CN113081671A - Method for improving on-demand auxiliary rehabilitation training participation degree based on Bayesian optimization - Google Patents

Abstract

Translated fromChinese

本发明公开了一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，通过评估被试多次在随机预设轨迹跟随任务中的运动表现指标与运动参与度指标，以提高被试训练中的主动参与度为目标建立评价函数，采用贝叶斯优化方法训练学习评价函数与按需辅助策略的超参数间的关系，寻到下一轮按需辅助策略中最合适的按需辅助策略。该方法同时量化评价被试的运动表现指标与运动参与度指标，可实时监测被试生理心理状态，并根据被试的运动表现与运动参与度变化情况，提供个性化、智能化的最优按需辅助策略，从而保证被试的积极投入状态，有效地刺激神经有效引发神经功能重组，提高了机器人技术辅助康复的训练效率，是康复机器人更快进入临床应用的关键要素。

The invention discloses a method for improving the participation of on-demand assisted rehabilitation training based on Bayesian optimization. The active participation in training is the goal to establish an evaluation function, and the Bayesian optimization method is used to train and learn the relationship between the evaluation function and the hyperparameters of the on-demand assistance strategy, and find the most suitable on-demand assistance in the next round of on-demand assistance strategies. Strategy. This method quantitatively evaluates the subjects' exercise performance indicators and exercise participation indicators at the same time, which can monitor the subjects' physiological and psychological states in real time, and provide personalized and intelligent optimal indicators according to the changes in the subjects' exercise performance and exercise participation. Auxiliary strategies are needed to ensure the active participation of the subjects, effectively stimulate the nerves to effectively trigger the reorganization of nerve functions, and improve the training efficiency of robotic-assisted rehabilitation.

Description

Translated fromChinese

一种基于贝叶斯优化提高按需辅助康复训练参与度的方法A method for improving participation in on-demand assisted rehabilitation training based on Bayesian optimization

技术领域technical field

本发明属于康复机器人、康复训练以及机器学习技术领域，具体涉及一种基于贝叶斯优化提高按需辅助康复训练参与度的方法。The invention belongs to the technical fields of rehabilitation robots, rehabilitation training and machine learning, and particularly relates to a method for improving participation in on-demand assisted rehabilitation training based on Bayesian optimization.

背景技术Background technique

脑卒中已经成为威胁人类身心健康和生命安全的重大疾病之一，超过半数的脑卒中患者存在上肢运动功能障碍，该障碍严重影响了他们的日常生活活动。传统的上肢康复治疗方式主要依靠康复治疗师进行人工辅助训练，这种方式需耗费康复治疗师大量体力，并且难以精确评估患者的康复状态。随着机器人技术的发展，康复机器人的出现为康复治疗提供了新的途径。康复机器人可以在无康复治疗师现场指导的情况下辅助患者进行康复训练，节约了大量人力成本。此外，康复机器人可通过多种传感器精确评估患者的康复状态，有助于康复治疗师为患者制定后续的治疗方案，具有广阔的市场应用前景。Stroke has become one of the major diseases that threaten human physical and mental health and life safety. More than half of stroke patients have upper limb motor dysfunction, which seriously affects their daily activities. The traditional upper extremity rehabilitation treatment method mainly relies on the rehabilitation therapist to perform manual training, which requires a lot of physical strength of the rehabilitation therapist, and it is difficult to accurately assess the patient's rehabilitation status. With the development of robotics, the emergence of rehabilitation robots provides a new approach for rehabilitation therapy. Rehabilitation robots can assist patients in rehabilitation training without the on-site guidance of a rehabilitation therapist, saving a lot of labor costs. In addition, the rehabilitation robot can accurately assess the patient's rehabilitation status through a variety of sensors, which helps rehabilitation therapists to formulate follow-up treatment plans for patients, and has broad market application prospects.

