A method and system for adaptive regulation of exercise load

Abstract

The application discloses a kind of motion load adaptive regulation and control method and system, it includes the electrocardiogram signal data and motion state data of target object in the rehabilitation exercise process;Wavelet multilayer decomposition and wavelet reconstruction processing are successively carried out to electrocardiogram signal data, baseline correction signal is obtained;Motion state data is input as reference noise, and non-stationary artifact is cancelled to baseline correction signal based on variable step adaptive filtering algorithm, and high-fidelity electrocardiogram signal is obtained;Respectively to high-fidelity electrocardiogram signal and motion state data are carried out feature extraction, and myocardial ischemia correlation feature and motion correlation feature are obtained;Myocardial ischemia correlation feature, motion correlation feature and target object portrait vector are spliced and handled, and multidimensional risk vector is constructed;Multidimensional risk vector is input in fuzzy reasoning model, and the current risk level of target object is obtained, and corresponding load regulation instruction is generated.The method of the application improves the safety and effectiveness of rehabilitation training.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a method and system for adaptive control of motion load. Background Technology

[0002] With the continuous development of cardiac rehabilitation medicine, exercise therapy has become a core means of rehabilitation intervention for patients with cardiovascular diseases. During rehabilitation exercise, real-time and accurate monitoring of the patient's electrophysiological status and dynamic adjustment of exercise load accordingly are crucial to ensuring patient safety and avoiding exercise-induced adverse cardiovascular events.

[0003] Currently, existing cardiac rehabilitation monitoring systems typically collect patients' electrocardiogram signals through equipment and mainly rely on single indicators such as heart rate or preset fixed physiological thresholds to assess current exercise risk, and then issue load adjustment instructions to rehabilitation equipment such as treadmills and exercise bikes.

[0004] However, the physiological response of the human body during exercise is highly nonlinear and ambiguous, and the basic cardiopulmonary function of different patients varies greatly. Relying solely on a single indicator and a uniform rigid threshold for assessment ignores the coupling relationship between multidimensional physiological characteristics, exercise status and individualized baseline, which can easily lead to misjudgment of risk and thus make it impossible to achieve safe and accurate adaptive regulation of exercise load. Summary of the Invention

[0005] This invention provides a method and system for adaptive regulation of exercise load, which improves the safety of rehabilitation training.

[0006] To address the aforementioned technical problems, the present invention provides, in one aspect, a method for adaptive control of exercise load, comprising: Acquire electrocardiogram (ECG) signal data and motion status data of the target subject during rehabilitation exercises; The electrocardiogram signal data is subjected to wavelet multi-layer decomposition and wavelet reconstruction processing in sequence to obtain the baseline correction signal; Using the motion state data as a reference noise input, the baseline correction signal is non-stationary artifact cancellation is performed based on a variable step size adaptive filtering algorithm to obtain a high-fidelity electrical signal. Feature extraction was performed on the high-fidelity electrical signal and the motion state data respectively to obtain myocardial ischemia-related features and motion-related features; Obtain individualized baseline parameters of the target object and encode the individualized baseline parameters into a target object profile vector; concatenate the myocardial ischemia-related features, the motion-related features, and the target object profile vector to construct a multidimensional risk vector; The multidimensional risk vector is input into a preset fuzzy inference model to obtain the current risk level of the target object, and a load control instruction corresponding to the current risk level is generated.

[0007] As one preferred embodiment, the step of sequentially performing wavelet multi-level decomposition and wavelet reconstruction processing on the electrocardiogram signal data to obtain a baseline correction signal includes: Wavelet multi-level decomposition was performed on the electrocardiogram signal data to obtain high-frequency detail coefficients and low-frequency approximation coefficients at each scale; The high-frequency detail coefficients and the low-frequency approximation coefficients are processed respectively to obtain denoised high-frequency detail coefficients and drift-free low-frequency approximation coefficients. Wavelet reconstruction is then performed on the denoised high-frequency detail coefficients and the drift-free low-frequency approximation coefficients to obtain the baseline correction signal.

[0008] As one preferred embodiment, the step of using the motion state data as a reference noise input and performing non-stationary artifact cancellation on the baseline correction signal based on a variable step-size adaptive filtering algorithm to obtain a high-fidelity electrical signal includes: The gravity component of the triaxial acceleration signal in the motion state data is removed to obtain the triaxial dynamic acceleration signal, and the first instantaneous motion intensity at the current moment is determined based on the triaxial dynamic acceleration signal. A reference input vector is constructed based on the first instantaneous motion intensity and the second instantaneous motion intensity at a historical moment; The baseline correction signal is used as the desired input signal, and the reference input vector is weighted according to the current weight vector of the adaptive filter to obtain the artifact estimation signal. The difference between the desired input signal and the artifact estimation signal is calculated to determine the filtering error signal, which represents the high-fidelity electrical signal after removing non-stationary motion artifacts.

[0009] As one preferred embodiment, after determining the filtering error signal based on the difference between the desired input signal and the artifact estimation signal, the method further includes: Based on the first instantaneous motion intensity and the amplitude of the filtering error signal, the step size factor of the adaptive filter is dynamically updated; wherein, the updated value of the step size factor is positively correlated with the first instantaneous motion intensity and the amplitude of the filtering error signal. Based on the step size factor, the filtering error signal, and the reference input vector, the current weight vector of the adaptive filter is iteratively updated to obtain the target weight vector at the next time step, wherein the target weight vector is used to generate the artifact estimation signal at the next time step.

[0010] As one preferred embodiment, the step of inputting the multidimensional risk vector into a preset fuzzy inference model to obtain the current risk level of the target object includes: The multidimensional risk vector is normalized, and the features of each dimension in the normalized multidimensional risk vector are mapped to the corresponding fuzzy set through the Gaussian membership function to obtain the fuzzy input vector. Based on a preset fuzzy rule base, rule reasoning is performed on the fuzzified input vector to determine the trigger strength of each fuzzy rule, and the conclusion fuzzy set output by all fuzzy rules is aggregated based on the trigger strength to obtain an aggregated risk fuzzy set. The aggregated risk fuzzy set is defuzzified to obtain continuous risk scores; The continuous risk score is mapped to a preset numerical range to determine the current risk level of the target object.

