A method for generating training schemes based on dynamic physiological data and user instructions

Abstract

This invention relates to the field of healthcare informatics technology and discloses a method for generating training schemes based on dynamic physiological data and user commands. The method includes: before establishing a mapping model between user subjective perception and physiological execution, performing a data access verification on the physiological data, which includes activation timing features and signal stability judgment; and establishing the mapping model based on the valid data that has passed the verification, thereby generating a training task and collaborative correction feedback targeting intrinsic perception. This invention, through the data access verification, ensures that the data anchors used to construct the core mapping model simultaneously satisfy both the dimensions of pure source and clear intent, avoiding the risk of reinforcing erroneous movement patterns during training due to compensation or unstable user control, and allowing training to return to the stability and calibration of intrinsic perception.

Description

Technical Field

[0001] This invention relates to a method for generating training schemes based on dynamic physiological data and user instructions, belonging to the field of healthcare informatics technology. Background Technology

[0002] Currently, in the field of personalized home rehabilitation, a commonly used technical approach is to use electromyography or pressure sensors to monitor the user's physiological state in real time and guide the user to complete training through biofeedback. This type of method usually presets an objective physiological data indicator as a training goal. The user's task in training is to use their own physiological activities to make the real-time physiological data generated reach or maintain the target value. In theory, this approach provides a foundation for the quantification and standardization of home training and has constituted the mainstream technical path in this field for a certain period of time.

[0003] However, when this technology is applied to users in the early stages of rehabilitation with relatively weak proprioception, its limitations begin to emerge. The training instructions received by the user are external objective values, which lack a direct correspondence with the user's internal sense of exertion. The user finds it difficult to internalize the instructions into a precise intention to exert force, and the training process often manifests as repeated attempts to reach unknown goals. At the same time, the system only uses the final physiological data values ​​as the basis for evaluation and cannot distinguish the correctness of the neuromuscular recruitment patterns behind the data. In order to achieve the numerical goals, users often unconsciously call non-target auxiliary muscle groups for compensation or produce disjointed physiological outputs due to unstable intentions. Therefore, after collecting and adopting physiological data affected by compensatory behaviors or unstable control, the training programs generated by the existing technical approach may have biases from the beginning, and long-term training may even have the risk of solidifying incorrect movement patterns.

[0004] Specifically, existing technologies have the following shortcomings in application: 1. Training instructions are external objective values, which are disconnected from the user's internal perception, making it difficult for the user to accurately understand and control their own exertion level; 2. Feedback mechanisms only present the results of physiological activities but fail to verify the purity of the movement patterns that produce these results, making it impossible to prevent compensatory behavior; 3. The data source used for personalized program development itself lacks a quality verification process, and the non-target signals introduced by compensation or unstable control affect the accuracy of subsequent program development. Therefore, how to establish a mechanism that can verify the purity of movement patterns and the stability of intent during the data collection stage, and on this basis, construct a method that can accurately map the user's subjective perception with objective physiological execution, and then guide the user to calibrate their internal perception for closed-loop training, becomes the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a method for generating training schemes based on dynamic physiological data and user instructions. Its main purpose is to solve the problem in the prior art that the inability to verify the purity and stability of physiological data sources leads to deviations in establishing the mapping between subjective perception and objective physiological data, making it difficult to generate effective and uncompensated personalized training schemes.

[0006] To achieve the above objectives, the present invention provides a method for generating training schemes based on dynamic physiological data and user instructions, the method comprising the following steps:

[0007] Step a: Receive the physiological data stream generated when the user performs physiological operations on preset subjective perception tags;

[0008] Step b: Before establishing the subsequent personalized mapping model, a data admission verification is performed on the physiological data stream. This data admission verification includes: analyzing the activation timing characteristics of the physiological data stream during the initiation phase; and authorizing the use of the physiological data stream for subsequent stability verification only if the activation timing characteristics conform to a template representing the pure activation of the target muscle group; monitoring the stability index of the physiological data stream that passes the verification in real time; and confirming the physiological data stream as valid physiological data for modeling only if the stability index meets the stability conditions.

[0009] Step c: Based on subjective perception labels and effective physiological data, establish a personalized mapping model that represents the correspondence between the user's subjective perception and physiological execution.

[0010] Step d: Based on the training logic and personalized mapping model, generate a training task that includes subjective perception target, and determine the range of physiological data target corresponding to the subjective perception target.

[0011] Step e involves comparing the real-time collected physiological data with the target range of physiological data during the user's training task and generating collaborative correction feedback.

[0012] Preferably, the method for establishing a personalized mapping model in step c includes: acquiring at least two subjective perception labels and their corresponding valid physiological data, wherein the subjective perception labels are terms that characterize different intensity levels of force perceived by the user; and generating a personalized mapping model by applying a curve fitting algorithm to the at least two subjective perception labels and their corresponding valid physiological data.

