Lower limb gait stability control method based on brain-computer interface multi-modal sensor fusion

Abstract

The application relates to the technical field of brain-computer interfaces, and proposes a lower limb gait stability control method based on a brain-computer interface multi-modal sensor fusion, which comprises the following steps: collecting electroencephalogram data, head movement data, lower limb gait data and auxiliary confirmation data, dividing a gait cycle, mapping a gait phase and extracting an electroencephalogram signal envelope; calculating motion artifact reference, synchronous change degree, spatial common mode disturbance degree and artifact attribution value; determining a purified electroencephalogram envelope signal and electroencephalogram reliability; combining the auxiliary confirmation data and the lower limb gait data to generate auxiliary torque of each joint of a lower limb exoskeleton, so that the stability control of the lower limb gait is realized. The application can improve the accuracy of motion intention recognition.

Description

Technical Field

[0001] This application relates to the field of brain-computer interface technology, specifically to a method for lower limb gait stability control based on brain-computer interface multimodal sensor fusion. Background Technology

[0002] Lower limb exoskeletons and intelligent powered prostheses, which employ reactive closed-loop control logic, can only passively compensate for existing joint angle shifts, body posture tilts, and changes in ground reaction forces. Furthermore, they struggle to coordinate device regulation and human balance in complex dynamic scenarios such as walking on uneven surfaces, climbing stairs, and experiencing sudden disturbances. While multimodal sensor fusion lower limb gait control technology based on brain-computer interfaces can directly decode human movement intentions and achieve proactive, preemptive control, non-invasive EEG signals in dynamic walking scenarios are highly susceptible to interference from head movements, neck muscle tension, and fluctuations in electrode contact, generating motion artifacts that are rhythmically coupled with movement intentions.

[0003] Existing artifact processing techniques are mostly adapted to static experimental environments, relying on general filtering, independent component separation, and offline correction methods. However, they cannot incorporate the complete gait periodic movement patterns, nor can they integrate multimodal data such as electromyography, inertia, and plantar pressure. They also cannot determine the source, interference intensity, and signal reliability of artifacts during dynamic walking in real time. These shortcomings easily lead to excessive suppression of effective EEG signals or artifact residue, directly causing inaccurate recognition of lower limb movement intentions and poor decoding stability. This results in insufficient reliability of the final output gait control commands, severely restricting the practical application of brain-computer interface lower limb gait control technology. Summary of the Invention

[0004] This application provides a lower limb gait stability control method based on brain-computer interface multimodal sensor fusion to solve the problem that EEG signals are easily interfered with by dynamic motion artifacts, leading to inaccurate recognition of movement intentions. The specific technical solution adopted is as follows: One embodiment of this application provides a lower limb gait stability control method based on brain-computer interface multimodal sensor fusion, the method comprising the following steps: Collect EEG data, head motion data, lower limb gait data and auxiliary confirmation data. Divide gait cycles based on lower limb gait data, map the acquisition time within the gait cycle to gait phase, and extract the EEG signal envelope of the target frequency band within the EEG channel. Based on head motion data and lower limb gait data, motion artifact references for gait phases are calculated. Based on the changes in the EEG signal envelope and the changes in motion artifact references for the same gait phase in adjacent gait cycles, the degree of synchronization of gait phases is calculated. Based on the differences in the EEG signal envelopes of adjacent gait phases within a gait cycle, the degree of spatial common-mode perturbation of gait phases is calculated. Combining motion artifact references and the degree of synchronization, the artifact attribution value for gait phases is determined. Based on the artifact attribution values ​​and EEG signal envelopes of all gait phases, the EEG signal envelopes are initially compensated and natural retention anchors are set. The cleaned EEG envelope signal of each gait phase is determined. Based on the cleaned EEG envelope signals of synchronous gait phases in different gait cycles, the degree of spatial common-mode perturbation and the degree of synchronization change determined by the cleaned EEG envelope signals, as well as the differences in the cleaned EEG envelope signals of adjacent gait phases in the gait cycle and the differences in auxiliary confirmation data, the EEG reliability of the gait cycle is calculated. Based on purified EEG envelope signals, auxiliary confirmation data, and lower limb gait data, auxiliary torques are generated for each joint of the lower limb exoskeleton to achieve stability control of lower limb gait.

[0005] Furthermore, the head motion data includes triaxial acceleration, triaxial angular velocity and corresponding acquisition time of the head, and the lower limb gait data includes left and right plantar pressure, lower limb angular velocity, lower limb acceleration and corresponding acquisition time of the lower limbs.

[0006] Furthermore, the method for calculating the motion artifact reference is as follows: Based on the triaxial acceleration and triaxial angular velocity, determine the resultant head acceleration and resultant head angular velocity of the gait phase, and calculate the head motion reference of the gait phase; based on the left and right plantar pressure, calculate the plantar impact reference of the gait phase; based on the lower limb angular velocity, calculate the lower limb motion reference of the gait phase. For any one type of data, including head acceleration, head angular velocity, left foot pressure, right foot pressure, and lower limb angular velocity, the absolute value of the median in the preset evaluation benchmark is used as the numerator, and the sum of the absolute value of the median and the interquartile range in the preset evaluation benchmark is used as the denominator. The value of the fraction is recorded as the characteristic stable value of the data, and the ratio of the characteristic stable value of the data to the sum of the characteristic stable values ​​of all types of data is recorded as the contribution weight of the data. We use contribution weights to perform a weighted summation of head motion references, foot impact references, and lower limb motion references to obtain motion artifact references.

[0007] Furthermore, the method for calculating the degree of synchronous change in the gait phase is as follows: Calculate the change in the EEG signal envelope and the change in the motion artifact reference under the same gait phase in adjacent gait cycles. Based on the changes in the EEG signal envelope and the changes in the motion artifact reference, identify the effective gait phase. Calculate the product of the changes in the EEG signal envelope and the changes in the motion artifact reference under the effective gait phase. When the product is positive, determine that the effective gait phase is changing in the same direction. The ratio of the number of effective gait phases that change in the same direction within the next gait cycle of an adjacent gait cycle to the total number of effective gait phases is denoted as the degree of synchronous change of gait phase within the next gait cycle of an adjacent gait cycle.