[0011] Another aspect of the present invention provides a motion load adaptive control system, comprising: The acquisition module is used to acquire electrocardiogram signal data and motion status data of the target object during rehabilitation exercise; The preprocessing module is used to sequentially perform wavelet multi-layer decomposition and wavelet reconstruction processing on the electrocardiogram signal data to obtain the baseline correction signal; The correction module is used to take the motion state data as a reference noise input and perform non-stationary artifact cancellation on the baseline correction signal based on a variable step size adaptive filtering algorithm to obtain a high-fidelity electrical signal. The extraction module is used to extract features from the high-fidelity electrical signal and the motion state data respectively to obtain myocardial ischemia-related features and motion-related features; The splicing module is used to obtain the individualized baseline parameters of the target object and encode the individualized baseline parameters into a target object profile vector; the myocardial ischemia-related features, the motion-related features and the target object profile vector are spliced ​​together to construct a multidimensional risk vector; The control module is used to input the multidimensional risk vector into a preset fuzzy inference model to obtain the current risk level of the target object and generate a load control instruction corresponding to the current risk level.

[0012] As one preferred embodiment, the correction module is specifically used for: Wavelet multi-level decomposition was performed on the electrocardiogram signal data to obtain high-frequency detail coefficients and low-frequency approximation coefficients at each scale; The high-frequency detail coefficients and the low-frequency approximation coefficients are processed respectively to obtain denoised high-frequency detail coefficients and drift-free low-frequency approximation coefficients. Wavelet reconstruction is then performed on the denoised high-frequency detail coefficients and the drift-free low-frequency approximation coefficients to obtain the baseline correction signal.

[0013] As one preferred embodiment, the correction module is specifically used for: The gravity component of the triaxial acceleration signal in the motion state data is removed to obtain the triaxial dynamic acceleration signal, and the first instantaneous motion intensity at the current moment is determined based on the triaxial dynamic acceleration signal. A reference input vector is constructed based on the first instantaneous motion intensity and the second instantaneous motion intensity at a historical moment; The baseline correction signal is used as the desired input signal, and the reference input vector is weighted according to the current weight vector of the adaptive filter to obtain the artifact estimation signal. The difference between the desired input signal and the artifact estimation signal is calculated to determine the filtering error signal, which represents the high-fidelity electrical signal after removing non-stationary motion artifacts.

[0014] As one preferred embodiment, the correction module is further configured to: Based on the first instantaneous motion intensity and the amplitude of the filtering error signal, the step size factor of the adaptive filter is dynamically updated; wherein, the updated value of the step size factor is positively correlated with the first instantaneous motion intensity and the amplitude of the filtering error signal. Based on the step size factor, the filtering error signal, and the reference input vector, the current weight vector of the adaptive filter is iteratively updated to obtain the target weight vector at the next time step, wherein the target weight vector is used to generate the artifact estimation signal at the next time step.

[0015] As one preferred embodiment, the control module is specifically used for: The multidimensional risk vector is normalized, and the features of each dimension in the normalized multidimensional risk vector are mapped to the corresponding fuzzy set through the Gaussian membership function to obtain the fuzzy input vector. Based on a preset fuzzy rule base, rule reasoning is performed on the fuzzified input vector to determine the trigger strength of each fuzzy rule, and the conclusion fuzzy set output by all fuzzy rules is aggregated based on the trigger strength to obtain an aggregated risk fuzzy set. The aggregated risk fuzzy set is defuzzified to obtain continuous risk scores; The continuous risk score is mapped to a preset numerical range to determine the current risk level of the target object.

[0016] Compared with the prior art, the beneficial effects of the present invention are at least one of the following: 1) At the signal processing level, this invention effectively solves the technical challenges of ECG signals being easily interfered with and key features being easily distorted during motion. Specifically, this invention first uses wavelet multi-layer decomposition technology to separate the ECG signal in the frequency domain, denoising high-frequency details and approximating low-frequency drift and reconstructing the signal, thus initially eliminating baseline drift interference. Based on this, it innovatively uses motion data as reference noise input and employs a variable step-size adaptive filtering algorithm to cancel non-stationary artifacts. This cascaded mechanism of "wavelet preprocessing + adaptive filtering" can dynamically adjust the filtering step size according to changes in motion intensity, thoroughly filtering out complex non-stationary motion artifacts while preserving key myocardial ischemia features such as ST segment changes with high fidelity to the greatest extent, significantly improving the quality of dynamic ECG signals and providing an extremely reliable data foundation for subsequent risk assessment.

[0017] 2) At the level of risk assessment and control, this invention overcomes the shortcomings of traditional single-indicator and fixed-threshold assessments, which are prone to misjudgment, and achieves highly personalized and precise adaptive load control. Specifically, this invention deeply integrates high-fidelity cardiac electrical signal characteristics, real-time motion state characteristics, and individualized baseline parameters of the target object to construct a multi-dimensional risk vector that comprehensively represents the patient's state; and further inputs it into a preset fuzzy inference model. Because the fuzzy inference model can perfectly match the nonlinear and boundary fuzziness characteristics of human physiological responses and fully consider individual differences among patients, it can more scientifically and accurately quantify the current risk level. The load control instructions generated based on this can match the patient's true tolerance in real time and dynamically, effectively avoiding excessive or insufficient exercise load caused by risk misjudgment, and greatly improving the safety and clinical intervention effect of cardiac rehabilitation exercise. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the adaptive control method for motion load in one embodiment of the present invention; Figure 2 This is a structural block diagram of the motion load adaptive control system in one embodiment of the present invention; Figure label: Among them, 11 is the acquisition module; 12 is the preprocessing module; 13 is the correction module; 14 is the extraction module; 15 is the splicing module; and 16 is the control module. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] In the description of this invention, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0021] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0022] One embodiment of the present invention provides a method for adaptive regulation of exercise load. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The diagram shown is a flowchart of an adaptive motion load control method according to one embodiment of the present invention, which includes steps S1-S6: S1: Acquire electrocardiogram signal data and motion status data of the target subject during rehabilitation exercise.