[0013] Preferably, the activation timing features include the onset delay from the issuance of the command to the first breakthrough of the resting baseline noise threshold by physiological data, and the peak slope characterized by the reciprocal of the time taken for physiological data to rise from the onset point to the first peak; the template is the numerical range of the onset delay and the peak slope.

[0014] Preferably, the method for determining the stability index and judging the stability condition includes: continuously calculating the coefficient of variation (CV) of the physiological data stream within a sliding time window, wherein... , The standard deviation of physiological data within the sliding time window. The value is the average of the physiological data within the sliding time window; and when the value of the coefficient of variation (CV) is consistently lower than the steady-state threshold, the stability condition is considered to be met.

[0015] Preferably, the collaborative correction feedback in step e is a non-numerical graphical user interface change, and the degree of change in the form of the graphical user interface is determined by the deviation between the real-time physiological data and the target range of the physiological data.

[0016] Preferably, after step e, a dynamic adjustment mechanism for compensating for fatigue during the session is further included. This mechanism includes: acquiring the corresponding actual physiological data for each training task successfully executed by the user; updating the fatigue compensation factor based on the difference between the actual physiological data and the initial physiological data recorded in the personalized mapping model; and applying the fatigue compensation factor to adjust the target range of the physiological data to be generated when generating the next training task.

[0017] Preferably, the method further includes: when the preset update triggering condition is met, repeating steps a to c to dynamically update the personalized mapping model.

[0018] Preferably, the method further includes an active detection mechanism for assessing the user's neuromuscular agility, the mechanism comprising: issuing an instruction to the user during the execution of a training task, instructing the user to transition from the current physiological execution state to another target state; measuring the response time taken from the time the user receives the instruction until the real-time physiological data reaches the physiological data value corresponding to the other target state; and using the response time as an indicator for assessing the user's neuromuscular agility.

[0019] Preferably, the method further includes, after step c: calculating a perceptual control gain index characterizing the slope of the personalized mapping model; and using the perceptual control gain index as an independent evaluation parameter for tracking changes in the user's neuromuscular control efficiency.

[0020] Preferably, the method further includes a system in-situ state self-verification mechanism before performing step c. This mechanism includes: applying a standardized mechanical micro-perturbation signal to the sensor carrying physiological data through an excitation source; using the sensor to collect the response signal to the mechanical micro-perturbation signal and generating a response feature characterizing the system in-situ state; comparing the response feature with a reference feature acquired during initial calibration; and performing step c only when the difference between the response feature and the reference feature is less than a threshold.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] 1. This invention introduces a dual data access verification mechanism, including activation timing feature analysis and physiological signal stability judgment, before establishing a personalized mapping model between subjective perception and physiological execution. This changes the situation in existing technologies where physiological data is directly adopted without being able to identify its intrinsic quality. The method first analyzes the activation timing features of the physiological operation initiation phase to exclude compensatory data generated by non-target muscle groups. Then, it judges whether the data stream that has passed the initial screening has entered a stable output state, filtering out oscillating data caused by user hesitation or unstable control. This serial verification logic of first identifying authenticity and then judging stability ensures that every data anchor point used to build the core mapping model simultaneously satisfies the two key dimensions of pure source and clear intent. This ensures that the training objectives of the training program generated based on this model are anchored to real and effective neuromuscular recruitment activities from the source, avoiding the risk of reinforcing incorrect movement patterns or ineffective physiological activities during the training process.

[0023] 2. This invention proposes an interactive method that returns the setting of training task goals and feedback methods to the user's subjective perception, avoiding the cognitive load and execution bias caused by users pursuing an external physiological data goal in existing technologies. This method is based on a pre-established user-personalized perception-physiology mapping model to generate training tasks centered on maintaining a certain internal sensation. At the same time, real-time physiological data is transformed into a non-numerical co-correction feedback, which is only used to confirm or fine-tune the user's maintenance of the current internal sensation. The user's attention is thus guided from the external display interface back to the perception and stabilization of their own internal state. The role of physiological data here changes from a passively executed command to an active learning reference. This change in interaction method makes the training process more in line with the learning laws of motor neurons, reduces the user's execution threshold and frustration, and helps maintain compliance in home rehabilitation.

[0024] 3. This invention establishes a dynamic closed-loop adjustment mechanism that can run through the entire process of a single training session and a long-term rehabilitation cycle. This allows the training program to respond to the user's immediate physiological state changes and long-term ability evolution. In a single training session, by tracking the decreasing trend of the user's actual physiological output while maintaining the same subjective feeling, the system can identify and quantify muscle fatigue during the session and make compensatory adjustments to the physiological target range of subsequent tasks accordingly, so that the user's training load matches their immediate ability. In the long term, by periodically re-anchoring perception and updating the model, the training program can not only adaptively adjust with the user's rehabilitation process, but more importantly, the change in the slope of the mapping model itself is defined as an evaluation index that can characterize the user's neuromuscular control efficiency, providing a deeper perspective beyond muscle strength growth for measuring rehabilitation effects. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the system interaction process for fatigue compensation according to the present invention;