康复机器人的控制策略是影响其康复治疗效果的关键因素之一。近年来，按需辅助控制策略成为了该领域的研究热点。顾名思义，按需辅助控制策略的主要思想是康复机器人按照被试的康复需求提供其完成康复训练任务所需的辅助力矩。该控制策略在保证被试完成康复训练任务的前提下，最小化康复机器人提供的辅助力矩，从而最大化被试提供的主动力矩。研究表明，康复训练重复的本质很容易使被试丧失兴趣、感到无聊，不合适的辅助训练策略极有可能使其丧失信心，对康复训练产生厌烦的情绪，对康复效果产生极其不利的影响。因此当前脑卒中康复的一个研究重点就是如何保持并提高被试在康复训练中的主动参与度。已有研究表明，被试保持主动参与度可以提高康复的效率。但是目前的康复机器人大多仅考虑被试运动表现改变按需辅助策略，缺乏对被试主动参与度的量化评价机制，不能实时监测被试的生理心理状态变化从而了解被试参与度的变化。亦缺乏调动被试积极性的有效机制，无法保证被试的主动参与度以及投入状态。换言之，目前的康复机器人从力和运动层面按需辅助训练被试，不能有效引导被试积极参与，且不能实时反馈被试状态并开展有针对性的按需辅助策略调整。因此，同时量化评价被试的运动表现指标与运动参与度指标，结合机器学习方法开发一种可实时监测被试生理心理状态，并根据被试的运动表现与运动参与度变化情况，提供个性化、最优的按需辅助策略，从而保证被试的积极投入状态，有效地刺激神经有效引发神经功能重组，有望提高机器人技术辅助康复的训练效率，亦是康复机器人更快进入临床应用的关键要素。The control strategy of the rehabilitation robot is one of the key factors affecting its rehabilitation treatment effect. In recent years, on-demand assisted control strategies have become a research hotspot in this field. As the name suggests, the main idea of the on-demand assistance control strategy is that the rehabilitation robot provides the auxiliary torque required to complete the rehabilitation training task according to the rehabilitation needs of the subjects. The control strategy minimizes the auxiliary torque provided by the rehabilitation robot and maximizes the active torque provided by the subject under the premise of ensuring that the subjects complete the rehabilitation training task. Studies have shown that the repetitive nature of rehabilitation training can easily make the subjects lose interest and feel bored, and inappropriate auxiliary training strategies are very likely to make them lose their confidence, produce boredom to rehabilitation training, and have an extremely adverse impact on the rehabilitation effect. Therefore, a current research focus of stroke rehabilitation is how to maintain and improve subjects' active participation in rehabilitation training. Studies have shown that maintaining active participation can improve the efficiency of rehabilitation. However, most of the current rehabilitation robots only consider on-demand assistance strategies to change the subjects' exercise performance, lack a quantitative evaluation mechanism for the subjects' active participation, and cannot monitor the changes in the subjects' physiological and psychological states in real time to understand the changes in the subjects' participation. It also lacks an effective mechanism to mobilize the enthusiasm of the subjects, and cannot guarantee the active participation and investment status of the subjects. In other words, the current rehabilitation robot can not effectively guide the participants to actively participate in the on-demand training of the subjects from the force and motion level, and cannot provide real-time feedback on the subjects' status and carry out targeted on-demand assistance strategy adjustments. Therefore, at the same time, quantitative evaluation of the subjects' exercise performance indicators and exercise participation indicators, combined with machine learning methods to develop a real-time monitoring of the subjects' physiological and psychological state, and provide personalized services according to the changes of the subjects' exercise performance and exercise participation. , the optimal on-demand assistance strategy, so as to ensure the active participation of the subjects, effectively stimulate the nerves and effectively trigger the reorganization of nerve functions, which is expected to improve the training efficiency of robotic-assisted rehabilitation, and is also a key element for rehabilitation robots to enter clinical applications faster. .

发明内容SUMMARY OF THE INVENTION

为解决上述问题，本发明公开了一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，实时监测被试生理心理状态，提供个性化、智能化的最优按需辅助策略，从而保证被试的积极投入状态，有效地刺激神经有效引发神经功能重组，提高了机器人技术辅助康复的训练效率，是康复机器人更快进入临床应用的关键要素。In order to solve the above problems, the present invention discloses a method based on Bayesian optimization to improve participation in on-demand assisted rehabilitation training, monitor the physiological and psychological states of the subjects in real time, and provide personalized and intelligent optimal on-demand assistance strategies, thereby Ensuring the active participation of the subjects, effectively stimulating the nerves and effectively triggering the reorganization of nerve functions, and improving the training efficiency of robotic-assisted rehabilitation are the key elements for rehabilitation robots to enter clinical applications faster.