[0023] During cardiac rehabilitation exercises following percutaneous coronary intervention (PCI), patients' electrocardiogram (ECG) signals are highly susceptible to interference from factors such as limb movement, respiration, and muscle contraction. Furthermore, there is a strong temporal correlation between changes in exercise status and ECG abnormalities. Therefore, simultaneously acquiring high-quality ECG signal data and accurate exercise status data is a prerequisite for achieving adaptive load regulation.

[0024] It should be noted that the target group in this embodiment specifically refers to post-PCI patients who are in the cardiac rehabilitation period, wearing wearable monitoring devices and following a prescribed exercise prescription. This invention is not limited to this, and the method is also applicable to exercise monitoring scenarios for other cardiovascular disease rehabilitation populations or healthy populations.

[0025] Specifically, the acquisition of electrocardiogram (ECG) signal data adopts a multi-lead dynamic acquisition method. Preferably, at least three leads (such as modified V1, V5, and aVF leads) of analog ECG signals are continuously acquired at a sampling rate of not less than 250Hz by using a dry or wet electrode array attached to the chest of the target subject, and the digital ECG signals are obtained after front-end analog filtering and analog-to-digital conversion.

[0026] In this embodiment, to balance the skin sensitivity and signal quality of patients after PCI, textile-based flexible electrodes can be used to reduce motion friction noise. It should be noted that the number of leads, sampling rate, and electrode type can be adjusted according to the actual rehabilitation scenario and hardware conditions. This invention does not impose any limitations on these, as long as the electrocardiogram waveforms containing key morphological features of myocardial ischemia, such as the ST segment and T wave, can be captured.

[0027] Furthermore, motion state data is acquired through an inertial measurement unit (IMU) to characterize the target subject's body posture, motion intensity, and movement patterns during rehabilitation exercises. Preferably, the IMU is integrated into the electrocardiogram (ECG) acquisition device or worn independently in the core trunk region (such as the manubrium of the sternum), synchronously acquiring triaxial acceleration, triaxial angular velocity, and triaxial magnetometer data with the same sampling clock as the ECG signal, at a sampling rate of not less than 50Hz.

[0028] In this embodiment, placing the IMU on the torso rather than the limbs can effectively avoid local high-frequency vibration interference caused by upper limb arm swinging or lower limb stepping, making the motion state data more accurately reflect the overall motion load of the whole body, thereby providing a cleaner reference noise input for subsequent variable step size adaptive filtering.

[0029] To ensure strict time-domain alignment between ECG signal data and motion state data, this embodiment employs a hardware-level time synchronization mechanism. Specifically, the ECG acquisition module and the IMU module share the system clock of the same main control chip, or microsecond-level timestamps are achieved through a dedicated synchronization trigger line. It should be noted that if a distributed sensor architecture is used, software alignment can also be achieved through the clock synchronization service in the Bluetooth Low Energy (BLE) protocol stack, but additional compensation for transmission delay is required. This invention does not limit the specific synchronization implementation method; the core is to ensure that the time deviation between the two types of data is less than 1 / 10 of a single heartbeat cycle, to avoid motion artifact cancellation failure or the introduction of new artificial artifacts due to timing misalignment.

[0030] In terms of ECG signal data preprocessing, this embodiment applies anti-aliasing low-pass filtering and power frequency notch filtering at the acquisition end to suppress environmental electromagnetic interference. Preferably, the analog front end is equipped with a fourth-order Butterworth low-pass filter with a cutoff frequency of 40Hz and a 50Hz / 60Hz adaptive notch filter to maximize the signal-to-noise ratio while ensuring that the low-frequency components of the ST segment are not distorted. Simultaneously, the raw ECG data undergoes real-time quality assessment. When electrode detachment, saturation, or persistent baseline drift exceeding a preset threshold is detected, the data for that period is automatically marked as invalid, triggering re-acquisition or an alarm. It should be noted that the above preprocessing parameters can be adaptively configured according to the dynamic characteristics of different exercise types (such as cycling, upper limb power meter, and resistance band training), and this invention does not limit this.

[0031] For the preprocessing of motion state data, this embodiment first performs zero-bias calibration and coordinate system one adjustment on the raw IMU data. Specifically, 3 seconds of data are collected while the target object is standing still to calculate the gyroscope zero bias and subtract it in real time; then, the sensor body coordinate system is transformed to the human anatomical coordinate system (front-back, left-right, up-down) using a rotation matrix, so that the motion characteristics have physiological significance. Preferably, the acceleration data can also be separated by gravity components to extract pure dynamic acceleration as a characterization of motion intensity. It should be noted that the installation angle required for coordinate transformation can be obtained through static calibration or dynamic self-calibration algorithms, which are not limited in this invention, as long as the final output motion state data can accurately reflect the actual body movements of the target object during rehabilitation exercises.

[0032] Finally, this embodiment encapsulates the synchronized and preprocessed ECG signal data and motion state data into a unified data frame structure for subsequent steps. Each data frame includes a timestamp, multi-channel ECG sample values, raw values ​​from the nine-axis IMU, and data quality flags. In this embodiment, the data frame length can be set to 1 second or one complete heartbeat cycle to balance real-time performance and computational efficiency.

[0033] S2: Perform wavelet multi-layer decomposition and wavelet reconstruction on the ECG signal data in sequence to obtain the baseline correction signal.

[0034] When patients perform rehabilitation exercises such as cycling or treadmill work, changes in respiratory rate and body displacement can cause severe low-frequency baseline drift, while muscle contraction introduces high-frequency electromyographic interference. Traditional fixed bandpass filtering, while denoising, can easily weaken or distort the ST segment and T wave morphology, leading to misdiagnosis of myocardial ischemia characteristics. Wavelet transform, with its excellent multi-resolution time-frequency analysis capabilities, can decompose the signal into different frequency bands, achieving effective separation of baseline components from key ECG morphological components.

[0035] It should be noted that this embodiment uses wavelet multi-layer decomposition and reconstruction as the first step in cascaded artifact suppression. Its purpose is to initially reduce band-separable interference while preserving the QRS group, ST segment and T wave morphology, so as to provide a high-quality input signal for subsequent motion vector-based adaptive filtering.