[0008] Furthermore, the method for calculating the degree of spatial common-mode perturbation in the gait phase is as follows: The difference between the envelope of the EEG signal in the same EEG channel and the value of the previous adjacent gait phase is calculated and denoted as the adjacent gait phase difference of the EEG channel. The proportion of the same direction change of the adjacent gait phase differences of all EEG channels is denoted as the signal polarity consistency of the gait phase. The reciprocal of the coefficient of variation of the adjacent gait phase differences of all EEG channels is denoted as the common mode characteristic value of the gait phase. The normalized value of the average value of the adjacent gait phase differences of all EEG channels is denoted as the overall perturbation amplitude of the gait phase. The product of the signal polarity consistency of the gait phase, the common-mode eigenvalue, and the overall perturbation amplitude is denoted as the spatial common-mode perturbation degree of the gait phase.

[0009] Furthermore, the artifact attribution value of the gait phase is the product of the motion artifact reference of the gait phase, the degree of synchronization change, and the degree of spatial common-mode disturbance.

[0010] Furthermore, the method for determining the purified EEG envelope signal of the gait phase is as follows: The ratio of the artifact attribution value of the gait phase to the median of the artifact attribution values ​​of all gait phases preceding the current gait phase is used as the artifact attribution deviation value of the gait phase. The ratio of the value of the EEG signal envelope of the gait phase to the median of the value of the EEG signal envelope of all gait phases preceding the current gait phase is used as the EEG signal envelope deviation value of the gait phase. The artifact attribution deviation value is used as the independent variable, and the EEG signal envelope deviation value is used as the dependent variable. A linear fit is performed on the gait phase and the artifact attribution deviation values ​​of all gait phases preceding the current gait phase and the EEG signal envelope deviation value to obtain a fitted line. The value of the phase influence relationship of the gait phase is calculated based on the fitted line. The product of the value of the phase influence relationship, the artifact attribution deviation value of the gait phase, and the artifact attribution value is recorded as the interpretable artifact component. The interpretable artifact component is subtracted from the EEG signal envelope of the gait phase to obtain the purified EEG envelope signal of the gait phase. Low-artifact candidate cycles are selected based on the artifact attribution values ​​of all gait phases within the gait cycle, natural preservation anchors are determined, and some purified EEG envelope signals are revalued based on the natural preservation anchors.

[0011] Furthermore, the specific methods for screening low-artifact candidate cycles based on the artifact attribution values ​​of all gait phases within the gait cycle, determining the natural preservation anchor point, and reassigning values ​​to a portion of the purified EEG envelope signal based on the natural preservation anchor point are as follows: The sum of the first quartile and interquartile range of the artifact attribution values ​​of all gait cycles preceding the gait cycle in which the gait phase is located is denoted as the candidate cycle threshold for the gait phase. Gait cycles in which all gait phase artifact attribution values ​​are less than the candidate cycle threshold are denoted as low artifact candidate cycles. The normalized value of the reciprocal of the artifact attribution values ​​of all gait phases within the low artifact candidate cycle is used as a weight to perform a weighted summation of the EEG signal envelope of the low artifact candidate cycle to obtain the natural retention anchor point of the low artifact candidate cycle. The purified EEG envelope signal within the low artifact candidate cycle that is less than the natural retention anchor point is assigned as the natural retention anchor point.

[0012] Furthermore, the specific method for calculating the EEG reliability of the gait cycle is as follows: Gait phase stability is calculated based on the median and interquartile range of the purified EEG envelope signals of all gait cycles preceding the gait cycle in which the gait phase is located, and the purified EEG envelope signal of the gait phase. Based on the purified EEG envelope signal, the value of the spatial common mode perturbation of the gait phase is recalculated, and the difference between the number 1 and the recalculated value of the spatial common mode perturbation is recorded as the spatial specificity of the gait phase. Based on the purified EEG envelope signal, the value of the synchronous change degree of gait phase is recalculated, and the difference between the number 1 and the recalculated synchronous change degree is recorded as the motion independence of gait phase. The product of gait phase stability, spatial specificity, and movement independence is denoted as the effective EEG retention value of gait phase. The auxiliary confirmation data includes electromyographic activation envelope, joint angular velocity and position change data and corresponding acquisition time; the matching weight of gait phase is calculated based on the difference between the purified EEG envelope signal, electromyographic activation envelope and joint angular velocity of adjacent gait phases within the gait cycle, and the difference between the purified EEG envelope signal, electromyographic activation envelope and joint angular velocity of the gait phase and the previous adjacent gait phase. The product of the effective EEG retention value of the gait phase and the matching weight of the gait phase in the corresponding channel frequency band is denoted as the gait phase confidence level. The sum of the confidence levels of all gait phases corresponding to the EEG signal envelopes of all target frequency bands of all EEG channels in the same gait cycle is denoted as the gait cycle EEG confidence level.

[0013] Furthermore, the method for determining the matching weights of the gait phase is as follows: Calculate the difference values ​​of the purified EEG envelope, EMG activation envelope, and joint angular velocity of adjacent gait phases within the gait cycle. The probability that the signs of the difference values ​​are consistent is recorded as the feature weight of the gait cycle. The sum of the probabilities that the signs of the difference values ​​of the gait phase are consistent with those of the previous adjacent gait phase is recorded as the adjustment weight of the gait phase. The product of the adjustment weight of the gait phase and the feature weight of the gait cycle in which the gait phase is located is recorded as the matching weight of the gait phase.

[0014] The beneficial effects of this application are: This application first divides the gait cycle and maps the acquisition time within the gait cycle to the gait phase, achieving a unified gait phase coordinate transformation for gait cycles of different durations. This ensures that EEG artifact recognition and discrimination are based on the same gait movement phase, completely avoiding signal phase mixing problems caused by hemiplegic gait asymmetry, turning and obstacle avoidance movement disturbances, and dynamic fluctuations in gait speed. Since the effective feature components of the human body's true movement intention and the artifact interference components introduced by the body's dynamic movement overlap in the EEG signal envelope, this application comprehensively determines the artifact attribution degree of the EEG signal based on the synchronicity of the motion source, the spatial common mode of the EEG channel, and the intensity of real-time motion disturbance. It traces the physical motion source of the EEG signal disturbance, calculates the motion artifact reference of the gait phase, and then quantifies the degree of synchronous change of the EEG signal and motion artifact in the corresponding gait phase across the gait cycle. It also quantifies the common artifact disturbances of multiple channels to obtain the spatial common mode disturbance degree of the gait phase, thereby determining the artifact attribution value of the gait phase. Finally, based on the artifact attribution value, an interpretable artifact compensation mechanism is constructed. The initial EEG signal envelope was compensated and a natural retention anchor point was set. The purified EEG envelope signal for each gait phase was determined. Then, from three core dimensions—gait phase stability, spatial specificity, and movement independence—the effectiveness of the compensated EEG signal was analyzed. This included whether the compensated signal was within the normal natural fluctuation range of the synchronous phase, whether the compensated signal eliminated multi-channel common-mode artifacts, and whether the coupling between the compensated EEG signal and somatic movement disturbances was effectively removed. The effectiveness of the compensated EEG signal was comprehensively quantified to obtain the EEG reliability of the gait cycle. The EEG reliability of the gait cycle not only reflects the waveform purity of the compensated EEG signal but also characterizes the matching stability between the EEG signal and the actual limb movement trend. Finally, based on the purified EEG envelope signal, auxiliary confirmation data, and lower limb gait data, auxiliary torques for each joint of the lower limb exoskeleton were generated to achieve lower limb gait stability control and solve the problem that EEG signals are easily interfered with by dynamic movement artifacts, leading to inaccurate recognition of movement intentions. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of a lower limb gait stability control method based on brain-computer interface multimodal sensor fusion, provided as an embodiment of this application. Detailed Implementation