[0026] Figure 2 This is a schematic diagram illustrating the effect of the fatigue compensation mechanism of the present invention;

[0027] Figure 3 This is a schematic diagram of the overall process of the method for generating training schemes based on dynamic physiological data and user instructions according to the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are intended to explain this invention and are not intended to limit the scope of protection of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0029] The present invention provides a method for generating training schemes based on dynamic physiological data and user commands. Its architecture may include a data access verification module, a personalized mapping model construction module, a training task generation and feedback module, and a dynamic closed-loop adjustment system. The data flow process involves first receiving the physiological data stream generated when a user performs physiological operations on preset subjective perception labels. This data stream is processed by the data access verification module to identify valid physiological data. Subsequently, the valid physiological data and the corresponding subjective perception labels are transmitted to the personalized mapping model construction module to establish a user-specific perception-physiological mapping relationship, thereby generating and providing training tasks. The module then generates training tasks and collaborative correction feedback based on this model, aiming to calibrate intrinsic perception. During training, a dynamic closed-loop adjustment system adaptively adjusts to the user's immediate state and long-term ability evolution. Before the personalized mapping model is established, the key parameters required for activating the temporal feature template and judging signal stability are determined through a system-level initialization calibration procedure. This procedure first guides the user to repeatedly perform the target muscle group reference action 5 to 10 times and collect data. Then, the outlier is removed from the obtained onset delay and peak slope sample set using the interquartile range method, where IQR is the difference between the third quartile Q3 and the first quartile Q1. Outliers exceeding [ , Sample points outside the specified range are discarded, and then the mean is calculated based on the remaining valid sample points. with standard deviation The upper and lower limits of the activation time-series feature template are set as follows: Here, k is a confidence coefficient, which defines the verification width of the template, and its value is preset to 2.0. Meanwhile, the coefficient of variation (CV) steady-state threshold in stability verification is derived from the statistical analysis of a pre-set set of gold-standard stable signals generated by multiple subjects on various standard devices and confirmed by experts. The specific value of this threshold is determined to be the 95th percentile of the distribution of all CV sample values ​​in this set, thus statistically defining an upper limit for physiological signal fluctuations that can be accepted by the system. In the system's in-situ self-verification mechanism, the response characteristics are quantified into an N-dimensional feature vector. Each dimension represents the frequency and amplitude pairs of the N highest-energy resonant peaks extracted after performing a Fast Fourier Transform (FFT) on the sensor response signal, while the real-time response characteristics... Compared with the benchmark features The difference is calculated by measuring the weighted Euclidean distance between the two. To determine this, the calculation formula is: ,in, and The first The frequency and amplitude of each resonant peak, and These are the dimensionless weighting coefficients used to adjust the contribution of frequency and amplitude to the total difference, respectively. The threshold for judging this difference is set by summing the mean of the D values ​​and three times the standard deviation obtained when initially acquiring the baseline features by performing several small perturbations around the correct coupling position. These precisely calibrated underlying parameters and models not only support the execution of a single training task, but their longitudinal data also jointly construct a personal digital phenotype that can depict the long-term evolution trajectory of a user's neuromuscular control ability. This provides a data foundation for achieving higher-order rehabilitation state prediction and adaptive optimization of intervention strategies, demonstrating the application potential of the solution in personalized precision medicine scenarios.

[0030] In the application of personalized rehabilitation training, an objective challenge lies in the fact that the quality of the physiological data collected by the system itself is not verified. When users execute instructions, they may compensate by calling non-target muscle groups due to insufficient proprioceptive control or produce disjointed physiological outputs due to unstable intentions. If subsequent personalized programs are established based on such data, the training may deviate from the predetermined goals or even reinforce incorrect movement patterns. To address this challenge, the present invention configures a data access verification process. This process verifies the received physiological data stream before establishing the subsequent personalized mapping model. Specifically, the process first analyzes the activation timing characteristics of the physiological data stream in the initiation phase. These activation timing characteristics include the effective period from the issuance of the instruction to the first breakthrough of a resting baseline noise threshold by the physiological data. The onset delay, represented by the reciprocal of the time taken for physiological data to rise from the onset point to the first peak, is the peak slope. The system compares the real-time calculated onset delay and peak slope with a preset template representing pure activation of the target muscle group. The template is a numerical range of onset delay and peak slope obtained through statistical analysis of experimental data under pure activation of the target muscle group. For example, the range of onset delay can be set to 50 milliseconds to 200 milliseconds. Only when the real-time extracted activation timing features match this template is the physiological data stream authorized for subsequent stability verification. Furthermore, for physiological data streams that have passed the activation timing feature verification, the system will further monitor their stability indicators in real time. The stability indicator is determined by continuously calculating the coefficient of variation of the physiological data stream within a sliding time window. ,in , The standard deviation of physiological data within the sliding time window. The average value of physiological data within a sliding time window, when the coefficient of variation The physiological data stream is considered to meet the stability condition only when the value is consistently below a preset steady-state threshold, such as 0.05, and maintained for the shortest duration, such as 1 second. In this way, a data admission check that includes activation time-series feature analysis and signal stability judgment ensures that the data anchors used to build the core mapping model pass the dual checks of source purity and intent clarity before being adopted, thus providing physiological data that meets the preset standards for subsequent generation schemes.