[0036] Specifically, before performing wavelet multilevel decomposition on the ECG signal data, anti-aliasing preprocessing and power frequency notch filtering are first applied to each ECG lead signal from 0.05Hz to 150Hz to filter out environmental electromagnetic interference. Subsequently, within a sliding window of length N, Symlet wavelet basis functions are used to perform multilevel decomposition on the preprocessed ECG signal.

[0037] Preferably, this embodiment uses the Sym8 wavelet basis and performs an 8-level decomposition, the decomposition process of which can be represented as follows: in, For the first ECG signals from leads, The low-frequency approximation coefficients at the 8th scale reflect baseline drift caused by factors such as respiration; For the first High-frequency detail coefficients at the scale, where, , The primary parameters reflect electromyographic interference and high-frequency noise, while the detail coefficients at intermediate scales correspond to the energy distribution of the QRS complex and ST-T waves. It should be noted that the wavelet basis type and the number of decomposition layers can be adaptively adjusted according to the actual sampling rate and motion type; this invention does not impose any limitations on these parameters.

[0038] Furthermore, the high-frequency detail coefficients and low-frequency approximation coefficients obtained from the decomposition are processed in a targeted manner.

[0039] For high-frequency detail coefficients, this embodiment employs a tradeoff between hard and soft thresholding for denoising, overcoming the signal discontinuity issues caused by traditional hard thresholding and the amplitude attenuation problems caused by soft thresholding. Preferably, the thresholding function is defined as: when hour, ;when hour, Among them, the regulatory factor Thresholds at various scales Through formula Calculate, where the noise standard deviation estimate is... Through the analysis of , The high-frequency detail coefficients are subjected to the above threshold processing to obtain denoised high-frequency detail coefficients that effectively suppress electromyographic noise.

[0040] For low-frequency approximation coefficients This embodiment uses low-pass smoothing or zeroing followed by reconstruction with intermediate scaling coefficients to remove baseline drift components, thereby obtaining drift-free low-frequency approximation coefficients. In an optional implementation, the original signal can also be directly subtracted from the original signal... Reconstructed baseline signal Baseline correction is then achieved. Subsequently, using the processed denoised high-frequency detail coefficients and drift-free low-frequency approximation coefficients, inverse wavelet transform (i.e., wavelet reconstruction) is performed using the Mallat algorithm to obtain the baseline-corrected signal. .

[0041] This baseline correction signal significantly reduces baseline drift and high-frequency electromyography while fully preserving the ST segment and T wave low-frequency morphological features required for myocardial ischemia assessment. It should be noted that the specific form of the threshold function, the threshold calculation method, and the low-frequency coefficient processing strategy can all be dynamically configured according to signal quality. This invention does not impose limitations on these aspects, as long as the final reconstructed baseline correction signal meets the input requirements for subsequent adaptive filtering.

[0042] S3: Using motion state data as reference noise input, non-stationary artifact cancellation is performed on the baseline correction signal based on the variable step size adaptive filtering algorithm to obtain a high-fidelity electrical signal.

[0043] Because motion artifacts generated by limb swinging, trunk displacement, and muscle contraction during rehabilitation training on exercise bikes or treadmills are non-stationary, broadband, and aliased with the ST-T band spectrum of electrocardiograms, traditional fixed-parameter filters struggle to balance convergence speed and steady-state error. Therefore, this embodiment innovatively uses synchronously acquired motion state data as reference noise input and employs a variable-step-size adaptive filtering algorithm to dynamically and accurately cancel non-stationary artifacts. It should be noted that this step, together with wavelet preprocessing, constitutes a cascaded artifact suppression strategy. The former is responsible for separating band-separable baseline drift and high-frequency electromyography, while the latter specifically performs adaptive elimination of motion-related artifacts. Together, they ensure high fidelity of the final output signal.

[0044] Specifically, the gravity component is first removed from the triaxial acceleration signal in the motion state data. In this embodiment, a low-pass filter or attitude calculation method is used to separate the static gravity acceleration vector from the original triaxial acceleration, and this vector is subtracted from the original signal to obtain a triaxial dynamic acceleration signal that only reflects the dynamic motion of the body. , , .

[0045] Subsequently, the first instantaneous motion intensity at the current moment is calculated based on the triaxial dynamic acceleration signal. This instantaneous motion intensity can quantify the intensity of the patient's rehabilitation exercises in real time, providing a crucial basis for subsequent construction of reference input vectors and dynamic adjustment of the filter step size. It should be noted that the method for removing the gravity component and the calculation method for the instantaneous motion intensity can be adjusted according to the sensor installation location and the type of motion; this invention does not impose any limitations on these aspects.

[0046] Furthermore, based on the motion intensity at the first instant and the motion intensity at the second instant from previous times, a reference input vector for adaptive filtering is constructed. Preferably, the instantaneous motion intensities at the current time and the previous M-1 times are arranged in chronological order to form an M-dimensional reference input vector. .

[0047] This vector not only contains current motion state information but also captures the phase lag characteristics of motion artifacts relative to mechanical motion by introducing a historical delay term, thereby more accurately modeling the artifact generation mechanism. In this embodiment, the delay order M can be dynamically set according to the sampling rate and the duration of typical motion artifacts; for example, M=4~8 at a 50Hz sampling rate. It should be noted that the dimension and constituent elements of the reference input vector (such as whether it contains angular velocity or magnetometer data) can be flexibly configured according to the actual artifact correlation analysis results, and this invention does not impose any limitations on them.

[0048] Next, the baseline correction signal is used as the desired input signal. Based on the current weight vector of the adaptive filter For the reference input vector Weighted calculations are performed to obtain the artifact estimation signal. Subsequently, the difference between the desired input signal and the artifact estimation signal is calculated to determine the filtering error signal. The filtered error signal represents the high-fidelity electrical signal after removing non-stationary motion artifacts, preserving the complete morphological characteristics associated with myocardial ischemia. In this embodiment, the adaptive filter adopts a transverse FIR structure, and its output is directly used as a real-time estimate of artifacts. The weights are continuously optimized through an error feedback mechanism to minimize artifact energy.