[0016] Please see Figure 1 The diagram illustrates a flowchart of a lower limb gait stability control method based on brain-computer interface multimodal sensor fusion according to an embodiment of this application. The method includes the following steps: Step S001: Collect EEG data, head movement data, lower limb gait data and auxiliary confirmation data. Divide the gait cycle based on the lower limb gait data, map the collection time within the gait cycle to the gait phase, and extract the EEG signal envelope of the target frequency band in the EEG channel.

[0017] This embodiment uses a stroke patient with hemiplegia wearing a lower limb exoskeleton for turning and obstacle avoidance rehabilitation walking training as the analysis scenario. Specifically, in this analysis scenario, the patient's affected lower limb has insufficient supporting muscle strength. During turning, obstacle avoidance, deceleration, and dynamic weight transfer, problems such as head compensatory swaying, uneven force on the sole of the foot, and disordered lower limb swing rhythm are prone to occur. As a result, the collected EEG signals are mixed with the characteristics of the real lower limb movement intention, as well as the artifact interference components caused by the dynamic movement of the body.

[0018] It should be noted that this scenario is merely illustrative, and the technical solution of this application is not limited thereto. The subsequent analysis and calculation process of this solution can be widely applied to all lower limb assistive device control scenarios where dynamic walking artifacts exist, including lower limb exoskeletons, intelligent powered prostheses, and intelligent walking aids.

[0019] EEG data, head movement data, lower limb gait data, and auxiliary confirmation data were collected separately.

[0020] The EEG data is acquired by a non-invasive EEG acquisition unit. The EEG data includes the acquisition time, EEG channel number, original potential signal of the EEG channel, and channel contact quality parameters. The EEG data can be used as the basic signal for decoding motor intention.

[0021] The head motion data is collected by an inertial measurement unit placed on the head or an EEG fixed structure. The head motion data includes the head's three-axis acceleration, three-axis angular velocity and the corresponding acquisition time. The head motion data can be used to characterize various head motion disturbances such as head compensatory swinging, posture adjustment and foot impact transmission.

[0022] The lower limb gait data is jointly collected by a plantar pressure sensor and a lower limb inertial measurement unit. The lower limb gait data includes left and right plantar pressure, lower limb angular velocity, lower limb acceleration and corresponding collection time. The lower limb gait data can be used to identify the boundaries of continuous gait cycles and divide each gait phase stage.

[0023] It should be noted that when the plantar pressure value is 0 for more than 500 consecutive ms, an alarm will be triggered and the system will switch to the safety stop mode of electromyography + inertial sensing.

[0024] The auxiliary confirmation data is collected by the lower limb electromyography acquisition unit or the joint motion sensing unit. The auxiliary confirmation data includes electromyography activation envelope, joint angular velocity and position change data and corresponding acquisition time. The auxiliary confirmation data can be used to verify the matching stability between the compensated EEG characteristics and the actual muscle activation state and limb movement trend.

[0025] The EEG data, head motion data, lower limb gait data, and auxiliary confirmation data are preprocessed respectively. The preprocessing includes data time alignment, basic filtering and noise reduction, completion of accurate sampling points, removal of abnormal sampling points, extraction of EEG target frequency bands, and unification of sensor coordinates. The preprocessing process is a well-known technology and will not be described in detail.

[0026] During turning and obstacle avoidance training, stroke patients with hemiplegia exhibit unstable gait rhythms, with the duration of support, swing amplitude, and impact intensity of the affected lower limb dynamically changing with each movement phase. If traditional fixed-time-window segmentation is used to process EEG signals, the same time window may encompass multiple asynchronous phases, causing genuine motor intent components and motion artifacts to become mixed in the temporal dimension, severely interfering with artifact recognition accuracy. Therefore, this application maps EEG signals from the traditional time domain to the gait phase domain, ensuring that all artifact recognition and signal correction operations are performed based on synchronous phases, thus unifying the data comparison benchmark.

[0027] During the patient's dynamic walking, although there were fluctuations in gait speed and asymmetry in gait between the left and right limbs, the timing of foot contact, weight transfer, forward swing of the lower limbs, and foot landing and lifting off the ground exhibited stable and traceable characteristics. Dynamic motion artifacts were often accompanied by head swaying, foot landing impact, and lower limb swinging movements, appearing regularly at specific gait stages. Only by establishing a precise gait phase benchmark can it be effectively determined whether EEG signal disturbances are stably correlated with specific gait movements, and to distinguish between valid intention signals and artifact interference signals. Compared to fixed time windows, gait phase segmentation can adapt to the asymmetric characteristics of hemiplegic gait and the dynamic rhythmic changes of turning and obstacle avoidance, avoiding the shortcomings of fixed windows that can only guarantee consistent duration but cannot match the actual movement state.

[0028] Based on the left and right plantar pressures, lower limb angular velocities, and corresponding acquisition times in the lower limb gait data, a finite state machine gait event detection algorithm is used to divide the complete gait cycle.

[0029] It is understandable that a complete gait cycle includes four key gait events: foot strike, foot lift, swing initiation, and foot transition. Foot strike and foot lift are identified by left and right foot pressure, while swing initiation and foot transition are identified by lower limb angular velocity.

[0030] The time interval between all acquisition moments within a gait cycle and the start moment of the gait cycle is mapped to the gait phase within the gait cycle, thereby achieving a unified gait phase coordinate transformation for gait cycles of different durations.