[0031] To establish the correspondence between a user's subjective perception and objective physiological performance, after acquiring valid physiological data, the system executes the following procedure to build a personalized mapping model representing the correspondence between the user's subjective perception and physiological performance. This procedure first acquires at least two subjective perception labels and their corresponding valid physiological data. The subjective perception labels are terms representing different intensity levels of force perceived by the user, such as mild and moderate. Subsequently, the system applies curve fitting algorithms, such as piecewise linear interpolation or polynomial fitting, to these at least two subjective perception labels and their corresponding valid physiological data to generate the personalized mapping model. This model constructs a functional relationship that quantifies the user's subjective feelings into expected physiological data, providing a personalized technical basis for subsequent generation of training tasks targeting intrinsic perception. Based on the established personalized mapping model, the system can transform the training objective from a fixed physiological data value into an objective for the user. Calibration of subjective perception: Based on preset training logic and a personalized mapping model, the system generates a training task containing a subjective perception target, such as contraction with a moderate sensation. Simultaneously, the system uses the personalized mapping model to parse this moderate subjective perception target into a corresponding physiological data target range. During the user's execution of this training task, the system compares the real-time collected physiological data with the physiological data target range and generates collaborative correction feedback. This collaborative correction feedback can be designed as a non-numerical graphical user interface change, such as a change in the shape or color of a graphic. The degree of change in the graphical user interface shape is determined by the deviation between the real-time physiological data and the physiological data target range. The user's attention is thus guided back to the perception and stabilization of their internal state, and the changes in the graphical interface serve as a reference for confirmation or fine-tuning. This interaction method constitutes a closed-loop training process aimed at guiding the user to calibrate their internal perception.

[0032] Furthermore, to enable the training program to respond to the user's immediate physiological state changes and long-term ability evolution, this invention may also include a dynamic closed-loop adjustment mechanism. Firstly, to address muscle fatigue that may occur during a single training session, the method may include a dynamic adjustment mechanism for compensating for fatigue within the session. This mechanism acquires the corresponding actual physiological data for each training task successfully executed by the user, and updates the fatigue compensation factor based on the difference between the actual physiological data and the initial physiological data recorded in the personalized mapping model. When generating the next training task, the system can apply the fatigue compensation factor to adjust the target range of the physiological data to be generated, so that the training load matches the user's immediate ability. Secondly, to assess neuromuscular function, the method may include an active detection mechanism for assessing the user's neuromuscular agility. This mechanism issues a command to the user during the execution of a training task, instructing them to transition from the current physiological execution state to another target state, and measures the response time from receiving the command to the user's real-time physiological data reaching the physiological data value corresponding to the other target state. The method can be used as an indicator to assess the user's neuromuscular agility; thirdly, the personalized mapping model itself can also be used for evaluation. After the model is established, the present invention can also calculate a perceptual control gain index that characterizes the slope of the model and use this index as an independent evaluation parameter to track changes in the user's neuromuscular control efficiency; fourthly, in order to enable the training program to follow the user's rehabilitation process, the method can also repeatedly execute the process from data acquisition to model establishment when preset update trigger conditions are met, such as periodically or by the user's initiative, so as to dynamically update the personalized mapping model; fifthly, in order to improve the reliability of data acquisition, the method can also include a system in-situ self-verification mechanism. Before executing the modeling step, the mechanism applies a standardized mechanical micro-perturbation signal to the sensor carrying physiological data through an excitation source, and uses the sensor to collect response signals to generate a response feature characterizing the system in-situ state. By comparing the response feature with the baseline feature obtained during initial calibration, subsequent operations are only performed when the difference between the two is less than a threshold. This helps to reduce the risk of data inaccuracy caused by sensor physical position drift.

