Intelligent rehabilitation guidance system based on machine learning

Abstract

The invention discloses an intelligent rehabilitation guidance system based on machine learning, and relates to the technical field of virtual rehabilitation guidance, and the system comprises an integrated multi-source sensor, collects the physiological and motion data of a user in real time, and constructs health data; fusing multi-modal data, adjusting the resistance difficulty according to the condition of the patient through virtual reality, and calculating a comprehensive health score and an action offset early warning index; setting a rehabilitation cooperation particle swarm algorithm, combining the comprehensive health score and the risk early warning index, combining the resistance difficulty and the muscle-joint cooperation coefficient to compensate action compensation, optimizing rehabilitation action parameters, and returning to the upper layer; and analyzing an abnormal root based on a structural causal model, and generating targeted rehabilitation suggestions. The problem that the risk of secondary injury is increased due to lack of dynamic regulation and control of training intensity and muscle compensation behaviors of a patient in traditional rehabilitation training is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual rehabilitation guidance, and specifically to an intelligent rehabilitation guidance system based on machine learning. Background Art

[0002] Traditional rehabilitation training relies on the guidance of physical therapists. The rehabilitation plan is mainly formulated based on experience, and it is difficult to achieve personalized and precise adjustment. At the same time, patients may have problems such as movement compensation, overtraining or undertraining during the rehabilitation process, which affect the rehabilitation effect.

[0003] The intelligent rehabilitation guidance system based on machine learning can collect the physiological data of patients in real time, analyze the rehabilitation progress, so as to improve the rehabilitation efficiency and safety. However, traditional machine learning algorithms cannot dynamically coordinate the rehabilitation degree and movement risk of patients. In addition, although the introduction of virtual reality technology can enhance the immersion of patients and improve the compliance of rehabilitation training, traditional training uses fixed resistance, which is easy to cause undertraining or overfatigue. During the rehabilitation training of patients, it is impossible to judge the standard degree of patients' movements in real time. And as the rehabilitation process of patients and the recovery of their bodies, the existing rehabilitation guidance plan cannot adjust the rehabilitation difficulty according to the real-time physical state of patients. Moreover, during the rehabilitation process, patients often use compensatory movements due to insufficient muscle strength or pain, which will lead to joint wear or muscle atrophy in the long term. The traditional method cannot monitor the safety degree of patients' movements.

[0004] Therefore, the present invention proposes an intelligent rehabilitation guidance system based on machine learning. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent rehabilitation guidance system based on machine learning to solve the existing problems mentioned in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: An intelligent rehabilitation guidance system based on machine learning, comprising:

[0007] A health data acquisition module, which is used to integrate multi-source sensors, collect the physiological and movement data of users in real time, and construct health data.

[0008] A health status scoring module, which is used to fuse multi-modal data, adjust the resistance difficulty according to the patient's situation through virtual reality, and calculate the comprehensive health score and action deviation warning index.

[0009] A guidance parameter optimization module, which is used to set a rehabilitation collaborative particle swarm algorithm, combine the comprehensive health score and risk warning index, combine the resistance difficulty and muscle-joint coordination coefficient to compensate for action compensation, optimize the rehabilitation action parameters, and return them to the health status scoring module.

[0010] The causal reasoning and guidance generation module analyzes the root causes of anomalies based on the structural causal model and generates targeted rehabilitation suggestions.

[0011] A further improvement of the present invention lies in that the health data acquisition module integrates physiological sensors, motion sensors, bioelectric sensors and virtual reality interaction modules to construct a multi-modal spatio-temporal data set D = {x t |x t =(h t ,m t ,e t ,v t )}, where the physiological data vector h t =(HR t ,BP t ,SpO2 t ), HR t represents heart rate, BP t represents blood pressure, SpO2 t represents blood oxygen; the motion data vector m t =(θ joint ,a acc ,f step ), θ joint represents joint angle, a acc represents accelerometer data, f step represents gait frequency; the bioelectric vector e t =(EMG amp ,EMG frep ), EMG amp represents the amplitude feature of the electromyogram signal, EMG frep represents the frequency feature of the electromyogram signal; v t represents virtual reality feedback.