为达到上述目的，本发明的技术方案如下：For achieving the above object, technical scheme of the present invention is as follows:

一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，包括以下步骤：A method for improving participation in on-demand assisted rehabilitation training based on Bayesian optimization, comprising the following steps:

步骤1.设计轨迹跟随任务：目标轨迹设计为两个半圆组成的类正弦曲线，但是在整个轨迹中，仅在轨迹上均匀显示五个参考点，作为引导点，被试操作过程中控制机器人末端依据参考点进行轨迹跟随；Step 1. Design a trajectory following task: The target trajectory is designed as a sine-like curve composed of two semi-circles, but in the entire trajectory, only five reference points are evenly displayed on the trajectory as guide points, and the end of the robot is controlled during the test operation. Follow the trajectory according to the reference point;

步骤2.评价指标：选择被试的轨迹跟踪误差F^E作为被试的运动表现评估指标；选择表面肌电信号的均方根值(RMS)即肌肉激活度F^MA评估被试在训练中的运动参与度。Step 2. Evaluation index: Select the subject's trajectory tracking error^FE as the subject's exercise performance evaluation index; select the root mean square (RMS) value of the surface EMG signal, that is, the muscle activation degree F^MA to evaluate the subject's performance during training. sports participation.

步骤3.评价机制：以提高康复机器人辅助神经康复的训练效率为目标，以提高被试训练中的主动参与度为切入点，综合运动表现指标和运动参与度指标建立评价函数

Step 3. Evaluation mechanism: with the goal of improving the training efficiency of rehabilitation robot-assisted neurorehabilitation, and the starting point of improving the active participation of the subjects in training, the evaluation function was established by integrating the sports performance indicators and sports participation indicators.

其中，

为肌肉激活度最小下限值，一般设置为0.5；β为权重参数，一般设置为4000～8000。in,

is the minimum lower limit of muscle activation, generally set to 0.5; β is the weight parameter, generally set to 4000-8000.

步骤4.贝叶斯优化过程：首先经过随机过程，获得被试n轮的运动表现指标F^E与运动参与度F^MA以及评价函数J，利用贝叶斯优化训练学习评价函数J与按需辅助策略的超参数f_max即最大边界辅助力的函数关系，找到下一轮按需辅助策略中使得评价函数J最大的超参数

Step 4. Bayesian optimization process: First, through a random process, the exercise performance index^FE , exercise participation degree F^MA and evaluation function J of the n-round subjects are obtained, and the evaluation function J and on-demand assistance are trained by Bayesian optimization. The hyperparameter f_max of the strategy is the functional relationship of the maximum boundary auxiliary force, and find the hyperparameter that maximizes the evaluation function J in the next round of on-demand assistance strategy

进一步，步骤4所述随机过程为，被试进行n轮轨迹跟踪任务，每一轮训练中，超参数f_max为所选取范围内的随机数值，即自适应调整机制在随机过程的每一轮都不相同，保证能够在不同辅助的情况下能够根据被试能力获取到最真实的被试表现；Further, the random process described in step 4 is that the subject performs n rounds of trajectory tracking tasks, and in each round of training, the hyperparameter_fmax is a random value within the selected range, that is, the adaptive adjustment mechanism is performed in each round of the random process. are different, to ensure that the most realistic performance of the test subject can be obtained according to the subject's ability under different auxiliary conditions;

进一步，步骤4所述贝叶斯优化过程，在获取新一轮的评价结果后，将新的数据并入数据集

中，然后根据新的数据集进行贝叶斯优化获取下一轮的超参数f_max，重复该过程数次后，训练结束。Further, in the Bayesian optimization process described in step 4, after obtaining a new round of evaluation results, the new data is merged into the data set

, and then perform Bayesian optimization according to the new data set to obtain the hyperparameter f_max of the next round. After repeating this process several times, the training ends.

步骤5.按需辅助策略：在训练过程中，机器人自适应根据被试在操作过程中的位置误差△d与超参数f_max，按照辅助力场规律实时调整机器人对被试施加的力。Step 5. On-demand assistance strategy: During the training process, the robot adaptively adjusts the force applied by the robot to the subject in real time according to the law of the auxiliary force field according to the subject's position error Δd and hyperparameter f_max during the operation.

进一步，步骤5所述辅助力场公式如下：Further, the auxiliary force field formula described in step 5 is as follows:

f＝f_max*[1-exp(-(△d/λ)²)]f=f_max *[1-exp(-(Δd/λ)² )]

其中，f为机器人施加给被试的辅助力，f_max为最大边界辅助力，λ为辅助力场的激活区域值，在这里取值为0.2。Among them, f is the auxiliary force applied by the robot to the subject, f_max is the maximum boundary auxiliary force, and λ is the activation area value of the auxiliary force field, which is 0.2 here.