[0049] Furthermore, this embodiment introduces a dynamic update mechanism for the step size factor based on motion intensity and error amplitude. Specifically, the updated value of the step size factor... With the intensity of motion at the first instant and the amplitude of the filtering error signal They are positively correlated, and their update formula can be expressed as: in, and These are the lower and upper limits of the step size, respectively. To prevent division by zero regularization constants, the step size is automatically increased to accelerate convergence when the motion is vigorous or the error is large; conversely, the step size is automatically decreased when the motion is gentle or the error is small to reduce steady-state misalignment and preserve ECG signal details. This variable step size strategy effectively resolves the contradiction between convergence speed and steady-state accuracy in non-stationary motion scenarios with a fixed step size.

[0050] Finally, based on the updated step size factor Filtering error signal and reference input vector The current weight vector of the adaptive filter is updated according to the LMS iteration rule to obtain the target weight vector for the next time step. The target weight vector will be used to generate a new artifact estimation signal in the next time step, thereby achieving online adaptive learning of the filter. In this embodiment, the initial value of the weight vector can be set to zero or obtained through offline calibration. The iteration process updates in real time at each sampling point to ensure millisecond-level response capability to rapidly changing motion artifacts. It should be noted that the above weight update formula, step size dynamic adjustment function and its parameters can be equivalently replaced or optimized according to hardware computing power and signal characteristics. This invention does not limit these modifications, as long as it can achieve variable step size adaptive artifact cancellation based on motion state.

[0051] S4: Feature extraction is performed on high-fidelity electrocardiogram (HFC) signals and motion state data respectively to obtain myocardial ischemia-related features and motion-related features.

[0052] In cardiac rehabilitation exercises, relying solely on resting electrocardiograms or a single heart rate threshold is insufficient to comprehensively reflect the myocardial oxygen supply imbalance under dynamic load, and the traditional fixed ST segment determination window cannot adapt to electrophysiological changes at different heart rates. Therefore, this embodiment employs a multi-dimensional feature extraction strategy to construct myocardial ischemia-related features and exercise-related features, respectively, providing a quantitative basis for subsequent multi-dimensional risk vector fusion and adaptive load regulation.

[0053] Specifically, when extracting myocardial ischemia-related features from high-fidelity electrical signals, R-peak detection and heartbeat segmentation are performed first. In this embodiment, the Pan-Tompkins algorithm, derivative square integral, or equivalent QRS detection algorithm can be used to locate the R-peak. Let the time of the i-th R-peak be . The interval between adjacent R peaks is Instantaneous heart rate is calculated as follows: Subsequently, the starting point of the QRS group was determined based on the slope thresholds before and after the R peak and the energy envelope. and the end point and point J Defined as the inflection point after the end of the QRS or For different heart rates, this embodiment dynamically uses 60ms or 80ms after the J point as the ST determination point to adapt to the influence of heart rate changes on the repolarization process. It should be noted that the R-peak detection algorithm and the J-point positioning method can be dynamically switched according to the actual signal quality, and this invention does not limit them.

[0054] Furthermore, this embodiment calculates ST segment offset and ST slope based on point J, and extracts T wave morphological features to improve the myocardial ischemia-related features. The formula for calculating ST segment offset is: in, , The PR or TP segment is the isoelectric baseline. The ST slope is used to characterize the morphological evolution trend of the ST segment, and its calculation formula is: Simultaneously, extract the T-wave amplitude. T-wave area T-wave polarity And used to characterize the alternation trend of T waves Among them, T-wave alternation is an important marker of microvolt-level electrical instability in the early stage of myocardial ischemia, and incorporating it into the feature space can significantly improve the sensitivity of identifying the risk of occult ischemia.

[0055] Regarding the extraction of heart rate variability and heart rate reserve features, this embodiment calculates the rate of change of heart rate. This reflects the autonomic nervous system's instantaneous regulatory capacity to exercise load. Furthermore, heart rate reserve (HRR) is introduced to quantify the patient's relative load level; its calculation formula is as follows: .in, Current heart rate, Resting heart rate Based on age, The use of beta-blockers and the maximum heart rate after correction to the prescribed target heart rate. Preferably, this correction mechanism can effectively avoid the use of... The underestimation of workload in patients undergoing PCI with beta-blockers due to limited heart rate rise makes heart rate reserve indicators more consistent with the true physiological state of postoperative patients.

[0056] Regarding motion-related feature extraction, this embodiment extracts motion intensity, cadence, motion vector norm, and perceived fatigue level (RPE) compensation index from motion state data to quantify the real-time physiological load and fatigue compensation status of the target subject during rehabilitation exercise. Preferably, when the patient does not actively input RPE, the RPE compensation index is estimated based on multidimensional motion and physiological indicators, and its calculation formula is as follows: in, To normalize the motion power, For tread frequency volatility, The blood oxygen level decreased. to The regression coefficients are individually calibrated. The function is used to limit the estimated values ​​to the range of 6 to 20 on the Borg scale. This RPE compensation index, combined with normalized exercise intensity, constitutes an exercise-related feature, effectively compensating for the limitations of a single heart rate index in assessing patients' subjective fatigue and exercise tolerance.

[0057] In this embodiment, the extraction of myocardial ischemia-related features and motion-related features is strictly synchronized in time. Specifically, using the ECG R-wave moment as the anchor point, the mean or peak value of the motion features within the corresponding time period is output synchronously at the end of each cardiac cycle, ensuring a one-to-one temporal correspondence between the two types of features. This synchronization mechanism is a prerequisite for the subsequent construction of a multidimensional risk vector, avoiding misjudgment of risk due to temporal misalignment. Finally, both the extracted myocardial ischemia-related features and motion-related features are output in standardized numerical form. Before being fed into the risk assessment model, Z-score standardization or Min-Max normalization can be performed to eliminate the influence of dimensional differences on weight allocation. It should be noted that the temporal granularity of feature output, preprocessing methods, and feature dimension expansion can be flexibly adjusted according to actual hardware configuration and clinical validation results; this invention does not impose limitations on these aspects. Through the above implementation method, a reliable mapping from raw physiological-motor signals to a high-risk discriminative feature space is achieved, comprehensively supporting individualized and real-time risk perception during post-PCI rehabilitation exercises.