[0033] Example 1: In a home rehabilitation scenario for pelvic floor dysfunction, the user is an early-stage rehabilitation patient with weak proprioception. The challenge lies in internalizing external, objective training instructions into stable and pure internal force intentions. The execution process is often accompanied by compensatory force exertion of non-target muscle groups or fluctuations in physiological signals due to unstable control. Conventional biofeedback methods only use the final values ​​of physiological data as the evaluation criterion, failing to discern the correctness of the neuromuscular recruitment patterns behind the data, thus risking the solidification of incorrect movement patterns. When using the method of this invention, the system does not first issue an objective physiological data target to the user, but instead initiates the process of establishing a personalized mapping model. The first step of this process is to guide the user to execute an instruction based on subjective perception through the user interface, such as "Please contract with what feels like a very light force." During the user's first attempt, the abdominal muscles were mobilized to compensate for the loss. Upon receiving the physiological data stream generated by this physiological action, the system's data access verification mechanism was triggered. The activation timing feature analysis module within this mechanism extracted the onset delay and peak slope from the data stream's initiation phase and compared these two feature values ​​with a preset template representing pure activation of the target muscle group. Because the activation timing of the compensatory contraction did not conform to the template, the data stream was deemed invalid and discarded. The system then issued a non-numerical correction command to the user. After adjustment, the user attempted again. This time, the correct pelvic floor muscles were activated, but due to unstable control, the physiological data showed continuous fluctuations. At this point, the data stream passed the activation timing feature verification and was sent to the stability verification stage. This stage continuously calculated the coefficient of variation of the physiological data stream within a sliding time window. If the value consistently exceeds a preset steady-state threshold, the system determines that the user's physiological function has not yet reached a stable state and continues to wait without collecting data until the user finds a sense of control, thus reducing the coefficient of variation of the physiological data output. Only after the physiological data remains below the steady-state threshold for a preset time is the system recognized as valid physiological data within this stable window and bound to the subjective perception label of "very light". In this process, the activation of the temporal feature verification provides a pure signal for the subsequent stability verification, avoiding the analysis of invalid compensatory signals. The stability verification ensures that the signal that passes the temporal verification also represents a clear and stable subjective intention. The cascaded collaboration of the two verification steps will only confirm the physiological signal as a data anchor point that can be used for modeling when both the source and stability of the physiological signal meet the preset standards.

[0034] Based on the anchor point set obtained through the above verification process, which includes multiple subjective perception labels and their corresponding valid physiological data, the system generates a personalized mapping model using a curve fitting algorithm. Based on this model, a new training task is issued to the user: maintain a moderate sensation for ten seconds. During task execution, when the user's actual physiological output falls below the physiological data target range determined by the mapping model due to muscle fatigue while maintaining the same subjective sensation, a dynamic adjustment mechanism to compensate for intra-session fatigue is activated. This mechanism uses the user's actual physiological data from the last successful task execution to update a fatigue compensation factor and applies this factor to dynamically lower the physiological data target range for subsequent training tasks. This approach resolves the contradiction of requiring the user to achieve a physiological goal set in a non-fatigue state while in a fatigued physiological state, allowing the training load to adjust according to the user's real-time ability changes. In another training phase, to evaluate... The user's rapid reaction ability triggers an active detection mechanism to assess the user's neuromuscular agility. When the user maintains a moderate level of sensation, the system issues a command to transition to a stronger level of sensation and measures the response time, thus providing a parameter for assessing the speed of control in the rehabilitation process. The generation and execution of the entire training program no longer requires the user to match an externally fixed value, but is transformed into a dynamic calibration process based on the user's verified, pure, and stable subjective perception, and capable of adapting to changes in their immediate state. Throughout the training, the user's attention is guided to the identification, stabilization, and transformation of their own internal force sensation. The role of physiological data here changes from a passively executed command to a reference system for calibrating internal perception. Through the coordinated operation of a series of internal mechanisms, the system transforms a trial-and-error process that is difficult to succeed due to compensation and instability into a guided learning process in which the goal is dynamically matched with the user's ability.

[0035] Example 2: In an experiment, to verify the role of the data admission verification mechanism in distinguishing physiological signals of different natures and its impact on the accuracy of subsequent personalized mapping models, an experimental platform was built, comprising a physiological signal simulation source, a data acquisition module, and data processing software. The physiological signal simulation source consisted of a programmable pneumatic actuator and a pressure sensor, capable of repeatedly generating pressure signals with different characteristics according to preset waveform parameters to simulate the user's physiological operations under different conditions. The data acquisition module was responsible for digitizing the pressure signals. The data processing software integrated two algorithm paths: Path A implemented a data admission verification mechanism that included activation timing features and signal stability judgment; Path B adopted a data processing method without admission verification, i.e., directly averaging the received signal over a preset time period. In the experiment, the setting of the key parameter, the steady-state threshold, was considered in balancing the strictness of data screening with the inclusiveness of data acceptance. A threshold that was too low might classify physiologically acceptable small fluctuations as unstable, while a threshold that was too high might introduce invalid data due to control instability. In this experiment, the threshold was determined based on statistical analysis of the coefficient of variation of multiple stable physiological signal samples and was set to 0.05.

