user's motor intentions. The exoskeleton's auxiliary force output parameters are dynamically calculated based on the physiological fatigue index according to a preset mapping relationship. The lower limb exoskeleton is driven to perform corresponding movements according to the control commands and the auxiliary force output parameters; Provide users with real-time feedback and guidance information based on the physiological fatigue index and / or the fatigue level.

[0016] A third aspect of the present invention also provides an electronic device, comprising: The memory is used to store the processing program; The processor, when executing the processing program, implements the physiological fatigue adaptive control method for the motor imagery brain-computer interface exoskeleton as described above. Compared with the prior art, the present invention has the following beneficial effects: 1) Improve the effectiveness of long-term rehabilitation training: Through fatigue monitoring by multimodal physiological signal fusion and adaptive adjustment of EEG decoding model, the problem of decreased decoding accuracy caused by user fatigue is effectively solved, ensuring the control stability of long-term training.

[0017] 2) Improve user comfort and compliance: The intelligent assistive force compensation mechanism can dynamically reduce the burden on users according to their fatigue level. The higher the fatigue level, the greater the assistive force, thereby improving the comfort of training and enhancing users' compliance with long-term rehabilitation training.

[0018] 3) Enhanced system robustness and safety: The system can sense changes in the user's physiological state in real time and make corresponding adjustments, avoiding safety hazards caused by misoperation under fatigue conditions and improving the overall robustness of the system. The safety limiting unit monitors the exoskeleton's output torque and motor current in real time to ensure that the auxiliary force output does not exceed the safety threshold.

[0019] 4) Personalized Rehabilitation Training: By combining user physiological fatigue data, the system can provide more personalized training plans and auxiliary strategies to optimize rehabilitation outcomes. Multi-level fatigue thresholds are determined using a combination of personalized calibration and group statistical data to achieve individualized adaptive control. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall architecture of the physiological fatigue adaptive control system of the motor imagination brain-computer interface exoskeleton in an embodiment of the present invention; Figure 2 This is a flowchart of the physiological fatigue monitoring module in an embodiment of the present invention; Figure 3 This is a flowchart of the EEG decoding and feature adaptive reconstruction module in an embodiment of the present invention; Figure 4 This is a flowchart of the auxiliary force compensation module in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, this embodiment provides a physiological fatigue adaptive control system for a motor imagery brain-computer interface exoskeleton, including: The multi-physiological signal acquisition module is used to simultaneously acquire the user's motor imagery EEG signals and multimodal physiological signals of physiological fatigue state. The multimodal physiological signals include at least heart rate variability (HRV) related signals, skin conductance (GSR) signals, and EEG frequency band signals. The physiological fatigue monitoring module is connected to the multi-physiological signal acquisition module and is used to extract features from the multi-modal physiological signals and calculate the physiological fatigue index and corresponding fatigue level of the target user based on the extracted multi-dimensional features. The EEG decoding and feature adaptive reconstruction module is connected to the multi-physiological signal acquisition module and the physiological fatigue monitoring module, respectively. It is used to receive the motor imagery EEG signal and adaptively adjust the EEG decoding model according to the physiological fatigue index. When the model completes initialization and passes verification and the decoding confidence reaches the preset condition, it outputs a motor control command; otherwise, it outputs a safety control command to maintain, decelerate, or pause execution. An auxiliary force compensation module, connected to the physiological fatigue monitoring module, is used to dynamically calculate the exoskeleton auxiliary force output parameters according to the physiological fatigue index and a preset mapping relationship. The exoskeleton execution module is connected to the EEG decoding and feature adaptive reconstruction module and the auxiliary force compensation module, respectively, and is used to receive the control commands and the auxiliary force output parameters, and drive the lower limb exoskeleton to perform corresponding movements according to the control commands and the auxiliary force output parameters; The human-computer interaction and feedback module is connected to the physiological fatigue monitoring module and the auxiliary force compensation module, and is used to provide real-time feedback and guidance to the user based on the physiological fatigue index and / or the fatigue level.

[0023] This embodiment establishes a closed-loop control system based on multimodal physiological signal fusion monitoring technology, encompassing adaptive reconstruction of EEG decoding models and dynamic compensation of exoskeleton-assisted forces. Its working principle first involves synchronously acquiring motor imagery EEG signals and multimodal physiological signals through a multi-physiological signal acquisition module. Then, a physiological fatigue monitoring module performs multi-dimensional feature extraction and fusion analysis to calculate the fatigue index F, which characterizes the user's physiological load state, in real time. I The fatigue index, as the core state variable of the system, directly drives two parallel adaptive control branches.

[0024] The first branch focuses on the online adaptive reconstruction of the EEG decoding model. When F is detected... I An increase in the signal indicates that the user is entering a state of fatigue. The EEG decoding and feature adaptive reconstruction module activates an adaptive adjustment mechanism based on the fatigue state, compensating for the non-stationary drift of the EEG signal caused by fatigue through feature mapping reconstruction or model parameter updates. This process ensures that even when the user's physiological state changes, the system can still accurately interpret motor intentions and maintain the stability and accuracy of control commands.

[0025] The second branch focuses on the dynamic compensation of the auxiliary force output. The auxiliary force compensation module is based on F I The nonlinear mapping relationship between the assist force and the exoskeleton's assist torque parameters is calculated in real time. This mechanism makes the assist force level positively correlated with the user's fatigue level, that is, providing mechanical assistance to compensate for insufficient active muscle strength when fatigue deepens, and reducing the assist ratio to maintain active training intensity when the user is in good condition.

[0026] The exoskeleton execution module integrates the motion control commands and assistive force parameters output from the aforementioned branches, driving the lower limb exoskeleton to perform the expected movements. Simultaneously, the human-computer interaction and feedback module outputs status prompts and training guidance. Through this architecture, the system achieves coordinated adaptive adjustment between the brain-computer interface control layer and the physical execution layer, effectively solving the problems of decreased decoding performance and limited motor ability caused by physiological fatigue during long-term rehabilitation training. It achieves dynamic synchronization between the exoskeleton system and the user's physiological state, significantly improving the safety, continuity, and effectiveness of rehabilitation training.