[0058] S5: Obtain the individualized baseline parameters of the target object and encode the individualized baseline parameters into a target object profile vector; concatenate the myocardial ischemia-related features, motion-related features and the target object profile vector to construct a multidimensional risk vector.

[0059] In cardiac rehabilitation exercises after PCI, different patients exhibit significant individual differences in their tolerance to the same exercise load and myocardial ischemia trigger threshold, influenced by factors such as left ventricular ejection fraction (LVEF), stent implantation site, number of diseased vessels, postoperative recovery days, and medication use. Relying solely on real-time ECG or exercise indicators for risk assessment can easily lead to misjudgments by neglecting the patient's static or slowly changing pathological baseline. Therefore, this embodiment employs a multi-dimensional feature splicing strategy to deeply integrate real-time dynamic features with individualized static features, providing a highly discriminative input space for subsequent fuzzy logic risk assessment.

[0060] Specifically, when obtaining individualized baseline parameters for the target individual, key postoperative characteristic parameters can be extracted from the patient's medical records, PCI surgery records, echocardiograms, medication records, rehabilitation assessment records, and exercise prescriptions issued by the physician. These parameters are entered when the patient is first registered, read before each rehabilitation training session, and dynamically updated after follow-up examinations or updates to medical orders. Preferably, individualized baseline parameters include, but are not limited to, left ventricular ejection fraction (LVEF), stent location, number of diseased vessels, and postoperative days. Information on beta-blocker use, history of arrhythmias or angina, and prescribed target heart rate range. It should be noted that the sources and types of the above parameters can be expanded or modified based on the availability of actual clinical data; this invention does not limit these modifications.

[0061] Furthermore, in this embodiment, the extracted individualized baseline parameters are numerically encoded to construct the target object profile vector pp. The specific encoding method is as follows: LVEF is coded according to echocardiography or discharge summary. (For example, LVEF ≥ 50% is coded as 1.0, 40%-49% as 1.2, and < 40% as 1.5); the stent site is coded according to the PCI surgical record. (For example, the left main coronary artery is coded as 1.5, the proximal segment of the left anterior descending artery as 1.3, the proximal segment of the circumflex artery or right coronary artery as 1.2, and the distal or branch artery as 1.0); the number of diseased vessels is coded according to the coronary angiography report. (For example, 1.0 for single-vessel, 1.2 for double-vessel, and 1.5 for triple-vessel or left main coronary artery lesions); postoperative days are normalized based on the surgery date and training date. (1, postoperative days / 90) is used to characterize the additional safety weight for early rehabilitation; Receptor blocker use and previous arrhythmias were coded as binary variables. and (1 for use or presence, 0 for no use or absence); the lower limit of the target heart rate zone is directly extracted from the doctor's order. and upper limit Ultimately, the target object profile vector is defined as follows: Through this coding mechanism, static or slowly changing postoperative risk factors are transformed into a numerical space that can be calculated by the model, enabling the risk threshold to adapt to individual differences among patients after PCI.

[0062] After constructing the target object profile vector, this embodiment will use the myocardial ischemia-related features extracted in the aforementioned steps. Motion-related features The vector is concatenated with the target object profile vector pp to construct a multi-dimensional risk vector (i.e., the risk assessment input vector). In this embodiment, the multidimensional risk vector is defined as follows: This vector not only encompasses real-time ECG morphological changes (such as ST segment deviation, ST slope, and T wave alternation), heart rate reserve, and exercise load intensity, but also deeply integrates patient-specific anatomical and pathological baseline information, achieving a three-dimensional coupling of "real-time physiological state - dynamic exercise load - static medical history characteristics." It should be noted that the dimensions of the multidimensional risk vector can be expanded based on actual hardware configuration and clinical validation results (e.g., by adding respiratory rate or blood oxygen saturation characteristics), and this invention does not impose any limitations on this.

[0063] S6: Input the multidimensional risk vector into the preset fuzzy inference model to obtain the current risk level of the target object, and generate the load control instruction corresponding to the current risk level.

[0064] In cardiac rehabilitation exercises following PCI, there is a complex nonlinear coupling relationship between the patient's physiological indicators (such as ST segment deviation and heart rate reserve) and postoperative pathological parameters (such as LVEF and stent location), and the individual's tolerance boundary to exercise load has significant fuzziness and uncertainty. Traditional assessment methods relying on a single fixed threshold (such as the absolute value of ST segment depression) are insufficient to fully reflect the true risk and are prone to false alarms or false negatives. Therefore, this embodiment introduces a fuzzy logic reasoning mechanism, which transforms precise numerical inputs into semantic fuzzy sets, performs reasoning through an expert rule base, and outputs continuous load control instructions, thereby achieving individualized closed-loop rehabilitation training while ensuring medical safety.

[0065] Specifically, when inputting the multidimensional risk vector into the preset fuzzy inference model, the multidimensional risk vector is first normalized to eliminate the dimensional differences between different physiological characteristics and medical history parameters. Then, a Gaussian membership function is used to map each dimension of the normalized multidimensional risk vector to its corresponding fuzzy set, resulting in a fuzzy input vector. Taking ST segment depression as an example, its corresponding fuzzy set can be divided into Normal, Mild, and Severe. The Gaussian membership function calculation formula is as follows: in, The normalized feature input values, The center value of the fuzzy set The standard deviation parameter is used to control the width of the fuzzy set. It should be noted that, for features of different dimensions such as LVEF, HRR, RPE compensation index, and T-wave anomalies, corresponding fuzzy set and membership function parameters can be set separately, but this invention does not limit this.

[0066] Furthermore, rule inference is performed on the fuzzy input vector based on a preset fuzzy rule base. This embodiment preferably uses the Mamdani inference mechanism, with rule forms such as: "IF ST_depression is Severe AND LVEF isLow AND StentRegion is HighRisk THEN Risk is High".