[0012] A further improvement of the present invention lies in that the health status scoring module includes a health scoring unit and an action deviation warning index unit. The health scoring unit extracts the characteristics of individual health status changes through a hierarchical physiological-motion attention network, processes the time series data through an LSTM model, extracts the characteristics of health status changes in the physiological data vector, calculates the importance weights of different data modalities using an attention mechanism, and obtains a comprehensive health score S h .

[0013] A further improvement of the present invention lies in that the action deviation warning index unit calculates the action deviation warning index by combining the physiological data vector and the motion data vector. The calculation formula is:

[0014]

[0015] where α 1 , α 2 and α3 represents the weight coefficient, sig() represents the Sigmoid function, R k represents the action deviation warning index, represents the fluctuation threshold of the i-th item of data in the physiological data vector, represents the fluctuation threshold of the i-th item of data in the motion data vector, represents the time series difference of the physiological data vector, represents the time series difference of the motion data vector, F vir represents the difficulty resistance coefficient.

[0016] A further improvement of the present invention lies in that the difficulty resistance coefficient constructs a dynamic rehabilitation environment based on the Unity engine, adjusts the difficulty resistance coefficient in the scene according to the patient's ability, and sets the target action sequence represents the target action vector at time point t, extracts the patient t∈[1,2,...,T], and extracts the patient's actual action sequence represents the patient's actual action vector at time point t, and obtains the difficulty resistance coefficient by calculating the time series similarity of the target action sequence and the patient's actual action sequence and combining the bioelectric vector wherein, sim() represents the similarity calculation, sim max represents the maximum similarity value, represents the patient's muscle strength coefficient.

[0017] A further improvement of the present invention lies in that the known parameter optimization module adopts a rehabilitation collaborative particle swarm algorithm to optimize the rehabilitation actions, and the specific steps include:

[0018] Step 1: Define the action target and encode the action target as (Am, fr, du, η), including amplitude, frequency, joint angle and exercise intensity;

[0019] Step 2: Randomly generate a set of action target combinations for each group of targets;

[0020] Step 3: Set the action target function;

[0021] Step 4: Dynamically balance the safety and training intensity of the target function according to the real-time comprehensive health score and the action deviation warning index;

[0022] Step 5: Repeat Steps 2 to 4, and through iteration, select the minimum area of the target function and output the rehabilitation action.

[0023] A further improvement of the present invention lies in that the action target function is expressed as:

[0024] SF = λ 1 S h +λ 2 sim(Tac[1,T] , Aac [1,T] ) - λ 3 R k + λ 4 ||mjs act - mjs opt || 2 ;

[0025] Among them, λ 1 , λ 1 , λ 1 , λ 1 represent the rehabilitation demand weights, and mjs act represents the muscle - joint synergy coefficient, and mjs opt represents the ideal synergy threshold.

[0026] The further improvement of the present invention lies in that the rehabilitation demand weight is dynamically coordinated by the real - time comprehensive health score and the motion deviation warning index, where λ 3 = log(1 + R k ).

[0027] The further improvement of the present invention lies in that the muscle - joint synergy coefficient is obtained by, at each time node, combining the muscle activation intensity represented by the amplitude feature of the electromyogram signal and the Euclidean norm of the joint angle to obtain the muscle - joint synergy coefficient mjs act , and the calculation formula is

[0028] The further improvement of the present invention lies in that in the causal reasoning and guidance generation module, the structural causal model represents the causal relationship of the multi - modal spatio - temporal data set with a directed graph and gives rehabilitation guidance suggestions for muscle fatigue or abnormal heart rate through independence tests.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] 1. First, the present invention calculates the motion deviation warning index by combining virtual reality with physiological data vectors and motion data vectors, and dynamically adjusts the warning threshold, which can effectively detect the compensatory behaviors and abnormal motion patterns of patients, and judge the accuracy of the rehabilitation actions of patients in real - time. When the patient may have a large motion deviation due to fatigue or abnormal physiological conditions, the system can issue a warning in real - time, improving the safety and effectiveness of rehabilitation training;

[0031] 2. Based on the real - time comprehensive health score and the motion deviation warning index, the resistance level is dynamically adjusted through the rehabilitation collaborative particle swarm algorithm to match the training intensity with the muscle strength state of the patient. By adjusting the virtual rehabilitation environment, it is ensured that the patient can obtain the optimal rehabilitation effect while adapting to their own abilities;