[0031] Specifically, the mapping formula for gait phase is:

[0032] in, Indicates the time of data collection In the Gait phase within each gait cycle; It represents the gait phase length of a complete cycle; Indicates the first The start time of each gait cycle; Indicates the time of data collection Distance to The time interval between the start times of each gait cycle; Indicates the first The duration of each gait cycle.

[0033] Mapping to gait phase can eliminate the differences in gait duration caused by turning, obstacle avoidance, and gait speed fluctuations, and achieve accurate correspondence and comparative analysis of the same gait phase under different walking conditions.

[0034] Based on the mapping relationship between the acquisition time and gait phase, EEG data, head movement data, lower limb gait data, and auxiliary confirmation data for each gait phase can be obtained. The gait phase mapping rule allows EEG artifact recognition and discrimination to be carried out based on the same gait movement phase, completely avoiding the signal phase confusion caused by hemiplegic gait asymmetry, turning and obstacle avoidance disturbances, and dynamic fluctuations in gait speed.

[0035] Based on the above gait phase mapping rules, the original potential signals of each EEG channel are rearranged according to the corresponding gait phase to obtain the gait phase domain EEG signal. Through frequency band decomposition technology, the EEG signal envelope of each EEG channel and each target frequency band is extracted.

[0036] In this embodiment, 8-30Hz is selected as the target frequency band for the EEG signal envelope. This frequency band includes the Mu rhythm and Beta rhythm, which are closely related to lower limb motor perception and motor execution. At the same time, it avoids low-frequency baseline drift, power frequency interference and high-frequency EMG artifact concentration areas, and is adapted to the signal characteristics in rehabilitation scenarios such as dynamic walking and turning obstacle avoidance.

[0037] The EEG signal envelope is the contour line of the changes in the amplitude of the EEG signal waveform. It is used to represent the change in the instantaneous energy or intensity of neural activity within the target frequency band over time and can serve as the core input for subsequent artifact analysis and compensation.

[0038] At this point, the EEG signal envelopes for each EEG channel and each target frequency band are obtained.

[0039] Step S002: Based on head motion data and lower limb gait data, calculate the motion artifact reference for the gait phase. Based on the changes in the EEG signal envelope and the changes in the motion artifact reference in the same gait phase of adjacent gait cycles, calculate the degree of synchronization change in the gait phase. Based on the differences in the EEG signal envelopes of adjacent gait phases within the gait cycle, calculate the degree of spatial common-mode perturbation of the gait phase. Combining the motion artifact reference and the degree of synchronization change, determine the artifact attribution value of the gait phase.

[0040] The envelope of EEG signals contains both effective characteristic components representing the true intention of movement and artifact interference components introduced by dynamic bodily movements. These two types of components have a high degree of frequency overlap, making accurate differentiation impossible using traditional single-filtering and offline signal separation methods. Therefore, it is necessary to rely on multimodal gait phase domain motion reference data to trace the physical motion source of EEG signal perturbations, quantify artifact attribution values ​​for each EEG channel and target frequency band, and achieve accurate artifact identification.

[0041] EEG motion artifacts during dynamic walking are not random noise, but rather have a clear physical correlation with human body movements. When stroke patients with hemiplegia undergo turning and obstacle avoidance training, they maintain balance through compensatory head swings, and the impact stress is transmitted through the landing of their feet. At the same time, the swinging, deceleration, and turning movements of the lower limbs trigger changes in limb rhythm. These somatic motion disturbances regularly affect the EEG acquisition process and often recur in fixed gait phases, frequently manifesting as synchronous rises and falls of signals from multiple EEG channels.

[0042] In contrast, while the EEG characteristics corresponding to genuine motor intentions are related to walking tasks, they do not change in complete sync with bodily movements such as head shaking, foot impact, and lower limb posture adjustments. Furthermore, the signal changes in each EEG channel exhibit differentiated characteristics. Based on this difference, the degree of artifact attribution of EEG signals can be comprehensively determined by considering the synchronicity of the motor source, the spatial commonality of EEG channels, and the intensity of real-time motor perturbations, thereby obtaining a reference for motor artifacts.

[0043] First, head motion reference, foot impact reference, and lower limb motion reference are constructed.

[0044] Specifically: The square root of the sum of the squares of the three-axis accelerations at the acquisition time is recorded as the head resultant acceleration at the acquisition time; the square root of the sum of the squares of the three-axis angular velocities at the acquisition time is recorded as the head resultant angular velocity at the acquisition time. Based on the correspondence between gait phases and acquisition times within the gait cycle, the head resultant acceleration, head resultant angular velocity, left foot plantar pressure, right foot plantar pressure, and lower limb angular velocity are obtained for each gait phase; for the head resultant acceleration, head resultant angular velocity, left foot plantar pressure, right foot plantar pressure, and lower limb angular velocity... For any type of data, samples corresponding to multiple gait cycles are manually extracted. The median and interquartile range of the extracted samples are used as evaluation benchmarks. The ratio of the difference between the data value and the median in the evaluation benchmark to the interquartile range in the evaluation benchmark is recorded as the normalized value of the data. The mean of the normalized values ​​of the resultant acceleration and the resultant angular velocity of the head is recorded as the head motion reference. The mean of the normalized values ​​of the left and right plantar pressures is recorded as the plantar impact reference. The normalized value of the lower limb angular velocity is recorded as the lower limb motion reference.

[0045] In this embodiment, a total of 100 samples corresponding to gait cycles were extracted to determine the evaluation benchmark.

[0046] The head motion reference is used to characterize the intensity of disturbances to the overall head motion; the foot impact reference is used to quantify the intensity of disturbances to foot landing impact and gait support switching; and the lower limb motion reference is used to quantify the intensity of disturbances to lower limb swing, deceleration, and turning adjustments.

[0047] Secondly, determine the contribution weights corresponding to head motion reference, foot impact reference, and lower limb motion reference.

[0048] Specifically: For any one of the following data types: head acceleration, head angular velocity, left foot pressure, right foot pressure, and lower limb angular velocity, the absolute value of the median in the evaluation benchmark is used as the numerator, and the sum of the absolute value of the median and the interquartile range in the evaluation benchmark is used as the denominator. The value of the fraction is recorded as the characteristic stable value of the data, and the ratio of the characteristic stable value of the data to the sum of the characteristic stable values ​​of all types of data is recorded as the contribution weight of the data.