[0067] For the The formula for calculating the trigger strength of a fuzzy rule is as follows: The minimum membership degree of all input features in the antecedent of the rule is taken as the overall activation degree of the rule. After determining the trigger strength of each fuzzy rule, the conclusion fuzzy sets output by all fuzzy rules are aggregated based on the trigger strength to obtain the aggregated risk fuzzy set. Subsequently, the aggregated risk fuzzy set is defuzzified, preferably using the centroid method to calculate the continuous risk score, the formula of which is: in, For variables in the domain of risk, This represents the aggregated risk membership function. Through the above reasoning and defuzzification process, the multidimensional fuzzy input is transformed into a continuous risk score between 0 and 100.

[0068] After obtaining continuous risk scores, they are mapped to preset numerical ranges to determine the current risk level of the target object. In this embodiment, the risk level is divided into four levels: 0 to 25 is mapped to safe, 25 to 50 to low risk, 50 to 75 to intermediate risk, and 75 to 100 to high risk. Preferably, to prevent fuzzy inference from underestimating risk due to the dilution of multiple features in extreme critical situations, this embodiment also introduces a safety rule coverage mechanism. Specifically, if a hard risk event such as ST segment depression ≥0.1mV lasting for at least 3 cardiac cycles, ventricular arrhythmia, rapid drop in SpO2, or patient-reported chest pain is detected, the safety rule will directly cover the fuzzy inference result, forcibly raising the current risk level to intermediate or high risk. It should be noted that the triggering conditions for safety rule coverage and the target level for forced mapping can be configured according to clinical guidelines and doctor's prescriptions, and are not limited here.

[0069] Based on the determined current risk level, this embodiment generates a corresponding load control instruction. The action space for load control is defined as follows: These instructions represent increasing exercise intensity, maintaining, decreasing exercise intensity, smooth stopping, and emergency braking, respectively. In practice, when the risk level is safe, an instruction is generated to maintain the current exercise intensity or slightly increase it according to the prescription; when the risk level is low, an instruction is generated to decrease the exercise intensity by 5%-10% or to remain under observation; when the risk level is medium, an instruction is generated to automatically reduce the exercise intensity by 10%-30% and limit further load; when the risk level is high, an instruction is generated to smoothly stop or apply emergency braking, simultaneously triggering medical alarms. By defining a clearly defined action space, seamless integration from risk assessment to the physical control of rehabilitation equipment is achieved.

[0070] Specifically, when the generated load control command is a smooth stop, to avoid the rehabilitation equipment (such as a power bike or treadmill) from momentarily locking up and causing the patient to fall or experience drastic hemodynamic fluctuations, this embodiment uses an S-curve to smoothly decay the exercise intensity of the equipment. For exercise intensity targets such as speed, power, or resistance... Its smooth stopping control formula is defined as: in, The initial motion intensity to trigger a smooth stop. To ensure the minimum strength required for equipment safety, and These parameters are used to control the slope and center time of intensity reduction, respectively. It should be noted that the parameters of the S-curve described above can be adaptively calibrated based on the mechanical response characteristics of different rehabilitation devices and the patient's weight. Furthermore, if a communication interruption exceeding a preset threshold (e.g., 500ms) is detected during the execution of control commands, the device can automatically enter a local safety protection mode, reducing the exercise intensity to a safe lower limit; this invention does not impose limitations on this. Through the above implementation methods, a closed-loop safety control system is comprehensively ensured for patients after PCI during cardiac rehabilitation exercise.

[0071] Another embodiment of the present invention provides a motion load adaptive control system; for details, please refer to [link to relevant documentation]. Figure 2 , Figure 2 The diagram shown illustrates a structural block diagram of a motion load adaptive control system according to one embodiment of the present invention, comprising: Acquisition module 11 is used to acquire electrocardiogram signal data and motion status data of the target object during rehabilitation exercise; Preprocessing module 12 is used to sequentially perform wavelet multi-layer decomposition and wavelet reconstruction on the electrocardiogram signal data to obtain the baseline correction signal; The correction module 13 is used to take motion state data as reference noise input and perform non-stationary artifact cancellation on the baseline correction signal based on the variable step size adaptive filtering algorithm to obtain a high-fidelity true electrical signal. Extraction module 14 is used to extract features from high-fidelity electrical signals and motion state data respectively to obtain myocardial ischemia-related features and motion-related features; The splicing module 15 is used to obtain the individualized baseline parameters of the target object and encode the individualized baseline parameters into a target object profile vector; the myocardial ischemia-related features, motion-related features and the target object profile vector are spliced ​​together to construct a multidimensional risk vector; The control module 16 is used to input the multi-dimensional risk vector into the preset fuzzy inference model to obtain the current risk level of the target object and generate the load control instruction corresponding to the current risk level.

[0072] Preferably, in one embodiment of the present invention, the correction module is specifically used for: Wavelet multi-level decomposition was performed on the electrocardiogram signal data to obtain high-frequency detail coefficients and low-frequency approximation coefficients at each scale; The high-frequency detail coefficients and low-frequency approximation coefficients are processed separately to obtain denoised high-frequency detail coefficients and drift-free low-frequency approximation coefficients. Wavelet reconstruction is then performed on the denoised high-frequency detail coefficients and drift-free low-frequency approximation coefficients to obtain the baseline correction signal.

[0073] Preferably, in one embodiment of the present invention, the correction module is specifically used for: Gravity component removal is performed on the triaxial acceleration signal in the motion state data to obtain the triaxial dynamic acceleration signal, and the first instantaneous motion intensity at the current moment is determined based on the triaxial dynamic acceleration signal; A reference input vector is constructed based on the motion intensity at the first instant and the motion intensity at the second instant in the historical time. The baseline correction signal is used as the desired input signal, and the reference input vector is weighted according to the current weight vector of the adaptive filter to obtain the artifact estimation signal. The difference between the expected input signal and the artifact estimation signal is calculated to determine the filtering error signal. The filtering error signal represents the high-fidelity electrical signal after removing non-stationary motion artifacts.