[0036] After the experiment was initiated, the physiological signal simulation source was set to generate three signal waveforms sequentially, each waveform being generated ten times, and simultaneously sent to paths A and B for processing. The first signal was set to have a relatively long onset delay to simulate the situation caused by compensatory exertion of large non-target muscle groups. The second signal had an onset delay that conformed to the template, but after entering the plateau phase, it exhibited continuous oscillations in amplitude within a certain range to simulate the situation caused by the user's hesitation or unstable control. The third signal had an onset delay that conformed to the template and a stable output plateau phase. During data acquisition, when path A processed the first signal, its activation timing feature analysis module determined all ten signals to be invalid because it detected that the onset delay exceeded the preset template range. When processing the second signal, the signal was sent to the stability verification stage because it passed the timing verification. This stage detected its coefficient of variation. If the steady-state threshold remains above 0.05, all ten signals are deemed invalid. Only when processing the third signal are both verification steps of path A passed, and these ten signals are confirmed as valid physiological data. In contrast, path B calculates and outputs the average pressure value within a preset time period when processing all three signals. Table 1 shows examples of the processing results of the two paths for the three signals in this experiment.

[0037] Table 1: Data table for validating the data access verification mechanism.

[0038]

[0039] Referring to Table 1, the data set output by path B contains distorted data generated by compensatory and unstable signals. If this set is used to build a mapping model between subjective perception and physiological execution, it will lead to model inaccuracies. For example, a physiological value that should represent a low subjective effort may be incorrectly anchored by a high value generated by compensation. In contrast, the data set output by path A, through its serial dual verification mechanism, filters out compensatory and unstable signals, retaining only valid signals. This phenomenon occurs because the activation time-series feature verification, as the identification link of signal source, and the signal stability verification, as the signal quality filtering link, work together to select data sources that meet the preset standards for subsequent modeling steps.

[0040] Example 3: This example combines Figures 1 to 3 This paper describes the implementation of a method for generating training schemes based on dynamic physiological data and user instructions. Figure 1 As shown, this process involves five interactive entities: the user, the system, the fatigue monitoring module, the compensation calculation module, and the task generator. After the first training task is completed, the system records the actual physiological data and transmits it to the fatigue monitoring module as the initial baseline value for storage. In the subsequent second training task, when the user performs a task with the same perceptual target, the system acquires the current real-time physiological data and sends it to the fatigue monitoring module along with the initial baseline. This module then requests the compensation calculation module to process the data. The compensation calculation module updates a fatigue compensation factor by calculating the difference between the two sets of physiological data and returns the factor. When the system requests a new task from the task generator for the third training task, the task generator applies this fatigue compensation factor to adjust the target range of the physiological data and returns the adjusted task parameters to the system. Based on this, the system issues the compensated training task to the user. In the continuous monitoring cycle of training, this process is dynamically repeated. The fatigue monitoring module continuously updates the compensation factor, and the task generator adjusts the target range in real time to ensure that the training load matches the user's immediate capabilities.

[0041] like Figure 2 As shown in the figure, the horizontal axis represents the number of training tasks, and the vertical axis represents the physiological data values, in units of... The figure contains three curves. The solid line, marked as physiological data without compensation, shows that without compensation, the user's physiological output shows a significant and continuous downward trend due to fatigue accumulation as the number of training sessions increases. The dashed line, marked as the dynamically adjusted target range, represents the target range of physiological data after the system identifies fatigue and dynamically lowers it based on the fatigue compensation factor. Its downward trend is more gradual than the uncompensated data, forming a training target that matches the user's immediate ability. The dotted line, marked as the actual data after compensation, closely follows the dynamically adjusted target range, indicating that after introducing the compensation mechanism, the user can continuously and effectively achieve the adjusted training target throughout the training session, thus verifying the effectiveness of the mechanism in resolving training interruptions or target failures caused by fatigue.

[0042] like Figure 3 As shown, this method begins with the user performing a physiological operation based on a preset subjective perception label (such as mild or moderate). The resulting physiological data stream first enters the data admission verification module for screening. This verification includes two sequential steps. The first step is activation time-series feature analysis, used to determine whether the onset delay and peak slope conform to a preset template. If they do not conform, it is determined to be a compensatory signal and the data is rejected. This step aims to ensure the purity of the signal source. Data that passes the first step verification will enter the second step of stability verification. This step calculates the coefficient of variation of the physiological data within a sliding window. It also determines whether the value remains below a preset threshold to confirm whether the user's intent is clear and stable. If the value is too high, the data is rejected. In addition, a system in-situ self-verification mechanism can also be run before modeling to ensure the reliability of the sensor's physical state. Only data that passes both verifications is confirmed as valid physiological data and, together with the corresponding subjective perception label, is used in the personalized mapping model building module to generate a user-specific perception-physiological mapping model. The training task generation and feedback module generates a training task with subjective perception as the target based on this model, and compares the real-time physiological data with the target range parsed by the model during user execution, providing collaborative correction feedback through non-numerical graphical interface changes. At the same time, this model is also the basis of the dynamic closed-loop adjustment system, which can calculate fatigue compensation factors, trigger periodic model updates, and perform neuromuscular agility assessments, thus forming a complete adaptive closed-loop training process.