[0049] The contribution weight can automatically enhance the role of the plantar impact reference during the landing phase and automatically enhance the role of the head movement reference during the turning phase when the head compensation is obvious, thus avoiding the poor adaptability problem caused by manually fixed weights.

[0050] We use contribution weights to perform a weighted summation of head motion references, foot impact references, and lower limb motion references to obtain motion artifact references.

[0051] Then, the degree of synchronization between EEG signals and motion artifacts in the corresponding gait phases during the gait cycle is quantified.

[0052] The changes in the EEG signal envelope and the motion artifact reference in the same gait phase of adjacent gait cycles are calculated. When the change in the EEG signal envelope is greater than a preset first change scale and the change in the motion artifact reference is greater than a preset second change scale, the corresponding gait phase of adjacent gait cycles is recorded as an effective gait phase. The product of the change in the EEG signal envelope and the change in the motion artifact reference in the effective gait phase is calculated. When the product is positive, the effective gait phase is determined to be a unidirectional change. When the product is negative or takes a value of 0, the effective gait phase is determined to be a non-unidirectional change. The ratio of the number of all unidirectional effective gait phases in the next gait cycle of adjacent gait cycles, before the gait phase, to the total number of effective gait phases is recorded as the degree of synchronous change of the gait phase in the next gait cycle of adjacent gait cycles.

[0053] In this embodiment, samples corresponding to 100 gait cycles are manually extracted. The change in the EEG signal envelope under the same gait phase in adjacent gait cycles is calculated. The sum of the median and interquartile range of all changes in the EEG signal envelope is recorded as a preset first change scale. The change in motion artifact reference under the same gait phase in adjacent gait cycles is calculated. The sum of the median and interquartile range of all changes in motion artifact reference is recorded as a preset second change scale. When there is no effective gait phase changing in the same direction before the gait phase, the degree of synchronous change of the gait phase is set to 0. When there is no previous adjacent gait cycle before the gait cycle, the degree of synchronous change of all gait phases within the gait cycle is set to 0.

[0054] The degree of synchronous change is used to characterize the extent to which the electroencephalogram (EEG) signal exhibits regular synchronous fluctuations in response to bodily movement disturbances.

[0055] For any gait phase within any gait cycle, calculate the difference between the EEG signal envelope of the same EEG channel at the gait phase and the value of the previous adjacent gait phase, denoted as the adjacent gait phase difference of the EEG channel; the proportion of the same-direction changes of the adjacent gait phase differences of all EEG channels is denoted as the signal polarity consistency of the gait phase; the reciprocal of the coefficient of variation of the adjacent gait phase differences of all EEG channels is denoted as the common-mode characteristic value of the gait phase; the normalized value of the average value of the adjacent gait phase differences of all EEG channels is denoted as the overall perturbation amplitude of the gait phase; and the product of the signal polarity consistency, the common-mode characteristic value, and the overall perturbation amplitude of the gait phase is denoted as the spatial common-mode perturbation degree of the gait phase.

[0056] It should be noted that this embodiment uses the maximum-minimum normalization method to calculate the normalized value of the average value of adjacent gait phase differences. In this embodiment, the maximum and minimum values ​​of the adjacent gait phase differences of all EEG channels are used as the maximum and minimum values ​​of the maximum-minimum normalization method, respectively. In the maximum-minimum normalization method, when the denominator is 0, the sum of the original denominator and a preset non-zero small constant is used as the denominator to calculate the normalized value. In this embodiment, the preset non-zero small constant is set to [value missing]. In practical applications, implementers may use other methods from existing technologies, such as the tanh function and the sigmoid function, to calculate the normalized value; no restrictions are imposed here.

[0057] The spatial common-mode perturbation level of the gait phase enables the quantification of multi-channel common artifact perturbation.

[0058] Finally, the product of the motion artifact reference, the degree of synchronization change, and the degree of spatial common-mode disturbance in the gait phase is denoted as the artifact attribution value of the gait phase.

[0059] The artifact attribution value can accurately quantify the degree to which the EEG signal corresponding to the gait phase is interfered with by body movement artifacts, thereby improving the accuracy of artifact recognition. At the same time, it meets the three physical conditions for artifact determination. The three physical conditions are: synchronous fluctuation of EEG signal and body movement disturbance, common disturbance characteristics of multiple EEG channels, and body movement disturbance significantly higher than the natural walking baseline level.

[0060] At this point, the artifact attribution values ​​for each gait phase are obtained.

[0061] Step S003: Based on the artifact attribution values ​​and EEG signal envelopes of all gait phases, initially compensate the EEG signal envelopes and set natural retention anchors. Determine the cleaned EEG envelope signal for each gait phase. Calculate the EEG reliability of the gait cycle based on the cleaned EEG envelope signals of synchronous phases within different gait cycles, the degree of spatial common-mode perturbation and synchronization change determined by the cleaned EEG envelope signals, and the differences in the cleaned EEG envelope signals of adjacent gait phases within the gait cycle and the differences in auxiliary confirmation data.

[0062] Artifact attribution values ​​can only quantify the degree of interference from dynamic motion artifacts on single-channel, single-band EEG signals, and cannot directly output effective EEG signals for lower limb gait control. Directly deleting the signal segments corresponding to high artifact attributions would cause EEG signal interruption, disrupting the continuity of gait control; suppressing and denoising the entire target frequency band of EEG signals could easily lead to the accidental removal of effective EEG components corresponding to genuine movement intentions, resulting in distorted intention decoding. Therefore, an interpretable artifact compensation mechanism needs to be constructed based on artifact attribution values ​​to precisely remove only motion artifact components that can be explained by multimodal motion references, and to generate a standardized EEG reliability index based on the compensation results.

[0063] During dynamic walking, EEG artifacts caused by head compensatory swaying, foot impact, and lower limb posture adjustments exhibit significant gait phase correlation and can be explained through synchronous phase multimodal motion reference. EEG fluctuations that cannot be explained by motion reference and possess multi-channel spatial differentiation characteristics are more likely to contain real neuromotor regulation information. During turning and obstacle avoidance rehabilitation walking in hemiplegic patients, it is crucial to maintain continuous and stable EEG signals. Processing methods such as signal segmentation and full-band suppression will compromise the brain-computer interface's proactive forward control capabilities. Furthermore, residual artifacts can easily lead to misinterpretations of motor intentions. Therefore, an interpretable and precise compensation and effective signal adaptive preservation processing logic should be adopted to replace the traditional, crude segmentation and full-band suppression methods.

[0064] During the same gait phase, the correlation between motion perturbation and EEG signals was established.