[0032] 3. By calculating the muscle-joint coordination coefficient, the matching degree between muscle activation and joint movement is analyzed in real time. When the coordination index is low, the system can dynamically adjust the training parameters, guide the patient to adopt the correct force application mode, reduce compensatory movements, detect compensatory behaviors in real time, optimize the rehabilitation training effect, and improve the rehabilitation quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a framework diagram of an intelligent rehabilitation guidance system based on machine learning according to the present invention;

[0034] Figure 2 It is a flowchart of a rehabilitation collaborative particle swarm optimization algorithm of an intelligent rehabilitation guidance system based on machine learning according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] The technical solution of the present invention will be described in detail below with reference to the drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. Without conflict, the technical features in the embodiments of the present invention and the embodiments can be combined with each other.

[0036] The term "and / or" is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " generally represents an "or" relationship between the associated objects before and after.

[0037] Embodiment 1

[0038] Figure 1 A framework diagram of an intelligent rehabilitation guidance system based on machine learning disclosed in this embodiment is shown, including:

[0039] A health data acquisition module, which is used to integrate multi-source sensors, collect the user's physiological and movement data in real time, and construct health data;

[0040] A health status scoring module, which is used to fuse multi-modal data, adjust the resistance difficulty according to the patient's condition through virtual reality, and calculate the comprehensive health score and action deviation warning index;

[0041] A guidance parameter optimization module, which is used to set a rehabilitation collaborative particle swarm optimization algorithm, combine the comprehensive health score and risk warning index, combine the resistance difficulty and muscle-joint coordination coefficient to compensate for action compensation, optimize the rehabilitation action parameters, and return them to the health status scoring module;

[0042] A causal reasoning and guidance generation module, which analyzes the root cause of anomalies based on a structural causal model and generates targeted rehabilitation suggestions.

[0043] The health data acquisition module integrates physiological sensors, motion sensors, bioelectric sensors, and virtual reality interaction modules to construct a multi-modal spatio-temporal data set D = {x t |x t =(h t ,m t ,e t ,v t )}, where the physiological data vector h t =(HR t ,BP t ,SpO2 t ), HR t represents heart rate, BP t represents blood pressure, and SpO2 t represents blood oxygen; the motion data vector m t =(θ joint ,a acc ,f step ), θ joint represents joint angle, a acc represents accelerometer data, and f step represents gait frequency; the bioelectric vector e t =(EMG amp ,EMG frep ), EMG amp represents the amplitude feature of the electromyogram signal, and EMG frep represents the frequency feature of the electromyogram signal; v t represents virtual reality feedback.

[0044] The health status scoring module includes a health scoring unit and an action deviation warning index unit. The health scoring unit extracts the characteristics of individual health status changes through a hierarchical physiological-motion attention network, processes the time series data through an LSTM model, extracts the characteristics of health status changes in the physiological data vector, calculates the importance weights of different data modalities using an attention mechanism, and obtains a comprehensive health score S h through weighted calculation. The calculation formula is: where L represents the data dimension;

[0045] The action deviation warning index unit calculates the action deviation warning index by combining the physiological data vector and the motion data vector. The calculation formula is: where α 1 , α 2 and α 3 represent weight coefficients, sig() represents the Sigmoid function, R k represents the action deviation warning index, represents the fluctuation threshold of the i-th item of data in the physiological data vector, represents the fluctuation threshold of the i-th item of data in the motion data vector, Represents the time series difference of the physiological data vector, Represents the time series difference of the motion data vector, F vir Represents the difficulty resistance coefficient.

[0046] When S h Is relatively low, it may indicate that the patient has fatigue, compensatory movements, or physiological abnormalities. At this time, excessive simulated resistance may cause R k To increase (such as too fast heart rate, muscle fatigue), so it is necessary to dynamically adjust the resistance F vir , to balance the training effect and safety;

[0047] In traditional training, fixed resistance is likely to cause over-fatigue or insufficient training. This system dynamically optimizes F vir , to ensure that the resistance matches the patient's muscle strength; the difficulty resistance coefficient constructs a dynamic rehabilitation environment based on the Unity engine, and adjusts the difficulty resistance coefficient in the scene according to the patient's ability, and sets the target action sequence Represents the target action vector at time point t, extracts the patient t∈[1,2,...,T], and extracts the patient's actual action sequence Represents the patient's actual action vector at time point t. By calculating the time series similarity between the target action sequence and the patient's actual action sequence, combined with the bioelectric vector, the difficulty resistance coefficient is obtained Among them, sim() represents the similarity calculation, sim max Represents the maximum similarity value, Represents the patient's muscle strength coefficient.