本发明的有益效果是：The beneficial effects of the present invention are:

1.该方法实时监测被试的生理心理状态，同时评价被试的运动表现与运动参与度，采用贝叶斯优化学习机制辅助决策，制定调动被试积极性的有效机制，保证了被试的积极参与以及投入状态，为智能化、高效化的神经康复提供了解决方案。1. This method monitors the physical and psychological state of the subjects in real time, and evaluates the subjects' exercise performance and exercise participation. The Bayesian optimization learning mechanism is used to assist decision-making, and an effective mechanism to mobilize the enthusiasm of the subjects is formulated to ensure the enthusiasm of the subjects. Participation and investment status provide solutions for intelligent and efficient neurorehabilitation.

2.针对按需辅助策略对被试主动参与度的影响存在个体差异性的问题，该方法可自主制定个性化按需辅助训练策略，从而使每个被试的康复训练效率更大化。2. Aiming at the problem of individual differences in the impact of on-demand assistance strategies on subjects' active participation, this method can independently formulate personalized on-demand assistance training strategies, so as to maximize the efficiency of each subject's rehabilitation training.

3.该方法采用贝叶斯优化制定下轮最优的按需辅助策略，由于其迭代次数少，能够提升被试在短期内的运动能力，激发被试对于力和运动控制的神经感知，尽可能避免无用的训练过程。3. The method uses Bayesian optimization to formulate the optimal on-demand assistance strategy for the next round. Due to its small number of iterations, it can improve the subjects' exercise ability in the short term, stimulate the subjects' neural perception of force and motor control, and maximize the ability of the subjects. May avoid useless training procedures.

附图说明Description of drawings

图1为算法框架示意图；Figure 1 is a schematic diagram of the algorithm framework;

图2为轨迹跟随任务与辅助力调节示意图。Figure 2 is a schematic diagram of the trajectory following task and auxiliary force adjustment.

具体实施方式Detailed ways

下面结合附图和具体实施方式，进一步阐明本发明，应理解下述具体实施方式仅用于说明本发明而不用于限制本发明的范围。The present invention will be further clarified below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

如图1所示，本发明实例提供的一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，包括如下步骤：As shown in FIG. 1 , a method for improving participation in on-demand assisted rehabilitation training based on Bayesian optimization provided by an example of the present invention includes the following steps:

1.设计轨迹跟随任务：1. Design trajectory following tasks:

目标轨迹设计为两个半圆组成的类正弦平面曲线，如图1所示，在整个轨迹中，仅在轨迹上均匀显示五个参考点G1～G5，作为引导点，被试操作过程中控制机器人末端依据参考点进行轨迹跟随。尤其在熟悉阶段，由治疗师/技术人员对被试进行轨迹跟踪任务的讲解，包括期望轨迹的形状、引导点位置，起止条件等，被试通常在熟悉阶段进行约3-5次循环周期；The target trajectory is designed as a quasi-sinusoidal plane curve composed of two semi-circles, as shown in Figure 1. In the entire trajectory, only five reference points G1~G5 are evenly displayed on the trajectory as guide points, and the subject controls the robot during the operation. The end follows the trajectory according to the reference point. Especially in the familiarization stage, the therapist/technician will explain the trajectory tracking task to the subjects, including the shape of the desired trajectory, the position of the guide point, the starting and ending conditions, etc. The subjects usually perform about 3-5 cycles in the familiarization stage;

2.评价指标：2. Evaluation indicators:

(1)运动表现指标(1) Sports performance indicators

为了衡量被试训练过程中的运动表现，通常选择完成训练任务所用的时间、运动轨迹跟踪误差、运动轨迹柔顺度等评价指标。在这里，选择运动跟踪误差F^E评价被试的运动表现，其表达式如下：In order to measure the exercise performance of the subjects during the training process, evaluation indicators such as the time taken to complete the training task, the tracking error of the motion trajectory, and the compliance of the motion trajectory are usually selected. Here, the motion tracking error^FE is chosen to evaluate the subjects' athletic performance, and its expression is as follows:

其中，xs为平面横坐标起始点，xe为平面横坐标终止点；y_i为实际位置的纵坐标，y_e为各个位置对应的期望纵坐标。Among them, xs is the starting point of the abscissa of the plane, xe is the end point of the abscissa of the plane; y_i is the ordinate of the actual position, and y_e is the desired ordinate corresponding to each position.