[0065] The ratio of the artifact attribution value of the gait phase to the median of the artifact attribution values ​​of all gait phases preceding the current gait phase is used as the artifact attribution deviation value of the gait phase. The ratio of the value of the EEG signal envelope of the gait phase to the median of the EEG signal envelope values ​​of all gait phases preceding the current gait phase is used as the EEG signal envelope deviation value of the gait phase. The artifact attribution deviation value is used as the independent variable, and the EEG signal envelope deviation value is used as the dependent variable. A linear fit is performed on the gait phase and the artifact attribution deviation values ​​of all gait phases preceding the current gait phase and the EEG signal envelope deviation value to obtain a fitted straight line. The phase influence relationship of the gait phase is calculated based on the fitted straight line. The phase influence relationship indicates the magnitude of fluctuation in the EEG envelope caused by each unit deviation of the artifact attribution value. The product of the phase influence relationship value, the artifact attribution deviation value of the gait phase, and the artifact attribution value is denoted as the explainable artifact component. By subtracting interpretable artifacts from the gait phase EEG signal envelope, the initial compensation can be completed, and the purified gait phase EEG envelope signal can be obtained.

[0066] The interpretable artifact components are specific interference terms in the EEG signal that are affected by somatic movement disturbances; during the initial compensation process, the interpretable artifact components need to recover the amplitude dimension by means of a reference voltage constant, which is pre-calibrated by those skilled in the art.

[0067] To address the overcompensation problem and prevent the erroneous removal of effective EEG components, natural retention anchors corresponding to gait phases are constructed.

[0068] The sum of the first quartile and interquartile range of the artifact attribution values ​​of the corresponding gait phases in all gait cycles preceding the gait phase is denoted as the candidate cycle threshold for the gait phase. Gait cycles in which all artifact attribution values ​​of gait phases are less than the candidate cycle threshold are denoted as low artifact candidate cycles. The normalized value of the reciprocal of the artifact attribution values ​​of all gait phases within the low artifact candidate cycle is used as a weight to perform a weighted summation of the EEG signal envelope of the low artifact candidate cycle to obtain the natural retention anchor point of the low artifact candidate cycle. The purified EEG envelope signal within the low artifact candidate cycle that is less than the natural retention anchor point is assigned as the natural retention anchor point to prevent the effective EEG components from being excessively deducted.

[0069] Furthermore, from three core dimensions—gait phase stability, spatial specificity, and movement independence—we analyzed whether the compensated EEG signal was within the normal natural fluctuation range of synchronous phase, whether the compensated signal eliminated multi-channel common-mode artifacts, and whether the coupling between the compensated EEG signal and somatic movement disturbances was effectively resolved. The effectiveness of the compensated EEG signal was comprehensively quantified, and the effective EEG retention values ​​corresponding to the gait phase of each channel, frequency band, and gait cycle were calculated.

[0070] Gait phase stability is calculated based on the median and interquartile range of the purified EEG envelope signals of all gait cycles preceding the gait phase, and the purified EEG envelope signal of the gait phase.

[0071] Select the phase phases of all gait cycles preceding the selected gait phase. Record the median of the purified EEG envelope signals of all selected gait phases as the first historical median of the gait phase. Record the interquartile range of the purified EEG envelope signals of all selected gait phases as the first historical interquartile range of the gait phase. Record the absolute value of the difference between the purified EEG envelope signal of the gait phase and the first historical median as the first absolute value of the gait phase. Use the sum of the first absolute value of the gait phase and the first historical interquartile range as the denominator, and the first absolute value of the gait phase as the numerator. Record the difference between the number 1 and the fraction as the gait phase stability.

[0072] According to the method of determining the spatial common-mode perturbation degree of gait phase based on the EEG signal envelope of the EEG channel, the value of the spatial common-mode perturbation degree of gait phase is recalculated based on the purified EEG envelope signal, and the difference between the number 1 and the recalculated value of the spatial common-mode perturbation degree is recorded as the spatial specificity of gait phase.

[0073] According to the method of determining the degree of synchronous change of gait phase based on the EEG signal envelope of the EEG channel, the value of the degree of synchronous change of gait phase is recalculated based on the purified EEG envelope signal, and the difference between the number 1 and the recalculated value of the degree of synchronous change is recorded as the motion independence of gait phase.

[0074] The product of gait phase stability, spatial specificity, and movement independence is denoted as the effective EEG retention value of the gait phase. The differences in the cleaned EEG envelope, EMG activation envelope, and joint angular velocity of adjacent gait phases within the gait cycle are calculated. The probability that the signs of the differences are consistent is denoted as the feature weight of the gait cycle. The sum of the probabilities that the signs of the differences in the cleaned EEG envelope, EMG activation envelope, and joint angular velocity of the gait phase are consistent with those of the previous adjacent gait phase is denoted as the adjustment weight of the gait phase. The product of the adjustment weight of the gait phase and the feature weight of the gait cycle in which the gait phase is located is denoted as the matching weight of the gait phase. The product of the effective EEG retention value of the gait phase and the matching weight of the gait phase in the corresponding channel frequency band is denoted as the gait phase confidence level. The sum of the confidence levels of all gait phases corresponding to the EEG signal envelopes of all target frequency bands of all EEG channels in the same gait cycle is denoted as the gait cycle EEG confidence level.

[0075] The reliability of gait cycle EEG can not only reflect the waveform purity of the compensating EEG signal, but also characterize the matching stability between the EEG signal and the actual movement trend of the human body.

[0076] At this point, the reliability of the EEG data for gait cycles is obtained.

[0077] Step S004: Based on the purified EEG envelope signal, auxiliary confirmation data, and lower limb gait data, generate auxiliary torques for each joint of the lower limb exoskeleton to achieve stability control of the lower limb gait.

[0078] The stability of EEG signals fluctuates during walking. If fixed weights are used for decoding motor intentions, residual artifacts are easily misjudged as valid motor intentions during interference phases such as head compensation swings, foot impact, and turning / obstacle avoidance. This leads to a mismatch between the timing of assisted control and the actual limb support state. Therefore, EEG reliability is used as a dynamic weight adjustment parameter for the multimodal controller, allowing the results of artifact recognition, orientation compensation, and reliability assessment to directly participate in control decisions.

[0079] Using the support vector machine algorithm, candidate control state support results are calculated based on the purified EEG envelope signal, EMG activation envelope, lower limb angular velocity, and plantar pressure, respectively.