[0027] In this embodiment, the multi-physiological signal acquisition module is the front-end unit for the system to realize human-machine state perception. Its main function is to simultaneously acquire the user's motor imagery EEG signals and multimodal physiological fatigue signals. This module acquires scalp EEG signals through a high-density EEG electrode cap. These signals are not only used for decoding motor intentions but also serve as a data source for calculating specific EEG frequency band ratios (such as the (Theta+Alpha) / Beta ratio). To comprehensively assess the user's physiological load, the module integrates a set of physiological sensors, including an electrocardiogram (ECG) or photoplethysmography (PPG) sensor for acquiring heart rate variability (HRV) data, and a skin conductance (GSR) sensor for reflecting the level of autonomic nervous system activity. All of the above signals are synchronously acquired and transmitted at a sampling rate of not less than 250Hz to ensure strict alignment of EEG signals, ECG signals, and skin conductance signals on the time axis, providing high-precision multidimensional data support for subsequent physiological fatigue monitoring and adaptive adjustment of the EEG decoding model.

[0028] like Figure 2 As shown, the physiological fatigue monitoring module includes: a feature extraction unit, a fatigue index calculation unit, a fatigue level determination unit, and a threshold determination branch unit.

[0029] The feature extraction unit is used to extract fatigue-related features from the multimodal physiological signals. The fatigue-related features include at least two of the following: HRV time-domain features, HRV frequency-domain features, GSR features, and EEG frequency band power features.

[0030] Specifically, the feature extraction unit includes an EEG frequency band feature extraction subunit, which performs bandpass filtering (0.5-45Hz), notch filtering (50 / 60Hz), and artifact removal (such as independent component analysis (ICA) or regression analysis to remove electrooculography / electromyography artifacts) on the EEG frequency band signal. It extracts EEG segments through a preset first sliding window (e.g., 5 seconds, step size 1 second), performs time-frequency transformation (e.g., fast Fourier transform, FFT) on each segment to calculate the power spectral density (PSD) of the Delta band (0.5-4Hz), Theta band (4-8Hz), Alpha band (8-13Hz), and Beta band (13-30Hz), and calculates the (Theta+Alpha) / Beta ratio and / or Alpha / Beta ratio as the power features of the EEG frequency band. As fatigue levels increase, the power in the Theta and Alpha bands increases relatively, while the power in the Beta band decreases relatively. Therefore, the (Theta+Alpha) / Beta ratio can effectively reflect changes in fatigue status.

[0031] The heart rate variability feature extraction subunit is used to detect R waves or PPG peaks from the heart rate variability related signals (ECG or PPG signals) to calculate the RR interval sequence, perform artifact correction and interpolation on the RR interval sequence, and calculate the time-domain indices SDNN and / or RMSSD as HRV time-domain features within a preset second sliding window (e.g., 5 minutes), as well as the frequency-domain indices LF (low-frequency power, 0.04-0.15Hz), HF (high-frequency power, 0.15-0.4Hz), and / or LF / HF ratio as HRV frequency-domain features. Under fatigue conditions, sympathetic nerve activity is enhanced and parasympathetic nerve activity is weakened, which is manifested by an increased LF / HF ratio and a decreased RMSSD. Therefore, the above-mentioned HRV features can effectively reflect the fatigue response of the autonomic nervous system.

[0032] The skin conductance feature extraction subunit is used to perform low-pass filtering (e.g., 1Hz) on the skin conductance signal to remove high-frequency noise, extract GSR segments through a preset third sliding window (e.g., 30 seconds, step size 5 seconds), and calculate the amplitude and frequency of the average skin conductance level (SCL) and skin conductance response (SCR) of each segment as GSR features. Skin conductance reflects the activation level of the sympathetic nervous system. In a state of fatigue, sympathetic nerve activity increases, manifested as an increase in skin conductance level and an increase in SCR frequency.

[0033] Fatigue index calculation unit, configured to fuse the fatigue-related features, and calculate a physiological fatigue index representing the user's current physiological load level through a pre-trained machine learning model based on the fused features. Specifically, the fatigue index calculation unit splices the HRV time-domain feature vector, HRV frequency-domain feature vector, GSR feature vector and EEG band power feature vector extracted by the feature extraction unit into a unified fused feature vector, inputs the pre-trained machine learning model, and outputs a normalized physiological fatigue index F I , the physiological fatigue index F I has a value range of 0-100, where 0 indicates fully awake and 100 indicates extremely fatigued. The machine learning model is a Support Vector Regression (SVR) model or a Multi-Layer Perceptron (MLP) neural network. The SVR model adopts a Radial Basis Function (RBF) kernel function, and optimizes parameters C and gamma through grid search and cross validation. The MLP neural network includes an input layer, 2-3 hidden layers and an output layer, each hidden layer contains 32-64 neurons and adopts a ReLU activation function, and the output layer is a single neuron, which is trained using an Adam optimizer and Mean Squared Error (MSE) as the loss function. The training datasets of both models include multi-physiological signal data of at least 50 subjects under different fatigue levels, which are labeled through a subjective fatigue scale (such as the Visual Analogue Scale VAS). The SVR or MLP model can automatically learn the complex nonlinear relationship between different physiological features and perform weighting according to their contribution to fatigue, so as to output a comprehensive and robust physiological fatigue index F I .

[0034] Fatigue level determination unit, configured to compare the physiological fatigue index with preset multi-level fatigue thresholds to determine the current fatigue level of the user. In this embodiment, a multi-level threshold mechanism is adopted. A first threshold T1, a second threshold T2 and a third threshold T3 are preset, and T1<T2<T3. Exemplarily, T1=60 is the mild fatigue threshold, T2=75 is the moderate fatigue threshold, and T3=90 is the severe fatigue threshold. The fatigue levels are divided as follows: non-fatigue state: F I <T1; mild fatigue state: T1≤F I <T2; moderate fatigue state: T2≤F I <T3; severe fatigue state: F I≥T3. The multi-level fatigue thresholds are determined by a method combining personalized calibration and population statistical data. The initial thresholds are set based on the statistical results of physiological fatigue data of large-scale populations. For each new user, the system conducts a baseline test (non-fatigue state) and a fatigue induction experiment (simulated fatigue state), collects the user's baseline data in the non-fatigue state, induces the user to generate different degrees of fatigue and synchronously collects physiological data and subjective fatigue scores, combines the collected user personalized data with the population physiological fatigue statistical data, and uses machine learning algorithms (such as decision trees or logistic regression) to train or fine-tune a personalized fatigue classifier or regression model to determine the personalized values of each threshold.