(2)运动参与度指标(2) Sports participation index

运动上的参与度被定义为被试积极并努力运动的状态。康复训练中，运动状态一般由肌电信号(EMG)进行监测及表征。有学者在结合虚拟现实技术的步态康复训练中采用EMG信号的均方根值(RMS)评价被试在训练中的运功参与度。由于可以表征信号的能量，因此，均方根值被认为是最具有意义的对肌电信号幅值进行分析的方法。故选择上肢主要负责运动功能的肱二头肌、肱三头肌长头、肱三头肌短头以及肱桡肌作为待分析肌肉群，定义运动参与度如下：Participation in exercise was defined as a state in which the participant was actively and diligently exercising. In rehabilitation training, the movement state is generally monitored and characterized by electromyography (EMG). Some scholars have used the root mean square value (RMS) of the EMG signal to evaluate the exercise participation of the subjects in the training in the gait rehabilitation training combined with virtual reality technology. Because it can characterize the energy of the signal, the root mean square value is considered to be the most meaningful method for analyzing the amplitude of the EMG signal. Therefore, the biceps brachii, the long head of the triceps, the short head of the triceps, and the brachioradialis, which are mainly responsible for the motor function of the upper limbs, were selected as the muscle groups to be analyzed, and the exercise participation was defined as follows:

其中，

为第i通道的肌电信号幅值向量，M为信号的长度。in,

is the EMG signal amplitude vector of the i-th channel, and M is the length of the signal.

3.评价机制：以提高康复机器人辅助神经康复的训练效率为目标，以提高被试训练中的主动参与度为切入点，综合运动表现指标和运动参与度指标建立评价函数

3. Evaluation mechanism: with the goal of improving the training efficiency of rehabilitation robot-assisted neurorehabilitation, and the starting point of improving the active participation of the subjects in training, the evaluation function is established by integrating the sports performance indicators and sports participation indicators.

其中，

为肌肉激活度最小下限值，设置为0.5；β为权重参数，一般设置为4000～8000。in,

It is the minimum lower limit of muscle activation, which is set to 0.5; β is the weight parameter, which is generally set to 4000-8000.

1.贝叶斯优化过程：1. Bayesian optimization process:

首先经过随机过程，获得被试n轮的运动表现指标F^E与运动参与度F^MA以及评价函数J，利用贝叶斯优化训练学习评价函数J与按需辅助策略的超参数f_max即最大边界辅助力的函数关系，找到下一轮按需辅助策略中使得评价函数J最大的超参数

First, through a random process, the exercise performance index F^E , exercise participation degree F^MA and evaluation function J of the n-round test subjects are obtained, and the evaluation function J and the hyperparameter f_max of the on-demand auxiliary strategy are trained by Bayesian optimization. The functional relationship of the auxiliary force to find the hyperparameter that maximizes the evaluation function J in the next round of on-demand auxiliary strategy

进一步，所述随机过程为，被试进行n轮轨迹跟踪任务，每一轮训练中，超参数f_max为所选取范围内的随机数值，即自适应调整机制在随机过程的每一轮都不相同，保证能够在不同辅助的情况下能够根据被试能力获取到最真实的被试表现；Further, the random process is that the subject performs n rounds of trajectory tracking tasks, and in each round of training, the hyperparameter f_max is a random value within the selected range, that is, the adaptive adjustment mechanism does not change in each round of the random process. The same, to ensure that the most realistic performance of the subjects can be obtained according to the ability of the subjects under different auxiliary conditions;

进一步，所述贝叶斯优化过程，在获取新一轮的评价结果后，将新的数据并入数据集

中，然后根据新的数据集进行贝叶斯优化获取下一轮的超参数f_max，重复该过程数次后，训练结束。Further, in the Bayesian optimization process, after obtaining a new round of evaluation results, the new data is merged into the data set

, and then perform Bayesian optimization according to the new data set to obtain the hyperparameter f_max of the next round. After repeating this process several times, the training ends.

2.按需辅助策略：2. On-demand assistance strategy:

在训练过程中，机器人自适应根据被试在操作过程中的位置误差△d与超参数f_max，按照辅助力场规律实时调整机器人对被试施加的力。During the training process, the robot adaptively adjusts the force exerted by the robot on the subject in real time according to the auxiliary force field law according to the position error Δd and the hyperparameter f_max of the subject during the operation.