[0043] Example 4: In a scenario where a device is first delivered to a user for personalized configuration, to ensure that subsequent training schemes are based on a data model reflecting the user's individual characteristics, an initial system calibration process is executed. This process aims to address the issue that the control parameters and templates of multiple core algorithm modules within the system cannot be effectively personalized if fixed preset values ​​are used, as their applicable values ​​vary with the user or the physical state of the sensors. The calibration process first performs baseline feature acquisition for the system's self-verification mechanism in place. The user places the device in a predetermined position on their body and remains stationary. The system drives its internal excitation source to apply standardized mechanical micro-perturbation signals to the sensors carrying physiological data, and simultaneously acquires response signals using these sensors. The response signal undergoes rapid transformation, and the system obtains a response feature characterizing the current physical coupling state between the sensor and the body, namely, a spectral data containing the position and amplitude of the resonance peak. This response feature is stored as a benchmark feature for subsequent comparison. Subsequently, the process enters the personalized calibration stage of the activation timing feature template. Under guidance, the user repeatedly performs the activation of the target muscle group marked as the reference action several times. The system extracts the onset delay and peak slope of the physiological data stream generated by each activation, and performs statistical analysis on the collected multiple sets of feature values, calculating their mean and standard deviation, thereby generating an activation timing feature template specific to the user in this device state, including upper and lower limit thresholds. This template is used for the identification of compensatory behavior in subsequent data admission verification.

[0044] After the equipment enters the long-term use phase, in order to adapt the fatigue compensation response speed to the user's physiological characteristics, the fatigue compensation factor in the dynamic adjustment mechanism used to compensate for intra-session fatigue has a smoothing coefficient in its update algorithm. An adaptive adjustment logic is employed; the system periodically analyzes the decay rate of the user's physiological output in all training sessions within that period. If the average decay rate exceeds a preset baseline rate, the system will increase the rate according to a preset function. The value is positive, and vice versa. Meanwhile, to enable the personalized mapping model to evolve with the user's capabilities, its dynamic update triggering conditions, in addition to a fixed time period, also include a judgment logic based on performance changes. The system continuously tracks the changes in the user's perceptual control gain index. When the latest value of this index exceeds the preset update threshold compared to the value at the last calibration, the system will also initiate and guide the user to re-execute the personalized mapping model establishment process. By executing the above process, which includes initial calibration and long-term adaptive adjustment, the key parameters, thresholds, and templates in the system are changed from using general preset values ​​to being determined or dynamically adjusted through personalized and reproducible processes. This ensures that the values ​​of multiple variables affecting system performance are constrained by a series of processes with clear inputs, processing procedures, and outputs.

[0045] Example 5: Before a training task begins, when the system executes the system positioning self-verification mechanism, if the difference between the real-time acquired response features and the stored baseline features exceeds a preset threshold, and the difference remains even after the user adjusts the device position multiple times as prompted, the system will terminate the normal startup process. At this time, the system provides the user with an option to confirm whether a new baseline feature needs to be established for the current physical state. If the user confirms, the system guides the user to execute a complete baseline feature acquisition process and stores the newly generated response features as the current and subsequent baseline features, thereby completing the recalibration of the system positioning state under the current physical state.

[0046] In another scenario, during the personalized mapping model building process, after the user has completed the anchoring of corresponding valid physiological data for the three subjective perception labels of mild, moderate, and severe, the system will first perform an internal logical consistency check on the acquired multiple data pairs, subjective perception labels, and valid physiological data before executing the curve fitting algorithm. The specific steps of this check are to verify whether the valid physiological data values ​​bound to the subjective perception labels of different intensity levels follow a monotonically increasing relationship, that is, the physiological data value corresponding to the mild label is less than the physiological data value corresponding to the moderate label. If the system detects that this logical relationship does not hold, it will pause the model building and point out the contradictory data anchor points to the user, guiding them to re-perform physiological operations and data anchoring only for the contradictory subjective perception labels, until all data anchor points meet the logical consistency before executing the subsequent model generation steps.

[0047] Example 6: In a scenario for fine-tuning training programs, a standardized setup process for interaction parameters and assessment procedures is implemented to address the issue that generic settings for feedback sensitivity or assessment completion conditions may not be suitable for individual user differences. This setup process first parameterizes the changes in the graphical user interface (GUI) for collaborative correction feedback. The system establishes a mapping function that correlates the deviation of real-time physiological data with the degree of GUI morphological change. To prevent users from overreacting to minor natural fluctuations in physiological signals, a zero-feedback stable interval is set in this mapping function. That is, when the deviation is less than a preset stable interval threshold, the GUI morphology remains unchanged. When the deviation exceeds this stable interval, the degree of morphological change is calculated based on a non-linear function. This function maps larger deviations to more visually significant changes while suppressing smaller deviations. In this process, the size of the stable interval threshold and the slope parameter of the non-linear function can be adjusted by rehabilitation instructors based on user feedback.