[0035] A threshold judgment branch unit, configured to generate a corresponding control signal according to the fatigue level and output it to the corresponding module respectively: when the fatigue level is a non-fatigue level, that is, F I <T1, the system maintains a normal monitoring state, continues to cyclically collect and process physiological signals to update F in real time I , controls the EEG decoding and feature adaptive reconstruction module to call the initial motor imagery decoding model established based on the user's non-fatigue state EEG data to decode the motor imagery EEG signal, and the assistance force compensation module keeps the assistance force at the baseline level. When the fatigue level is a fatigue level, that is, F I ≥T1, controls the EEG decoding and feature adaptive reconstruction module to adaptively adjust the initial motor imagery decoding model according to the current fatigue level and current EEG features to obtain a motor imagery decoding model for the fatigue state, and controls it to generate a motion control instruction by using the motor imagery decoding model for the fatigue state. That is, when entering the fatigue state, the system triggers three parallel control paths: First path: Send a model adaptive trigger signal to the EEG decoding and feature adaptive reconstruction module, control it to adaptively adjust the initial motor imagery decoding model according to the current fatigue level and current EEG features to obtain the motor imagery decoding model for the fatigue state, and generate a motion control instruction by using the motor imagery decoding model for the fatigue state. Second path: Send the current physiological fatigue index F to the assistance force compensation module I , trigger dynamic assistance force compensation, and the higher the fatigue degree, the greater the assistance force. Third path: According to the fatigue level where F is located I , send fatigue state information to the human-computer interaction and feedback module to generate corresponding visual, auditory or tactile prompts and guidance information (for example, prompt "Please pay attention to adjustment" when in mild fatigue, prompt "It is recommended to rest and adjust breathing rhythm" when in moderate fatigue, prompt "It is recommended to suspend training" when in severe fatigue). At the same time, the threshold judgment branch unit also records key information such as the current fatigue state and system adjustment parameters into the database for subsequent analysis and optimization of personalized rehabilitation schemes.

[0036] For example Figure 3As shown, the EEG decoding and feature adaptive reconstruction module includes: an initial decoding unit, a feature adaptive reconstruction unit, a model update unit, a decoding strategy switching unit, and a real-time monitoring and calibration unit.

[0037] The initial decoding unit stores the initial motor imagery decoding model trained on EEG data from a user's non-fatigue state. In a non-fatigue state, EEG signal characteristics are stable and the signal-to-noise ratio is high, resulting in a decoding model with high classification accuracy and reliability. The initial decoding model can employ traditional machine learning methods such as CSP (Common Spatial Patterns) + SVM (Support Vector Machine), or deep learning models such as EEGNet. The training data includes various motor imagery categories (e.g., left-hand, right-hand, and bipedal motor imagery). This model can accurately decode the user's motor intentions in a non-fatigue state, providing a stable and reliable control command benchmark for the system.

[0038] The feature adaptive reconstruction unit is used to adaptively reconstruct the current EEG features when the received physiological fatigue index reaches or exceeds a first threshold T1, in order to compensate for the feature distribution drift caused by fatigue. During long-term rehabilitation training, user physiological fatigue leads to changes in the spectral characteristics and temporal morphology of EEG signals, i.e., feature distribution drift. If the initial decoding model continues to be used to process the drifted features, the decoding accuracy will significantly decrease. To solve this problem, the feature adaptive reconstruction unit performs domain adaptive reconstruction of the current EEG features, aligning the feature distribution under fatigue state with the feature distribution under non-fatigue state, thereby eliminating the influence of fatigue on the feature space.

[0039] The feature adaptive reconstruction unit comprises a fatigue feature library, a feature matching subunit, and a domain adaptation training subunit. The fatigue feature library pre-stores typical EEG feature patterns corresponding to multiple fatigue levels, along with their corresponding decoding model fine-tuning parameters. This feature library is constructed by extracting features and performing cluster analysis on EEG data from a large number of subjects at different fatigue levels (mild, moderate, and severe). Each cluster center represents a typical fatigue state feature pattern. The fatigue feature library contains typical EEG feature vectors (such as frequency domain power spectrum and time domain waveform features) for each fatigue level, along with their corresponding decoding model fine-tuning parameters or weights. By pre-constructing a typical feature library covering multiple fatigue patterns, the system can quickly match the closest reference pattern when a user experiences fatigue, eliminating the need for model training from scratch, significantly shortening the adaptive adjustment time, and achieving rapid response under fatigue conditions.

[0040] The feature matching subunit is used to calculate the similarity (such as cosine similarity or Euclidean distance) between the current EEG features and each feature pattern in the fatigue feature library when the user enters a fatigue state. It selects the feature pattern with the highest similarity as the matching result and loads the pre-tuned model parameters corresponding to that pattern as the starting point for adaptive adjustment. Through this similarity-based matching mechanism, the type of the user's current fatigue features can be accurately identified, providing precise initial parameters for subsequent domain adaptation training and improving the efficiency and accuracy of model optimization.

[0041] The domain adaptation training subunit uses EEG samples labeled with motor imagery categories from a user's non-fatigue state as source domain data, and EEG samples from a fatigued state determined by task prompts during calibration as labeled target domain data, or EEG samples from a fatigued state during normal operation as unlabeled target domain data. Domain adaptation algorithms (such as Domain Adversarial Neural Networks (DANN) or Maximum Mean Difference (MMD)) are used to optimize and update model parameters, reducing the feature distribution differences between the source and target domains. This allows the fatigued motor imagery decoding model to learn robust feature representations under fatigued conditions. DANN introduces adversarial training between a domain classifier and a feature extractor, enabling the feature extractor to learn domain-invariant feature representations. MMD transfers knowledge from the source domain to the target domain by minimizing the MMD distance between the feature distributions of the source and target domains. Through domain adaptation training, the model can effectively overcome the data distribution differences between the source (non-fatigue) and target (fatigue) domains, maintaining high decoding accuracy even under fatigued conditions.