进一步，所述辅助力场公式如下：Further, the auxiliary force field formula is as follows:

f＝f_max*[1-exp(-(△d/λ)²)]f=f_max *[1-exp(-(Δd/λ)² )]

其中，f为机器人施加给被试的辅助力，f_max为最大边界辅助力，λ为辅助力场的激活区域值，在这里取值为0.2。Among them, f is the auxiliary force applied by the robot to the subject, f_max is the maximum boundary auxiliary force, and λ is the activation area value of the auxiliary force field, which is 0.2 here.

进一步，由于机器人的操作平面为二维平面，θ为被试所操作的机器人末端和当前半圆轨迹圆心的连线与水平面的夹角，根据末端所在的轨迹计算机器人末端输出力的方法如下：Further, since the operation plane of the robot is a two-dimensional plane, and θ is the angle between the line connecting the robot end operated by the subject and the center of the current semicircle trajectory and the horizontal plane, the method of calculating the output force of the robot end according to the trajectory of the end is as follows:

f_x＝f*cosθ*sig(f)f_x =f*cosθ*sig(f)

f_y＝f*sinθ*sig(f)f_y =f*sinθ*sig(f)

其中，sig(f)为机器人的调整力的符号值。Among them, sig(f) is the sign value of the adjustment force of the robot.

以上所述仅为本发明的较佳实施例，并不用以限制本发明，凡在本发明的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within the range.

Claims (7)

Translated fromChinese

1.一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，其特征在于，包括：1. a method for improving on-demand assisted rehabilitation training participation based on Bayesian optimization, is characterized in that, comprising:步骤1.设计轨迹跟随任务：目标轨迹设计为两个半圆组成的类正弦曲线，但是在整个轨迹中，仅在轨迹上均匀显示五个参考点，作为引导点，被试操作过程中控制机器人末端依据参考点进行轨迹跟随；Step 1. Design a trajectory following task: The target trajectory is designed as a sine-like curve composed of two semi-circles, but in the entire trajectory, only five reference points are evenly displayed on the trajectory as guide points, and the end of the robot is controlled during the test operation. Follow the trajectory according to the reference point;步骤2.评价指标：选择被试的轨迹跟踪误差F^E作为被试的运动表现评估指标；选择表面肌电信号的均方根值即肌肉激活度F^MA评估被试在训练中的运动参与度；Step 2. Evaluation index: select the subject's trajectory tracking error^FE as the subject's exercise performance evaluation index; select the root mean square value of the surface EMG signal, namely the muscle activation degree F^MA to evaluate the subject's exercise participation in training ;步骤3.评价机制：以提高康复机器人辅助神经康复的训练效率为目标，以提高被试训练中的主动参与度为切入点，综合运动表现指标和运动参与度指标建立评价函数J；Step 3. Evaluation mechanism: with the goal of improving the training efficiency of rehabilitation robot-assisted neurorehabilitation, and the starting point of improving the active participation of the subjects in the training, the evaluation function J was established by integrating the sports performance indicators and sports participation indicators;步骤4.贝叶斯优化过程：首先经过随机过程，获得被试n轮的运动表现指标F^E与运动参与度F^MA以及评价函数J，利用贝叶斯优化训练学习评价函数J与按需辅助策略的超参数f_max即最大边界辅助力的函数关系，找到下一轮按需辅助策略中使得评价函数J最大的超参数

Step 4. Bayesian optimization process: First, through a random process, the exercise performance index^FE , exercise participation degree F^MA and evaluation function J of the n-round subjects are obtained, and the evaluation function J and on-demand assistance are trained by Bayesian optimization. The hyperparameter f_max of the strategy is the functional relationship of the maximum boundary auxiliary force, and find the hyperparameter that maximizes the evaluation function J in the next round of on-demand assistance strategy

步骤5.按需辅助策略：在训练过程中，机器人自适应根据被试在操作过程中的位置误差△d与超参数f_max，按照辅助力场规律实时调整机器人对被试施加的力。Step 5. On-demand assistance strategy: During the training process, the robot adaptively adjusts the force applied by the robot to the subject in real time according to the law of the auxiliary force field according to the subject's position error Δd and hyperparameter f_max during the operation.2.根据权力要求1所述的一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，其特征在于，步骤2所述运动表现评估指标F^E公式如下：2. a kind of method based on Bayesian optimization to improve the degree of participation in on-demand assisted rehabilitation training according to claim 1, is characterized in that, the described sports performance evaluation index^FE formula of step 2 is as follows:

其中，xs为平面横坐标起始点，xe为平面横坐标终止点；y_i为实际位置的纵坐标，y_e为各个位置对应的期望纵坐标。Among them, xs is the starting point of the abscissa of the plane, xe is the end point of the abscissa of the plane; y_i is the ordinate of the actual position, and y_e is the desired ordinate corresponding to each position.3.根据权力要求1所述的一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，其特征在于，步骤2所述的运动参与度F^MA公式如下：3. a kind of method based on Bayesian optimization to improve the degree of participation in on-demand assisted rehabilitation training according to claim 1, is characterized in that, the^FMA formula of the degree of sports participation described in step 2 is as follows:

其中，

为第i通道的肌电信号幅值向量，M为信号的长度。in,

is the EMG signal amplitude vector of the i-th channel, and M is the length of the signal.4.根据权力要求1所述的一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，其特征在于，步骤3所述的评价函数形式如下：4. a kind of method based on Bayesian optimization to improve the degree of participation in on-demand assisted rehabilitation training according to claim 1, is characterized in that, the described evaluation function form of step 3 is as follows:

其中，

为肌肉激活度最小下限值，设置为0.5；β为权重参数，设置为4000～8000。in,

is the minimum lower limit of muscle activation, set to 0.5; β is the weight parameter, set to 4000-8000.5.根据权力要求1所述的一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，其特征在于，步骤4所述随机过程为，被试进行n轮轨迹跟踪任务，每一轮训练中，超参数f_max为所选取范围内的随机数值，即自适应调整机制在随机过程的每一轮都不相同，保证能够在不同辅助的情况下能够根据被试能力获取到最真实的被试表现。5. A method for improving participation in on-demand assisted rehabilitation training based on Bayesian optimization according to claim 1, wherein the random process in step 4 is that the subject performs n rounds of trajectory tracking tasks, and each In round training, the hyperparameter f_max is a random value within the selected range, that is, the adaptive adjustment mechanism is different in each round of the random process, ensuring that the most realistic results can be obtained according to the subject's ability under different auxiliary conditions. test performance.6.根据权力要求1所述的一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，其特征在于，步骤4所述贝叶斯优化过程，在获取新一轮的评价结果后，将新的数据并入数据集

中，然后根据新的数据集进行贝叶斯优化获取下一轮的超参数f_max，重复该过程数次后，训练结束。6. A method for improving participation in on-demand assisted rehabilitation training based on Bayesian optimization according to claim 1, wherein the Bayesian optimization process in step 4 is performed after obtaining a new round of evaluation results. , incorporating the new data into the dataset

, and then perform Bayesian optimization according to the new data set to obtain the hyperparameter f_max of the next round. After repeating this process several times, the training ends.7.根据权力要求1所述的一种基于贝叶斯优化提高按需辅助康复训练参与度的方法，其特征在于，步骤5所述所述辅助力场公式如下：7. A method for improving participation in on-demand auxiliary rehabilitation training based on Bayesian optimization according to claim 1, wherein the auxiliary force field formula described in step 5 is as follows:f＝f_max*[1-exp(-(△d/λ)²)]f=f_max *[1-exp(-(Δd/λ)² )]其中，f为机器人施加给被试的辅助力，f_max为最大边界辅助力，λ为辅助力场的激活区域值，在这里取值为0.2。Among them, f is the auxiliary force applied by the robot to the subject, f_max is the maximum boundary auxiliary force, and λ is the activation area value of the auxiliary force field, which is 0.2 here.进一步，由于机器人的操作平面为二维平面，θ为被试所操作的机器人末端和当前半圆轨迹圆心的连线与水平面的夹角，根据末端所在的轨迹计算机器人末端输出力的方法如下：Further, since the operation plane of the robot is a two-dimensional plane, and θ is the angle between the line connecting the robot end operated by the subject and the center of the current semicircle trajectory and the horizontal plane, the method of calculating the output force of the robot end according to the trajectory of the end is as follows:f_x＝f*cosθ*sig(f)f_x =f*cosθ*sig(f)f_y＝f*sinθ*sig(f)f_y =f*sinθ*sig(f)其中，sig(f)为机器人的调整力的符号值。Among them, sig(f) is the sign value of the adjustment force of the robot.

Images