[0048] Subsequently, the process defines the measurement protocol within the active detection mechanism used to assess the user's neuromuscular agility. When the system issues a command to the user to transition from the current physiological execution state to another target state, the other target state is defined as a physiological data target range surrounding the target physiological data value, rather than a single numerical point. The measurement of response time begins at the moment the command is issued, and the termination condition is set when the real-time collected physiological data enters the physiological data target range and is maintained continuously within that range for more than a preset minimum duration, which can be set to 200 milliseconds, before confirming that the user has stably reached the target state and stopping the timer. This measurement protocol, by introducing a target range and a minimum duration, reduces the possibility of incorrectly recording response time due to instantaneous fluctuations in physiological signals. By executing the above process, the operation of the human-computer interaction and evaluation-related aspects of the system is defined by a set of clearly configurable parameters and judgment conditions, thereby enabling the training program to be adjusted according to the user's specific needs.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for generating training schemes based on dynamic physiological data and user instructions, characterized in that, The method includes the following steps: Step a: Receive the physiological data stream generated when the user performs physiological operations on preset subjective perception tags; Step b: Before establishing the subsequent personalized mapping model, a data admission verification is performed on the physiological data stream. This data admission verification includes: analyzing the activation timing characteristics of the physiological data stream during the initiation phase; wherein the activation timing characteristics include the onset delay from the issuance of the command to the first breakthrough of the resting baseline noise threshold by the physiological data, and the peak slope represented by the reciprocal of the time taken for the physiological data to rise from the onset point to the first peak; and, only when the activation timing characteristics conform to a template representing the pure activation of the target muscle group, wherein the template is the numerical range of the onset delay and the peak slope, is the physiological data stream authorized for subsequent stability verification; real-time monitoring of the stability index of the verified physiological data stream; and, only when the stability index meets the stability condition is the physiological data stream confirmed as valid physiological data for modeling. The method for determining the stability index and judging the stability condition includes: continuously calculating the coefficient of variation (CV) of the physiological data stream within a sliding time window, wherein... , The standard deviation of physiological data within the sliding time window. The value is the average of the physiological data within the sliding time window; and when the value of the coefficient of variation (CV) is consistently below the steady-state threshold, the stability condition is considered to be met. Step c: Based on subjective perception labels and effective physiological data, establish a personalized mapping model that represents the correspondence between the user's subjective perception and physiological execution. The method for establishing the personalized mapping model includes: acquiring at least two subjective perception labels and their corresponding effective physiological data, wherein the subjective perception labels are terms that represent different intensity levels of the user's perceived force; and generating the personalized mapping model by applying a curve fitting algorithm to at least two subjective perception labels and their corresponding effective physiological data. Step d: Based on the training logic and personalized mapping model, generate a training task that includes subjective perception target, and determine the range of physiological data target corresponding to the subjective perception target. Step e involves comparing the real-time collected physiological data with the target range of physiological data during the user's training task and generating collaborative correction feedback.

2. The method for generating a training scheme based on dynamic physiological data and user instructions according to claim 1, characterized in that, The collaborative correction feedback in step e is a non-numerical graphical user interface change. The degree of change in the form of the graphical user interface is determined by the deviation between the real-time physiological data and the target range of the physiological data.

3. The method for generating training schemes based on dynamic physiological data and user instructions according to claim 1, characterized in that, Following step e, a dynamic adjustment mechanism for compensating for in-session fatigue is also included. This mechanism includes: acquiring the corresponding actual physiological data for each training task successfully executed by the user; updating the fatigue compensation factor based on the difference between the actual physiological data and the initial physiological data recorded in the personalized mapping model; and applying the fatigue compensation factor to adjust the target range of the physiological data to be generated when generating the next training task.

4. The method for generating a training scheme based on dynamic physiological data and user instructions according to claim 1, characterized in that, The method further includes: when the preset update triggering condition is met, repeating steps a to c to dynamically update the personalized mapping model.

5. The method for generating a training scheme based on dynamic physiological data and user instructions according to claim 1, characterized in that, The method also includes an active detection mechanism for assessing the user's neuromuscular agility, which includes: issuing an instruction to the user during the execution of a training task to transition from the current physiological execution state to another target state; measuring the response time taken from the time the user receives the instruction until the real-time physiological data reaches the physiological data value corresponding to the other target state; and using the response time as an indicator for assessing the user's neuromuscular agility.

6. The method for generating a training scheme based on dynamic physiological data and user instructions according to claim 1, characterized in that, The method further includes, after step c: calculating a perceptual control gain index that characterizes the slope of the personalized mapping model; and using the perceptual control gain index as an independent evaluation parameter for tracking changes in the user's neuromuscular control efficiency.

7. The method for generating a training scheme based on dynamic physiological data and user instructions according to claim 1, characterized in that, The method also includes a system in-situ state self-verification mechanism before performing step c. This mechanism includes: applying a standardized mechanical micro-perturbation signal to the sensor carrying physiological data through an excitation source; using the sensor to collect the response signal to the mechanical micro-perturbation signal and generating a response feature characterizing the system in-situ state; comparing the response feature with a reference feature acquired during initial calibration; and performing step c only when the difference between the response feature and the reference feature is less than a threshold.

Images