[0042] The model update unit is used to update the parameters of the initial motion imagery decoding model based on the adaptively reconstructed features, generating a fatigue state motion imagery decoding model. Specifically, the model update unit is configured as follows: Online data acquisition: Acquire the user's current motor imagery EEG data and corresponding motor intention tags online (through exoskeleton sensors or user feedback); Feature library matching and parameter initialization: Based on the current physiological fatigue index and EEG characteristics, the most matching fatigue pattern is retrieved from the pre-stored fatigue feature library, and the corresponding pre-tuned model parameters are used as the initial parameters for model updates. Parameter initialization using the pre-stored fatigue feature library provides a good starting point for subsequent domain adaptation training, reduces the number of training iterations, and accelerates model convergence.

[0043] Domain adaptation training: EEG samples with motor imagery category labels in the user's non-fatigued state are used as source domain data, and fatigued state EEG samples determined by task prompts during the calibration task are used as labeled target domain data, or unlabeled fatigued state EEG samples during normal operation are used as unlabeled target domain data. Model parameters are optimized and updated through a domain adaptation algorithm to obtain a feature representation robust to the fatigued state. This step enables the model to effectively adapt to the individual's fatigue feature pattern and overcome performance degradation caused by cross-state data distribution differences. The adopted domain adaptation algorithms include Domain-Adversarial Neural Network (DANN) or Maximum Mean Discrepancy (MMD). DANN introduces a domain classifier and a feature extractor for adversarial training to learn domain-invariant feature representations; MMD minimizes the distance between the feature distributions of the source domain and the target domain to obtain a feature representation robust to the fatigued state.

[0044] Online incremental update: The collected new batch of sample data is used to perform incremental update on the fatigued state motor imagery decoding model through a gradient descent algorithm (e.g., Mini-batch Gradient Descent). Since the user's fatigued state will dynamically change with the extension of training time, a single domain adaptation training cannot maintain the optimal performance of the model for a long time. Through the online incremental update mechanism, the system can continuously track the changes of the user's fatigued state throughout the rehabilitation training process and dynamically adjust the model parameters. The update can be triggered in an event-driven manner (e.g., every 10-20 valid fatigue samples are collected) or a time-driven manner (e.g., every 5 minutes) to ensure the real-time adaptability of the model.

[0045] Model validation: The performance of the updated model is evaluated using validation set data (e.g., recently collected fatigued state data not used for training), including indicators such as decoding accuracy and Kappa coefficient. Although online incremental update can continuously optimize the model, there is a risk of model performance degradation caused by noise samples or abnormal data. By performing strict performance evaluation after each update, the updated model is officially enabled only when the performance indicators meet the preset conditions, otherwise the current model is maintained, so as to ensure that the decoding performance does not deteriorate due to blind updates.

[0046] A decoding strategy switching unit, configured to smoothly switch the decoding strategy between the initial motor imagery decoding model and the fatigued state motor imagery decoding model according to the physiological fatigue index: When the physiological fatigue index is less than T1 (F I <T1), the output of the initial motor imagery decoding model is directly used as the control instruction to obtain the optimal decoding effect.

[0047] When the physiological fatigue index is greater than or equal to T1 (F IWhen the fatigue level reaches ≥T1, the system enters a fatigue state. However, at this time, the motion imagery decoding model for the fatigue state is still in its early stages of development. Direct switching may lead to fluctuations in decoding results due to the model not being fully adapted. To avoid user discomfort or misoperation caused by sudden changes in control commands, the system enters a transition phase. A weighted fusion calculation is performed on the decoding results of the initial motion imagery decoding model and the fatigue state motion imagery decoding model to generate the final control command. The weight of the initial motion imagery decoding model's decoding result is negatively correlated with the physiological fatigue index, while the weight of the fatigue state motion imagery decoding model's decoding result is positively correlated with the physiological fatigue index. This achieves a smooth transition from the initial model to the adaptive model during the decoding process. Specifically, in the initial switching phase, the initial model has a higher weight and the adaptive model has a lower weight. As the fatigue level increases, the initial model's weight gradually decreases, while the adaptive model's weight gradually increases. Through this weighted fusion mechanism, a smooth transition from the initial model to the adaptive model is achieved, avoiding sudden changes in control commands and user discomfort.

[0048] In this embodiment, the EEG decoding and feature adaptive reconstruction module also includes a real-time monitoring and calibration unit, which is configured to continuously monitor the decoding performance of the fatigue state motor imagery decoding model and trigger recalibration when the performance declines.

[0049] Because users' physiological and psychological states, as well as environmental factors, can change unpredictably, even models that have undergone domain adaptation and incremental updates may experience performance degradation in subsequent use. Continuous monitoring of the online classification accuracy and / or decoding confidence of the fatigue state motion imagery decoding model is essential. The online classification accuracy is determined based on the consistency between the control commands output by the fatigue state motion imagery decoding model and the user's actual motion intentions. The decoding confidence is determined based on the entropy value or maximum probability value of the probability distribution output by the fatigue state motion imagery decoding model; low confidence or high entropy values ​​indicate a potential decrease in decoding accuracy.

[0050] When it is detected that the online classification accuracy continuously falls below a preset accuracy threshold (e.g., 70%) within a preset duration (e.g., 30 seconds), or the decoding confidence continuously falls below a preset confidence threshold, it indicates that the decoding performance of the current model can no longer meet the system requirements. At this time, the system generates a recalibration trigger signal, and according to the recalibration trigger signal, controls the human-computer interaction and feedback module to guide the user to perform a preset calibration task (e.g., repeating motor imagery several times), and retrains or fine-tunes the fatigue state motor imagery decoding model based on the data collected from the calibration task, so as to restore the decoding performance. Through this real-time monitoring and recalibration mechanism, the system can actively intervene and restore when the decoding performance degrades, constructing a complete performance guarantee closed loop of monitoring-warning-recovery, which ensures the continuous and stable operation of the system in long-term rehabilitation training.