[0048] The patient's muscle strength coefficient is set by setting the patient's muscle strength ms st Standard value and standard coefficient The muscle strength coefficient gradually increases as the patient's actual muscle strength ms ac Increases step by step,

[0049] The calculation formula of the function sim() is:

[0050]

[0051] Among them, π represents the optimal alignment path, Represents the distance between the actual action and the ideal action in the feature space. The output of the function sim() represents the matching degree between the actual action and the ideal action, and the smaller the value, the closer they are;

[0052] By calculating sim(), the system can quantify the accuracy of the patient's actions; for example, if sim(Tac [1,T] ,Aac [1,T]) is large, indicating that the patient's movement deviates greatly from the standard template, and there may be compensatory behaviors or movement errors; if sim(Tac [1,T] ,Aac [1,T] ) gradually decreases, it indicates that the quality of the patient's movement is improving.

[0053] Example 2

[0054] Figure 2 shows the flow chart of the rehabilitation collaborative particle swarm algorithm of an intelligent rehabilitation guidance system based on machine learning according to the present invention. Based on the inventive concept of Example 1, this example proposes a rehabilitation collaborative particle swarm algorithm to optimize rehabilitation movements. The specific steps include:

[0055] Step 1: Define the movement target and encode the movement target as (Am, fr, du, η), including amplitude, frequency, joint angle, and exercise intensity;

[0056] Step 2: Randomly generate a set of movement target combinations for each group of targets;

[0057] Step 3: Set the movement target function:

[0058] SF = λ 1 S h +λ 2 sim(Tac [1,T] ,Aac [1,T] )-λ 3 R k +λ 4 ||mjs act -mjs opt || 2 ;

[0059] where λ 1 、λ 1 、λ 1 、λ 1 represent the rehabilitation demand weights, mjs act represents the muscle-joint coordination coefficient, and mjs opt represents the ideal coordination threshold.

[0060] Step 4: Dynamically balance the safety and training intensity of the target function according to the real-time comprehensive health score and the movement deviation warning index;

[0061] The rehabilitation demand weights are dynamically coordinated by the real-time comprehensive health score and the movement deviation warning index, where λ 3 = log(1 + R k ).

[0062] For the rehabilitation demand weight of the comprehensive health score, when S h is large, Rk hour will approach 1. Conversely, it will tend to 0. When measuring the proportion of the health score S h in the overall goal, when the system detects that the risk value R k is continuously increasing, it will automatically reduce the weight of S h (positive return part), and instead pay more attention to risk control; when the risk is very low, it emphasizes more on improving or maintaining the health score; through such a fractional structure, λ 1 will adaptively adjust according to the current state S h and R k 's relative size, without the need to manually define multiple interval rules, it can achieve the effect of reducing the priority of the health score at high risk and increasing the priority of the health score at low risk.

[0063] For the rehabilitation demand weight of the similarity between the actual action and the ideal target action, by the method of "1 - relative difference degree", the "difference degree" is converted into "similarity", then λ 2 the larger it is, the closer the action is to the standard action;

[0064] For the rehabilitation demand weight of the action deviation warning index, when R k = 0, λ 3 = 0, indicating that there is no risk and no additional penalty is required. When R k increases, λ 3 also monotonically increases. However, due to the "slow increase" characteristic of the logarithmic function, it will not make the weight increase too fast and cause the system to be too sensitive or unstable; if the simple form of "R k = λ 3 " is adopted, it may cause extreme changes in the objective function when the risk is slightly large.

[0065] Step Five: Repeat Step Two to Step Four. Through iteration, select the area with the minimum objective function and output the rehabilitation action.

[0066] The muscle - joint coordination coefficient is obtained by, at each time node, the Euclidean norm of the muscle activation intensity represented by the amplitude characteristics of the combined electromyogram signal and the joint angle, to obtain the muscle - joint coordination coefficient mjs act , and the calculation formula is

[0067] EMG amp(t) represents the electromyogram amplitude measured at the t-th time point, that is, the degree of muscle activation. This ratio will fall within the interval [-1, 1] or [0, 1]. Generally, if the two signals are positively correlated and synchronous in the trend of change, the synergy index approaches 1; if their change directions are opposite or asynchronous, the index will become smaller, even close to 0 or negative. At this time, the larger the value, the higher the degree of synergy or matching between muscle activation and joint movement; the smaller the value, the weaker their coupling degree, which may mean that the muscle force generation and the joint movement stage are not synchronous. The present invention can realize adjusting the virtual resistance during the rehabilitation guidance process, detecting the compensatory behavior in real time, and guiding the patient to adopt the correct force generation mode.