[0080] The candidate control states include six categories of lower limb assistive movement modes: start-up preparation, support maintenance, swing assistance, steering adjustment, disturbance recovery, and stop protection. As is known to those skilled in the art, after inputting a single-channel sensor signal into a support vector machine, the model outputs a corresponding confidence score for each type of control state. This confidence score represents the candidate control state support result, characterizing the degree of bias of the current signal towards each assistive control mode.

[0081] In this context, the weight of EEG signals in multimodal fusion is determined by EEG reliability. The higher the EEG reliability, the greater the proportion of motor intention in active assisted control; the lower the EEG reliability, the weaker the role of EEG intention, and the more the control system relies on the actual motor state reflected by electromyography, inertial sensing, and plantar pressure to perform stability protection control.

[0082] In particular, it should be noted that when the EEG reliability is less than 0.2 for three consecutive gait cycles, it automatically switches to a pure reactive control mode of electromyography + inertial sensing. The pure reactive control mode is a well-known technology and will not be described in detail here.

[0083] During the control output phase, the current gait control state is determined using an existing finite state machine, and auxiliary torques for each joint of the lower limb exoskeleton are generated through an impedance control algorithm. The above processing dynamically allocates the proportion of control quantities based on EEG reliability and the consistency of multimodal signals: when EEG reliability is high and the control states determined by multimodal signals tend to be consistent, the proportion of active intention control quantities in the final command is increased; when EEG reliability decreases or multimodal signal determination results show discrepancies, the proportion of EEG intention control quantities is reduced, while simultaneously increasing the proportion of stabilization and protection control quantities based on plantar pressure, lower limb angular velocity, and gait phase.

[0084] The specific adjustment logic is as follows: Calculate the difference in support results of the four types of signals for the same candidate control state within the gait cycle; take the median of the differences in support results of all corresponding candidate control states before the gait cycle as the normal difference benchmark. When the difference in support results of the candidate control state is less than or equal to the normal difference benchmark, the multimodal signal is in the normal consistency range. When the difference in support results of the candidate control state is greater than the normal difference benchmark, the multimodal signal is in a state of obvious divergence.

[0085] The percentage of proactive intent control is the ratio of the difference between the baseline difference and the support result of the candidate control state. When the percentage of proactive intent control is greater than 1, the percentage of proactive intent control is assigned the value of 1.

[0086] The normal difference benchmark represents the upper limit of the tolerable normal signal divergence. The greater the difference in the support results of candidate control states, the lower the proportion of active intention control.

[0087] It is particularly important to note that in all steps involving division operations, if the denominator is zero, a preset non-zero small constant is uniformly added to the denominator to avoid division by zero errors and ensure normal execution of the operation; all the above calculations are accelerated by FPGA hardware to ensure that the system end-to-end latency is less than 100ms.

[0088] The above mechanism can fully leverage its forward control advantages when the EEG signal is reliable, and continuously reduce the weight of the EEG components when the signal is interfered with by artifacts, effectively preventing residual motion artifacts from dominating the control output of lower limb assistive devices.

[0089] It is understandable that the stability control instructions for lower limb gait specifically include: gait phase coordinates, phase domain EEG envelope, cleaned EEG envelope signal, motion artifact reference, artifact attribution value, EEG confidence, weight of EEG signal participation in multimodal fusion, and candidate control state support results.

[0090] This achieves stable control of lower limb gait.

Claims

1. A method for lower limb gait stability control based on brain-computer interface multimodal sensor fusion, characterized in that, The method includes the following steps: Collect EEG data, head motion data, lower limb gait data and auxiliary confirmation data. Divide gait cycles based on lower limb gait data, map the acquisition time within the gait cycle to gait phase, and extract the EEG signal envelope of the target frequency band within the EEG channel. Based on head motion data and lower limb gait data, motion artifact references for gait phases are calculated. Based on the changes in the EEG signal envelope and the changes in motion artifact references for the same gait phase in adjacent gait cycles, the degree of synchronization of gait phases is calculated. Based on the differences in the EEG signal envelopes of adjacent gait phases within a gait cycle, the degree of spatial common-mode perturbation of gait phases is calculated. Combining motion artifact references and the degree of synchronization, the artifact attribution value for gait phases is determined. Based on the artifact attribution values ​​and EEG signal envelopes of all gait phases, the EEG signal envelopes are initially compensated and natural retention anchors are set. The cleaned EEG envelope signal of each gait phase is determined. Based on the cleaned EEG envelope signals of synchronous gait phases in different gait cycles, the degree of spatial common-mode perturbation and the degree of synchronization change determined by the cleaned EEG envelope signals, as well as the differences in the cleaned EEG envelope signals of adjacent gait phases in the gait cycle and the differences in auxiliary confirmation data, the EEG reliability of the gait cycle is calculated. Based on the purified EEG envelope signal, auxiliary confirmation data, and lower limb gait data, auxiliary torques are generated for each joint of the lower limb exoskeleton to achieve stability control of the lower limb gait. The head motion data includes triaxial acceleration, triaxial angular velocity and corresponding acquisition time of the head, and the lower limb gait data includes left and right plantar pressure, lower limb angular velocity, lower limb acceleration and corresponding acquisition time of the lower limbs. The head motion reference, foot impact reference, and lower limb motion reference are weighted and summed using contribution weights to obtain the motion artifact reference. The artifact attribution value of the gait phase is the product of the motion artifact reference of the gait phase, the degree of synchronization change, and the degree of spatial common-mode disturbance.

2. The lower limb gait stability control method based on brain-computer interface multimodal sensor fusion according to claim 1, characterized in that, The method for calculating the motion artifact reference is as follows: Based on the triaxial acceleration and triaxial angular velocity, determine the resultant head acceleration and resultant head angular velocity of the gait phase, and calculate the head motion reference of the gait phase; based on the left and right plantar pressure, calculate the plantar impact reference of the gait phase; based on the lower limb angular velocity, calculate the lower limb motion reference of the gait phase. For any one of the following data types: head acceleration, head angular velocity, left foot pressure, right foot pressure, and lower limb angular velocity, the absolute value of the median in the preset evaluation benchmark is used as the numerator, and the sum of the absolute value of the median and the interquartile range in the preset evaluation benchmark is used as the denominator. The value of the fraction is recorded as the characteristic stable value of the data, and the ratio of the characteristic stable value of the data to the sum of the characteristic stable values ​​of all types of data is recorded as the contribution weight of the data.