[0051] As Figure 4 shown, the assistance force compensation module is configured to dynamically adjust the assistance force output parameter of the lower limb exoskeleton according to the physiological fatigue index output by the physiological fatigue monitoring module. By establishing the association between the physiological state and the mechanical assistance, this module realizes the coordinated operation of the human-machine organism. Specifically, the assistance force compensation module includes an input interface unit, a mapping calculation unit, an instruction conversion unit and a safety amplitude limiting unit.

[0052] The input interface unit is configured to receive in real time the physiological fatigue index F output by the physiological fatigue monitoring module I , as the front-end input of the control loop.

[0053] The mapping calculation unit is configured to, according to a preset nonlinear mapping function, convert the physiological fatigue index F I to the corresponding assistance force output ratio A(F I ). Since the user's fatigue accumulation presents nonlinear characteristics, the assistance demand increases gently during mild fatigue, while the assistance demand rises sharply from moderate to severe fatigue. If linear mapping is adopted, the assistance force will be too large during mild fatigue (interfering with independent force generation) or the assistance force will be insufficient during severe fatigue (failing to effectively reduce the load). Therefore, this embodiment adopts a piecewise nonlinear mapping function to more precisely match the assistance demands under different fatigue degrees, so that the change of the assistance force matches the actual physiological demand of the user, avoiding premature intervention affecting the rehabilitation effect or delayed intervention leading to potential safety hazards. The piecewise nonlinear mapping function is specifically as follows: When the physiological fatigue index is less than the first threshold T1, the assistance force output ratio is the minimum assistance force A min , that is, when F I <T1, A(F I )=A min(minimum assisting force), at this time the user is in a non-fatigued state, the EEG signal is of good quality, the active movement ability is strong, the exoskeleton does not provide additional assistance, and the user completes the training entirely by autonomous force, so as to maximize the active rehabilitation effect.

[0054] When the physiological fatigue index is greater than or equal to T1 and less than the second threshold T2, that is, T1≤F I <T2, the assisting force output ratio A(F I ) increases monotonically between the minimum assisting force A min and the first assisting force threshold, the user enters a mild fatigue state, and the active movement ability begins to decline. At this time, the exoskeleton starts to intervene to provide slight assistance, so as to slightly reduce the user's burden, while not significantly weakening the stimulating effect of autonomous force exertion on nerve remodeling.

[0055] When the physiological fatigue index is greater than or equal to T2 and less than the third threshold T3, that is, T2≤F I <T3, the assisting force output ratio A(F I ) increases monotonically between the first assisting force threshold and the second assisting force threshold at a rate of change greater than that of the first mapping interval. The user enters a moderate fatigue state, and the movement ability decreases significantly. At this time, the assisting force of the exoskeleton increases significantly, so as to greatly reduce the user's active force exertion requirement, ensure that the training can continue, and avoid movement deformation or compensation caused by excessive fatigue.

[0056] When the physiological fatigue index F I is greater than or equal to T3, the assisting force output ratio reaches the maximum assisting force A max , that is, when F I ≥T3, A(F I )=A max (maximum assisting force). The user enters a severe fatigue state, the movement ability is severely reduced, and the exoskeleton provides the maximum assisting force, which almost completely bears the movement load, ensures the user's safety and guides the user to stop training in due time.

[0057] In this embodiment, the minimum assisting force A min is 0%, and the maximum assisting force A max is 80%. The first assisting force threshold is 30%, and the second assisting force threshold is 60%. The above values ensure that the assisting force increases slightly between 0% and 30% in the mild fatigue stage, increases at an accelerated rate between 30% and 60% in the moderate fatigue stage, and reaches the maximum value of 80% in the severe fatigue stage.

[0058] In another embodiment, the nonlinear mapping function is a Sigmoid function: Wherein, k is a slope parameter, which controls the steepness of the curve; F midThe fatigue index is set at a midpoint (e.g., 50). This Sigmoid function also achieves a mapping characteristic where the assist force increases steadily at mild fatigue, accelerates at moderate fatigue, and tends towards saturation at severe fatigue, enabling precise matching of assist needs under different fatigue levels. Through this nonlinear mapping mechanism, the assist force transitions across the entire fatigue range from a low-sensitivity zone (mild fatigue, where the assist force increases gradually to avoid premature intervention and interference with the autonomous exertion pattern), through a high-sensitivity zone (moderate fatigue, where the assist force increases significantly to quickly respond to the surge in assist needs), to a saturation zone (severe fatigue, where the assist force converges to its upper limit, providing amplitude limiting protection). This allows the exoskeleton's assist output to dynamically balance with the user's real-time physiological load level, effectively avoiding both insufficient assistance (in a fatigued state, the user's active muscle strength is insufficient to maintain effective movement) and excessive assistance (in a non-fatigued state, the exoskeleton compensates by weakening the stimulation of the central nervous system by active movement). It enables precise and continuous adjustment from zero assistance to maximum assistance across the entire fatigue range.

[0059] The instruction conversion unit converts the auxiliary force output ratio into motor torque commands or impedance parameter adjustments for the exoskeleton execution module. The auxiliary force output ratio is a normalized dimensionless value (0%-80%), while the exoskeleton's motor drivers require specific physical control quantities to execute. This unit converts the abstract auxiliary force ratio into concrete motor torque commands or impedance parameter adjustments, enabling the exoskeleton execution module to output appropriate auxiliary torque (such as auxiliary torque for hip and knee flexion and extension) according to the user's movement intentions. Specifically, the motor torque commands increase joint output torque to directly assist movement, while the impedance parameter adjustments regulate the exoskeleton's response sensitivity to user actuation.

[0060] A safety limiting unit is used to perform a safety limiting check on the motor torque command or impedance parameter adjustment before it is output to the exoskeleton execution module, ensuring that it does not exceed a preset safety threshold. During auxiliary force compensation, if the auxiliary force command exceeds the safe range due to sensor failure, signal interference, or system anomalies, it may cause overload damage to the exoskeleton or even pose a safety threat to the user. Therefore, the safety limiting unit monitors the exoskeleton's output torque and motor current in real time. When it detects that the output torque or motor current exceeds the preset safety threshold, it automatically limits the output value of the motor torque command or impedance parameter adjustment and switches to a safe mode. Through this dual safety mechanism, the auxiliary force output is ensured to always operate within a safe range, preventing overload or injury to the user.