[0068] Since it is difficult for patients to understand abstract suggestions such as "why the movement amplitude needs to be adjusted", the structural causal model in the causal reasoning and guidance generation module represents the causal relationship of the multi-modal spatio-temporal data set with a directed graph and gives rehabilitation guidance suggestions for muscle fatigue and abnormal heart rate through independence tests. Specifically, it includes:

[0069] Assume that there are potential relationships between all variables in the multi-modal spatio-temporal data set, and then construct a complete undirected graph with an edge between each pair of variables in the graph;

[0070] Test whether each pair of variables x κ and x τ in the graph are independent under the condition of not including any other variables, that is, the empty set condition. If the statistical test shows that x κ and x τ are conditionally independent, then remove the edge between them; at this time, extract any still-connected data x σ and x γ in the multi-modal spatio-temporal data set again, and then extract any other data connected to x σ and x γ to test whether x σ and x γ are independent. If they are independent, then while removing the edge between x σ and x γ , record the conditional data that makes them independent and send it into the separated data set;

[0071] Increase the size of the conditional set step by step in this way, starting from a single variable until the preset upper limit is reached or all variable combinations have been tested.

[0072] In the undirected graph obtained in the first stage, if there are two variables x κ and x τ both connected to the variable , but there is no direct edge between x κ and , and at the same time if x τ does not belong to xκ and For the separated data sets, it is determined that their structures are Denote x τ It is possible that x k and Common consequences, x k and There is no direct relationship;

[0073] Through the transitive rules, the undirected edges are further oriented to obtain a directed acyclic graph, and a causal model is established.

[0074] For example, if it is found that "heart rate" and "electromyogram signal amplitude characteristics" are correlated when other variables are not considered, but become independent under the condition of "joint angle", it can be inferred that "joint angle" may be the mediating variable between the two.

[0075] Similarly, if the relationship between "acceleration" and "heart rate" is strong when no other variables are included, but remains significant after adding other variables other than "gait frequency", it indicates that "acceleration" may have a direct causal effect on "heart rate".

[0076] Finally, through this systematic test, the automatically inferred causal graph will show the possible causal edges between variables, and this graph can be used as a visual and easy-to-understand advice explanation for patients. For example, generate text or voice for patients, indicating that interventions are made for the key node of "gait frequency" to improve problems such as "acceleration" and "heart rate".

[0077] The setting of the threshold and weight can be based on the default settings of the present invention or can be set by the operator himself.

[0078] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one or more flows and / or blocks Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one or more blocks.

[0080] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one or more flows and / or blocks Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one or more blocks.

[0081] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows and / or blocks Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one or more blocks.

[0082] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit and scope protected by the present invention and the claims. All of these are within the protection scope of the present invention.

Claims

1. An intelligent rehabilitation guidance system based on machine learning, characterized in that: include: Health data collection module, which is used to integrate multi-source sensors, collect user physiological and motion data in real time, and construct health data; The health status scoring module is used to integrate multimodal data, adjust the resistance difficulty according to the patient's condition through virtual reality, and calculate the comprehensive health score and movement deviation warning indicators; The guidance parameter optimization module is used to set the rehabilitation collaborative particle swarm algorithm, combine the comprehensive health score and risk warning indicators, combine the resistance difficulty and muscle-joint synergy coefficient to compensate for the action, optimize the rehabilitation action parameters, and return to the health status scoring module; The causal reasoning and guidance generation module analyzes the root causes of abnormalities based on the structural causal model and generates targeted rehabilitation suggestions.