3. The lower limb gait stability control method based on brain-computer interface multimodal sensor fusion according to claim 1, characterized in that, The method for calculating the degree of synchronous change in the gait phase is as follows: Calculate the change in the EEG signal envelope and the change in the motion artifact reference under the same gait phase in adjacent gait cycles. Based on the changes in the EEG signal envelope and the changes in the motion artifact reference, identify the effective gait phase. Calculate the product of the changes in the EEG signal envelope and the changes in the motion artifact reference under the effective gait phase. When the product is positive, determine that the effective gait phase is changing in the same direction. The ratio of the number of effective gait phases that change in the same direction within the next gait cycle of an adjacent gait cycle to the total number of effective gait phases is denoted as the degree of synchronous change of gait phase within the next gait cycle of an adjacent gait cycle.

4. The lower limb gait stability control method based on brain-computer interface multimodal sensor fusion according to claim 1, characterized in that, The method for calculating the degree of spatial common-mode perturbation in the gait phase is as follows: The difference between the envelope of the EEG signal in the same EEG channel and the value of the previous adjacent gait phase is calculated and denoted as the adjacent gait phase difference of the EEG channel. The proportion of the same direction change of the adjacent gait phase differences of all EEG channels is denoted as the signal polarity consistency of the gait phase. The reciprocal of the coefficient of variation of the adjacent gait phase differences of all EEG channels is denoted as the common mode characteristic value of the gait phase. The normalized value of the average value of the adjacent gait phase differences of all EEG channels is denoted as the overall perturbation amplitude of the gait phase. The product of the signal polarity consistency of the gait phase, the common-mode eigenvalue, and the overall perturbation amplitude is denoted as the spatial common-mode perturbation degree of the gait phase.

5. The lower limb gait stability control method based on brain-computer interface multimodal sensor fusion according to claim 1, characterized in that, The method for determining the purified EEG envelope signal of the gait phase is as follows: The ratio of the artifact attribution value of the gait phase to the median of the artifact attribution values ​​of all gait phases preceding the current gait phase is used as the artifact attribution deviation value of the gait phase. The ratio of the value of the EEG signal envelope of the gait phase to the median of the value of the EEG signal envelope of all gait phases preceding the current gait phase is used as the EEG signal envelope deviation value of the gait phase. The artifact attribution deviation value is used as the independent variable, and the EEG signal envelope deviation value is used as the dependent variable. A linear fit is performed on the gait phase and the artifact attribution deviation values ​​of all gait phases preceding the current gait phase and the EEG signal envelope deviation value to obtain a fitted line. The value of the phase influence relationship of the gait phase is calculated based on the fitted line. The product of the value of the phase influence relationship, the artifact attribution deviation value of the gait phase, and the artifact attribution value is recorded as the interpretable artifact component. The interpretable artifact component is subtracted from the EEG signal envelope of the gait phase to obtain the purified EEG envelope signal of the gait phase. Low-artifact candidate cycles are selected based on the artifact attribution values ​​of all gait phases within the gait cycle, natural preservation anchors are determined, and some purified EEG envelope signals are revalued based on the natural preservation anchors.

6. The lower limb gait stability control method based on brain-computer interface multimodal sensor fusion according to claim 5, characterized in that, The specific methods for screening low-artifact candidate cycles based on the artifact attribution values ​​of all gait phases within a gait cycle, determining naturally preserved anchor points, and re-assigning values ​​to a portion of the purified EEG envelope signal based on these naturally preserved anchor points are as follows: The sum of the first quartile and interquartile range of the artifact attribution values ​​of all gait cycles preceding the gait cycle in which the gait phase is located is denoted as the candidate cycle threshold for the gait phase. Gait cycles in which all gait phase artifact attribution values ​​are less than the candidate cycle threshold are denoted as low artifact candidate cycles. The normalized value of the reciprocal of the artifact attribution values ​​of all gait phases within the low artifact candidate cycle is used as a weight to perform a weighted summation of the EEG signal envelope of the low artifact candidate cycle to obtain the natural retention anchor point of the low artifact candidate cycle. The purified EEG envelope signal within the low artifact candidate cycle that is less than the natural retention anchor point is assigned as the natural retention anchor point.

7. The lower limb gait stability control method based on brain-computer interface multimodal sensor fusion according to claim 1, characterized in that, The specific method for calculating the EEG reliability of the gait cycle is as follows: Gait phase stability is calculated based on the median and interquartile range of the purified EEG envelope signals of all gait cycles preceding the gait cycle in which the gait phase is located, and the purified EEG envelope signal of the gait phase. Based on the purified EEG envelope signal, the value of the spatial common mode perturbation of the gait phase is recalculated, and the difference between the number 1 and the recalculated value of the spatial common mode perturbation is recorded as the spatial specificity of the gait phase. Based on the purified EEG envelope signal, the value of the synchronous change degree of gait phase is recalculated, and the difference between the number 1 and the recalculated synchronous change degree is recorded as the motion independence of gait phase. The product of gait phase stability, spatial specificity, and movement independence is denoted as the effective EEG retention value of gait phase. The auxiliary confirmation data includes electromyographic activation envelope, joint angular velocity and position change data and corresponding acquisition time; the matching weight of gait phase is calculated based on the difference between the purified EEG envelope signal, electromyographic activation envelope and joint angular velocity of adjacent gait phases within the gait cycle, and the difference between the purified EEG envelope signal, electromyographic activation envelope and joint angular velocity of the gait phase and the previous adjacent gait phase. The product of the effective EEG retention value of the gait phase and the matching weight of the gait phase in the corresponding channel frequency band is denoted as the gait phase confidence level. The sum of the confidence levels of all gait phases corresponding to the EEG signal envelopes of all target frequency bands of all EEG channels in the same gait cycle is denoted as the gait cycle EEG confidence level.

8. The lower limb gait stability control method based on brain-computer interface multimodal sensor fusion according to claim 7, characterized in that, The method for determining the matching weight of the gait phase is as follows: Calculate the difference values ​​of the purified EEG envelope, EMG activation envelope, and joint angular velocity of adjacent gait phases within the gait cycle. The probability that the signs of the difference values ​​are consistent is recorded as the feature weight of the gait cycle. The sum of the probabilities that the signs of the difference values ​​of the gait phase are consistent with those of the previous adjacent gait phase is recorded as the adjustment weight of the gait phase. The product of the adjustment weight of the gait phase and the feature weight of the gait cycle in which the gait phase is located is recorded as the matching weight of the gait phase.

Images