[0061] Furthermore, the safety limiting unit is also equipped with an emergency stop mechanism and a fall detection mechanism. This unit is connected to an emergency stop button and a fall detection module (based on multi-axis IMU data) located on the exoskeleton. When the user presses the emergency stop button, or the fall detection module determines that the user has fallen based on acceleration and angular velocity data, or detects that the motor current or output torque exceeds a preset safety threshold, the safety limiting unit immediately cuts off the output of the auxiliary force command, controls the exoskeleton to switch to a safety stop mode, locks the joints, or switches to a high-damping state to prevent secondary injury to the user.

[0062] In one embodiment, the exoskeleton execution module includes a motor driver, a high-performance brushless DC motor, a high-precision encoder, a torque sensor, and a multi-axis inertial measurement unit (IMU). The motor driver receives motor torque commands or impedance parameter adjustments from the command conversion unit and performs precise position or torque control on the motor output through a built-in PID controller. The high-precision encoder and torque sensor collect the angle information and actual output torque of the exoskeleton joints in real time and feed them back to the controller to form a closed-loop control circuit, ensuring that the exoskeleton can smoothly and safely execute movement intentions and provide the necessary auxiliary force. In addition, the module also includes hardware limit switches to trigger a forced stop when the motor movement exceeds the physical safety range, providing underlying hardware safety protection.

[0063] In one embodiment, the human-computer interaction and feedback module may include a display screen, a vibration motor, and a speaker. The display screen (such as a small LCD screen) uses a Qt or Web technology-based interface to visually present the physiological fatigue index, current assistive force ratio, fatigue level, and training progress. When the system determines that the user has entered a moderate or severe fatigue state, the display screen shows a breathing rhythm guidance animation to assist the user in adjusting their breathing. The vibration motor is integrated into the exoskeleton straps and provides tactile alerts when the fatigue level reaches a preset threshold. The speaker is used to play voice prompts (such as "You have entered a moderate fatigue state, rest is recommended") or warning sounds. Furthermore, this module is also responsible for recording the user's fatigue data, training performance, and system adjustment parameters to a local database for rehabilitation therapists to review and use for optimizing subsequent personalized rehabilitation plans.

[0064] Examples of rehabilitation training scenarios To more clearly illustrate the working process of the assistive force compensation module in this embodiment, the following description is based on a typical rehabilitation training scenario: Scenario 1: In the early stages of training, the user is full of energy. The system detects F I A low value (e.g., 30) is considered a non-fatigue state. The auxiliary force compensation module 400 calculates A(F). I The exoskeleton provides no additional assistance and is fully under the user's control to maximize the effectiveness of active rehabilitation.

[0065] Scenario 2: After 30 minutes of training, the user feels slightly fatigued. The system detects F I The value rose to 65, indicating a state of mild fatigue. The auxiliary force compensation module 400 calculated A(F) based on a piecewise nonlinear mapping function. I The exoskeleton begins to provide 20% of the assistive force, slightly reducing the user's burden. At the same time, the human-computer interaction module 600 prompts "You have entered a state of mild fatigue, please adjust accordingly."

[0066] Scenario 3: After one hour of training, the user is clearly fatigued. The system detects F. I The value rises to 80, indicating a moderate fatigue state. The auxiliary force compensation module calculates A(F). I The exoskeleton provides 50% of the assist force, significantly reducing the user's need for active exertion. The human-computer interaction module issues a voice prompt, "You have entered a state of moderate fatigue. We recommend resting for 5 minutes and adjusting your breathing rhythm," while the display screen shows a breathing guidance animation.

[0067] Scenario 4: The user continues training, reaching a high level of fatigue. The system detects F. I The value rises to 95, indicating a state of severe fatigue. The auxiliary force compensation module calculates A(F). I The exoskeleton provides maximum assist force, almost entirely bearing the exercise load. The system triggers safety mode, recommending a forced pause in training and recording fatigue data for future rehabilitation adjustments.

[0068] Through the aforementioned dynamic and adaptive adjustment mechanism, the assistive force compensation module in this embodiment can keep the exoskeleton synchronized with the user's physiological state at all times, which can maximize the rehabilitation training effect while ensuring user comfort and safety.

[0069] In one embodiment of the system hardware implementation, the multi-physiological signal acquisition module, physiological fatigue monitoring module, EEG decoding and feature adaptive reconstruction module, assistive force compensation module, and human-computer interaction and feedback module of this embodiment can be integrated into an embedded processing unit, for example, using an NVIDIA Jetson AGX Xavier as the core processor. A signal acquisition card or wireless transmission module is used to synchronously transmit data acquired by the EEG electrode cap, ECG / PPG sensor, and GSR sensor to this processing unit. At the software level, signal preprocessing (such as filtering and artifact removal) and feature extraction algorithms can be implemented using Python language combined with the MNE-Python library and the SciPy library; the machine learning fatigue prediction model (SVR or MLP) and deep learning EEG decoding model (such as EEGNet, DANN) can be trained and deployed using the TensorFlow or PyTorch framework. The exoskeleton execution module is connected to the embedded processing unit via a CAN bus or EtherCAT industrial Ethernet protocol to ensure low-latency real-time transmission of control commands.

[0070] Based on the same concept, this invention provides a control method for a physiological fatigue adaptive control system employing a motor imagery brain-computer interface exoskeleton as described above, comprising: The system simultaneously acquires the user's motor imagery EEG signals and multimodal physiological signals of physiological fatigue state. The acquired raw signals are preprocessed by denoising, filtering, and baseline drift correction. The multimodal physiological signals include at least heart rate variability-related signals, skin conductance signals, and EEG frequency band signals. Feature extraction is performed on the preprocessed multimodal physiological signals. The extracted multidimensional features are then fused and the physiological fatigue index is calculated in real time using a pre-trained machine learning model. The user's current fatigue level is determined based on the physiological fatigue index and preset multi-level fatigue thresholds. The system receives the motor imagery EEG signal and adaptively adjusts the EEG decoding model based on the physiological fatigue index to output control commands that conform to the user's motor intentions. The exoskeleton's auxiliary force output parameters are dynamically calculated based on the physiological fatigue index according to a preset mapping relationship. The lower limb exoskeleton is driven to perform corresponding movements according to the control commands and the auxiliary force output parameters; Real-time feedback and guidance information are provided to the user based on the physiological fatigue index and / or the fatigue level. The data flow and control flow relationships between the above steps have been described in the foregoing system embodiments, and therefore will not be repeated here.