2. The intelligent rehabilitation guidance system based on machine learning according to claim 1, characterized in that: The health data acquisition module integrates physiological sensors, motion sensors, bioelectric sensors and virtual reality interaction modules to construct a multimodal spatiotemporal data set D = {x t |x t =(h t ,m t ,e t ,v t )}, where the physiological data vector h t =(HR t ,BP t ,SpO2 t ), HR t Indicates heart rate, BP t Indicates blood pressure, SpO2 t Represents blood oxygen; motion data vector m t =(θ joint ,a acc ,f step ), θ joint represents the joint angle, a acc represents the accelerometer data, f step represents the gait frequency; the bioelectric vector e t =(EMG amp ,EMG frep ), EMG amp Represents the amplitude characteristics of electromyographic signals, EMG frep Represents the frequency characteristics of electromyographic signals; v t Represents virtual reality feedback.

3. The intelligent rehabilitation guidance system based on machine learning according to claim 1, characterized in that: The health status scoring module includes a health scoring unit and a motion deviation warning indicator unit. The health scoring unit extracts the individual health status change characteristics through a hierarchical physiological-motor attention network, processes time series data through an LSTM model, extracts the health status change characteristics in the physiological data vector, and uses an attention mechanism to calculate the importance weights of different data modalities. After weighted calculation, a comprehensive health score S is obtained. h .

4. The intelligent rehabilitation guidance system based on machine learning according to claim 3, characterized in that: The motion deviation warning indicator unit calculates the motion deviation warning indicator by combining the physiological data vector and the motion data vector, and the calculation formula is: Among them, α1, α2 and α3 represent weight coefficients, sig() represents Sigmoid function, R k Indicates the action deviation warning indicator, represents the fluctuation threshold of the i-th item in the physiological data vector, represents the fluctuation threshold of the i-th item in the motion data vector, represents the temporal difference of physiological data vector, represents the time difference of motion data vector, F vir Indicates the difficulty resistance coefficient.

5. The intelligent rehabilitation guidance system based on machine learning according to claim 4, characterized in that: The difficulty resistance coefficient is based on the Unity engine to build a dynamic rehabilitation environment, and the difficulty resistance coefficient in the scene is adjusted according to the patient's ability to set the target action sequence Represents the target action vector at time point t, extracts the patient t∈[1,2,...,T], and extracts the patient's actual action sequence The actual motion vector of the patient at time point t is represented. The difficulty resistance coefficient is obtained by calculating the time series similarity between the target motion sequence and the actual motion sequence of the patient and combining the bioelectric vector. Among them, sim() represents similarity calculation, sim max Indicates the maximum similarity. Represents the patient's muscle strength coefficient.

6. The intelligent rehabilitation guidance system based on machine learning according to claim 1, characterized in that: The guidance parameter optimization module adopts the rehabilitation collaborative particle swarm algorithm to optimize the rehabilitation action. The specific steps include: Step 1: Define the action target and encode it as (Am, fr, du, η), including amplitude, frequency, joint angle and movement intensity; Step 2: Randomly generate a set of action target combinations for each group of targets; Step 3: Set the action objective function; Step 4: Dynamically balance the safety and training intensity of the objective function based on the real-time comprehensive health score and action deviation warning indicators; Step 5: Repeat steps 2 to 4, select the minimum area of ​​the objective function through iteration, and output the rehabilitation action.

7. The intelligent rehabilitation guidance system based on machine learning according to claim 6, characterized in that: The action objective function is expressed as: SF=λ1S h +λ2sim(Tac [1,T] ,Aac [1,T] )-λ3R k +λ4||mjs act -mjs opt ||2; Among them, λ1, λ1, λ1, λ1 represent the weights of rehabilitation demand, mjs act represents the muscle-joint synergy coefficient, mjs opt represents the ideal synergy threshold.

8. The intelligent rehabilitation guidance system based on machine learning according to claim 7, characterized in that: The rehabilitation demand weight is dynamically coordinated by the real-time comprehensive health score and the action deviation warning indicator, where λ3=log(1+R k ).

9. The intelligent rehabilitation guidance system based on machine learning according to claim 7, characterized in that: The muscle-joint synergy coefficient mjs is obtained by combining the muscle activation intensity represented by the amplitude characteristics of the electromyographic signal and the Euclidean norm of the joint angle at each time node. act , the calculation formula is 10. The intelligent rehabilitation guidance system based on machine learning according to claim 1, characterized in that: The structural causal model in the causal reasoning and guidance generation module represents the causal relationship of multimodal spatiotemporal data sets with a directed graph, and provides rehabilitation guidance suggestions for muscle fatigue or abnormal heart rate through independence test.

Images