[0071] It should be noted that the division of the various modules in this device / system embodiment is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software through processing element calls; they can also be implemented entirely in hardware; or some units can be implemented by processing element calls to software, while others can be implemented in hardware.

[0072] Based on the same concept, an electronic device is also provided in some embodiments of this application. This electronic device includes a memory and a processor, wherein the memory stores a processing program, and the processor executes the processing program according to instructions. When the processor executes the processing program, the physiological fatigue adaptive control method of the motor imagery brain-computer interface exoskeleton described in the foregoing embodiments is realized.

[0073] In some embodiments of this application, a readable storage medium is also provided, which can be a non-volatile readable storage medium or a volatile readable storage medium. The readable storage medium stores instructions that, when executed on a computer, cause an electronic device containing such a readable storage medium to perform the aforementioned physiological fatigue adaptive control method for a motor imagery brain-computer interface exoskeleton.

[0074] It is understood that, regarding the aforementioned physiological fatigue adaptive control method for the motor imagery brain-computer interface exoskeleton, if all of these methods are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0075] Computer-readable storage media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0076] The program code for executing the technical solutions disclosed in this application can be written in any combination of one or more programming languages. These programming languages ​​include object-oriented programming languages ​​such as Python and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A physiological fatigue adaptive control system for a motor imagery brain-computer interface exoskeleton, characterized in that, include: The multi-physiological signal acquisition module is used to simultaneously acquire the user's motor imagery EEG signals and multimodal physiological signals of physiological fatigue state. The multimodal physiological signals include at least heart rate variability (HRV) related signals, skin conductance (GSR) signals, and EEG frequency band signals. The physiological fatigue monitoring module is connected to the multi-physiological signal acquisition module and is used to extract features from the multi-modal physiological signals and calculate the physiological fatigue index and corresponding fatigue level of the target user based on the extracted multi-dimensional features. The EEG decoding and feature adaptive reconstruction module is connected to the multi-physiological signal acquisition module and the physiological fatigue monitoring module, respectively. It is used to receive the motor imagery EEG signal and adaptively adjust the EEG decoding model according to the physiological fatigue index. When the model completes initialization and passes verification and the decoding confidence reaches the preset condition, it outputs a motor control command; otherwise, it outputs a safety control command to maintain, decelerate, or pause execution. An auxiliary force compensation module, connected to the physiological fatigue monitoring module, is used to dynamically calculate the exoskeleton auxiliary force output parameters according to the physiological fatigue index and a preset mapping relationship. The exoskeleton execution module is connected to the EEG decoding and feature adaptive reconstruction module and the auxiliary force compensation module, respectively, and is used to receive the control commands and the auxiliary force output parameters, and drive the lower limb exoskeleton to perform corresponding movements according to the control commands and the auxiliary force output parameters; The human-computer interaction and feedback module is connected to the physiological fatigue monitoring module and the auxiliary force compensation module, and is used to provide real-time feedback and guidance to the user based on the physiological fatigue index and / or the fatigue level.

2. The physiological fatigue adaptive control system for the motor imagery brain-computer interface exoskeleton as described in claim 1, characterized in that, The physiological fatigue monitoring module includes: The feature extraction unit is used to extract fatigue-related features from the multimodal physiological signals. The fatigue-related features include at least two of the following: HRV time-domain features, HRV frequency-domain features, GSR features, and EEG frequency band power features. The fatigue index calculation unit is used to fuse the fatigue-related features and calculate the physiological fatigue index, which represents the user's current physiological load level, based on the fused features using a pre-trained machine learning model. The fatigue level determination unit is used to compare the physiological fatigue index with a preset multi-level fatigue threshold to determine the user's current fatigue level. The threshold determination branch unit is used to generate corresponding control signals according to the fatigue level and output them to the corresponding modules respectively: when the fatigue level is a non-fatigue level, the EEG decoding and feature adaptive reconstruction module is controlled to call the initial motor imagery decoding model established based on the user's non-fatigue state EEG data to decode the motor imagery EEG signal. When the fatigue level is at the fatigue level, the EEG decoding and feature adaptive reconstruction module is controlled to adaptively adjust the initial motor imagery decoding model according to the current fatigue level and the current EEG features to obtain a fatigue state motor imagery decoding model, and the module is controlled to generate motor control commands using the fatigue state motor imagery decoding model.

3. The physiological fatigue adaptive control system for the motor imagery brain-computer interface exoskeleton as described in claim 2, characterized in that, The feature extraction unit includes: The EEG frequency band feature extraction subunit is used to perform bandpass filtering, notch filtering and artifact removal on the EEG frequency band signal, extract EEG segments through a preset first sliding window, perform time-frequency transformation on each segment to calculate the power spectral density of Delta band, Theta band, Alpha band and Beta band, and calculate the (Theta+Alpha) / Beta ratio and / or Alpha / Beta ratio. The heart rate variability feature extraction subunit is used to detect R wave or PPG peak from the heart rate variability related signal to calculate RR interval sequence, perform artifact correction and interpolation processing on the RR interval sequence, and calculate time domain indices SDNN and / or RMSSD, as well as frequency domain indices LF, HF and / or LF / HF ratio within a preset second sliding window. The skin conductance feature extraction subunit is used to perform low-pass filtering on the skin conductance signal to remove high-frequency noise, extract GSR segments through a preset third sliding window, and calculate the average skin conductance level and the amplitude and frequency of the skin conductance response for each segment.

4. The physiological fatigue adaptive control system for the motor imagery brain-computer interface exoskeleton as described in claim 1, characterized in that, The EEG decoding and feature adaptive reconstruction module includes: The initial decoding unit is used to store the initial motor imagery decoding model trained based on the user's non-fatigue EEG data; The feature adaptive reconstruction unit is used to adaptively reconstruct the current EEG features when the physiological fatigue index reaches or exceeds the first threshold T1, so as to compensate for the feature distribution drift caused by fatigue. The model update unit is used to update the parameters of the initial motion imagery decoding model based on the adaptively reconstructed features, and generate a fatigue state motion imagery decoding model. A decoding strategy switching unit is used to smoothly switch decoding strategies between the initial motor imagery decoding model and the fatigued state motor imagery decoding model based on the physiological fatigue index. When the physiological fatigue index is less than T1, the output of the initial motor imagery decoding model is directly used as the control command. When the physiological fatigue index is greater than or equal to T1, a transition phase begins. The decoding results of the initial motion imagery decoding model and the decoding results of the fatigued state motion imagery decoding model are weighted and fused to generate the final control command. The weight of the decoding result of the initial motion imagery decoding model is negatively correlated with the physiological fatigue index, while the weight of the decoding result of the fatigued state motion imagery decoding model is positively correlated with the physiological fatigue index, so as to achieve a smooth transition from the initial model to the adaptive model in the decoding process.

5. The physiological fatigue adaptive control system for the motor imagery brain-computer interface exoskeleton as described in claim 4, characterized in that, The model update unit is configured as follows: Online acquisition of users' current motor imagery EEG data and corresponding motor intention tags; Based on the current physiological fatigue index and EEG characteristics, the most matching fatigue pattern is retrieved from the pre-stored fatigue feature database, and the pre-fine-tuned model parameters corresponding to that pattern are called as the initial parameters for model updates. Using EEG samples with motor imagery category labels in a non-fatigue state as source domain data, and EEG samples in a fatigue state determined by task prompts during calibration tasks as labeled target domain data, or EEG samples in a fatigue state that are not labeled during normal operation as unlabeled target domain data, the model parameters are optimized and updated through a domain adaptation algorithm to obtain a feature representation robust to fatigue states. Using newly collected batch sample data, the fatigue state motion imagery decoding model is incrementally updated using the gradient descent algorithm; The updated model is evaluated using validation set data, and it is officially activated only when the performance metrics meet preset conditions.

6. The physiological fatigue adaptive control system for the motor imagery brain-computer interface exoskeleton as described in claim 4, characterized in that, The EEG decoding and feature adaptive reconstruction module further includes a real-time monitoring and calibration unit, which is configured as follows: Continuously monitor the online classification accuracy and / or decoding confidence of the fatigue state motion imagery decoding model; the online classification accuracy is determined based on the consistency between the control commands output by the fatigue state motion imagery decoding model and the user's actual motion intention, and the decoding confidence is determined based on the entropy value or maximum probability value of the probability distribution output by the fatigue state motion imagery decoding model. When it is detected that the online classification accuracy is continuously lower than the preset accuracy threshold within a preset time period, or the decoding confidence is continuously lower than the preset confidence threshold, a recalibration trigger signal is generated. Based on the recalibration trigger signal, the human-computer interaction and feedback module is controlled to guide the user to perform a preset calibration task, and the fatigue state motion imagery decoding model is retrained or fine-tuned based on the data collected by the calibration task to restore decoding performance.

7. The physiological fatigue adaptive control system for the motor imagery brain-computer interface exoskeleton as described in claim 1, characterized in that, The auxiliary force compensation module includes: The input interface unit is used to receive the physiological fatigue index output by the physiological fatigue monitoring module in real time. The mapping calculation unit is used to map the physiological fatigue index to the corresponding auxiliary force output ratio according to a preset nonlinear mapping function. The instruction conversion unit is used to convert the auxiliary force output ratio into a motor torque command or impedance parameter adjustment amount for the exoskeleton execution module to execute; The safety limiting unit is used to perform a safety limiting check on the motor torque command or impedance parameter adjustment amount before the motor torque command or impedance parameter adjustment amount is output to the exoskeleton execution module, so as to ensure that it does not exceed a preset safety threshold.

8. The physiological fatigue adaptive control system for the motor imagery brain-computer interface exoskeleton as described in claim 7, characterized in that, The nonlinear mapping function is a piecewise nonlinear function: When the physiological fatigue index is less than the first threshold T1, the auxiliary force output ratio is the minimum auxiliary force A. min ; When the physiological fatigue index is greater than or equal to T1 and less than the second threshold T2, the auxiliary force output ratio is at the minimum auxiliary force A. min It monotonically increases with respect to the first auxiliary force threshold. When the physiological fatigue index is greater than or equal to T2 and less than the third threshold T3, the auxiliary force output ratio increases monotonically between the first auxiliary force threshold and the second auxiliary force threshold at a rate greater than that of the first mapping interval. When the physiological fatigue index is greater than or equal to T3, the auxiliary force output ratio reaches the maximum auxiliary force A. max .

9. A control method employing the system described in any one of claims 1-8, characterized in that, include: Simultaneously acquire the user's motor imagery EEG signals and multimodal physiological signals of physiological fatigue state, and perform noise reduction, filtering and baseline drift correction preprocessing on the acquired raw signals; The multimodal physiological signals include at least heart rate variability-related signals, skin conductance signals, and electroencephalogram (EEG) frequency band signals; Feature extraction is performed on the preprocessed multimodal physiological signals. The extracted multidimensional features are then fused and the physiological fatigue index is calculated in real time using a pre-trained machine learning model. The user's current fatigue level is determined based on the physiological fatigue index and preset multi-level fatigue thresholds. The system receives the motor imagery EEG signal and adaptively adjusts the EEG decoding model based on the physiological fatigue index to output control commands that conform to the user's motor intentions. The exoskeleton's auxiliary force output parameters are dynamically calculated based on the physiological fatigue index according to a preset mapping relationship. The lower limb exoskeleton is driven to perform corresponding movements according to the control commands and the auxiliary force output parameters; Provide users with real-time feedback and guidance information based on the physiological fatigue index and / or the fatigue level.

10. An electronic device, characterized in that, include: The memory is used to store the processing program; A processor, which, when executing the processing program, implements the physiological fatigue adaptive control method for the motor imagery brain-computer interface exoskeleton as described in claim 9.

Images