[0074] Preferably, in one embodiment of the present invention, the correction module is further configured to: The step size factor of the adaptive filter is dynamically updated based on the amplitude of the motion intensity and the filtering error signal at the first instant; wherein the updated value of the step size factor is positively correlated with the amplitude of the motion intensity and the filtering error signal at the first instant. Based on the step size factor, the filtering error signal, and the reference input vector, the current weight vector of the adaptive filter is iteratively updated to obtain the target weight vector for the next time step. The target weight vector is used to generate the artifact estimation signal in the next time step.

[0075] Preferably, in one embodiment of the present invention, the control module is specifically used for: The multidimensional risk vector is normalized, and the features of each dimension in the normalized multidimensional risk vector are mapped to the corresponding fuzzy set through the Gaussian membership function to obtain the fuzzy input vector. Based on a pre-defined fuzzy rule base, rule reasoning is performed on the fuzzified input vector to determine the trigger strength of each fuzzy rule. Based on the trigger strength, the fuzzy set of conclusions output by all fuzzy rules is aggregated to obtain an aggregated risk fuzzy set. The aggregated risk fuzzy set is defuzzified to obtain continuous risk scores; The continuous risk score is mapped to a preset numerical range to determine the current risk level of the target object.

[0076] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for adaptive regulation of exercise load, characterized in that, include: Acquire electrocardiogram (ECG) signal data and motion status data of the target subject during rehabilitation exercises; The electrocardiogram signal data is subjected to wavelet multi-layer decomposition and wavelet reconstruction processing in sequence to obtain the baseline correction signal; Using the motion state data as a reference noise input, the baseline correction signal is non-stationary artifact cancellation is performed based on a variable step size adaptive filtering algorithm to obtain a high-fidelity electrical signal. Feature extraction was performed on the high-fidelity electrical signal and the motion state data respectively to obtain myocardial ischemia-related features and motion-related features; Obtain individualized baseline parameters of the target object and encode the individualized baseline parameters into a target object profile vector; concatenate the myocardial ischemia-related features, the motion-related features, and the target object profile vector to construct a multidimensional risk vector; The multidimensional risk vector is input into a preset fuzzy inference model to obtain the current risk level of the target object, and a load control instruction corresponding to the current risk level is generated.

2. The adaptive control method for exercise load as described in claim 1, characterized in that, The process of sequentially performing wavelet multi-level decomposition and wavelet reconstruction on the electrocardiogram signal data to obtain a baseline correction signal includes: Wavelet multi-level decomposition was performed on the electrocardiogram signal data to obtain high-frequency detail coefficients and low-frequency approximation coefficients at each scale; The high-frequency detail coefficients and the low-frequency approximation coefficients are processed respectively to obtain denoised high-frequency detail coefficients and drift-free low-frequency approximation coefficients. Wavelet reconstruction is then performed on the denoised high-frequency detail coefficients and the drift-free low-frequency approximation coefficients to obtain the baseline correction signal.

3. The adaptive adjustment method for exercise load as described in claim 1, characterized in that, The process of using the motion state data as reference noise input and performing non-stationary artifact cancellation on the baseline correction signal based on a variable step-size adaptive filtering algorithm to obtain a high-fidelity electrical signal includes: The gravity component of the triaxial acceleration signal in the motion state data is removed to obtain the triaxial dynamic acceleration signal, and the first instantaneous motion intensity at the current moment is determined based on the triaxial dynamic acceleration signal. A reference input vector is constructed based on the first instantaneous motion intensity and the second instantaneous motion intensity at a historical moment; The baseline correction signal is used as the desired input signal, and the reference input vector is weighted according to the current weight vector of the adaptive filter to obtain the artifact estimation signal. The difference between the desired input signal and the artifact estimation signal is calculated to determine the filtering error signal, which represents the high-fidelity electrical signal after removing non-stationary motion artifacts.

4. The adaptive regulation method for exercise load as described in claim 3, characterized in that, After determining the filtering error signal based on the difference between the expected input signal and the artifact estimation signal, the method further includes: Based on the first instantaneous motion intensity and the amplitude of the filtering error signal, the step size factor of the adaptive filter is dynamically updated; wherein, the updated value of the step size factor is positively correlated with the first instantaneous motion intensity and the amplitude of the filtering error signal. Based on the step size factor, the filtering error signal, and the reference input vector, the current weight vector of the adaptive filter is iteratively updated to obtain the target weight vector at the next time step, wherein the target weight vector is used to generate the artifact estimation signal at the next time step.

5. The adaptive regulation method for exercise load as described in claim 1, characterized in that, The step of inputting the multidimensional risk vector into a preset fuzzy inference model to obtain the current risk level of the target object includes: The multidimensional risk vector is normalized, and the features of each dimension in the normalized multidimensional risk vector are mapped to the corresponding fuzzy set through the Gaussian membership function to obtain the fuzzy input vector. Based on a preset fuzzy rule base, rule reasoning is performed on the fuzzified input vector to determine the trigger strength of each fuzzy rule, and the conclusion fuzzy set output by all fuzzy rules is aggregated based on the trigger strength to obtain an aggregated risk fuzzy set. The aggregated risk fuzzy set is defuzzified to obtain continuous risk scores; The continuous risk score is mapped to a preset numerical range to determine the current risk level of the target object.

6. A motion load adaptive control system, characterized in that, include: The acquisition module is used to acquire electrocardiogram signal data and motion status data of the target object during rehabilitation exercise; The preprocessing module is used to sequentially perform wavelet multi-layer decomposition and wavelet reconstruction processing on the electrocardiogram signal data to obtain the baseline correction signal; The correction module is used to take the motion state data as a reference noise input and perform non-stationary artifact cancellation on the baseline correction signal based on a variable step size adaptive filtering algorithm to obtain a high-fidelity electrical signal. The extraction module is used to extract features from the high-fidelity electrical signal and the motion state data respectively to obtain myocardial ischemia-related features and motion-related features; The splicing module is used to obtain the individualized baseline parameters of the target object and encode the individualized baseline parameters into a target object profile vector; the myocardial ischemia-related features, the motion-related features and the target object profile vector are spliced ​​together to construct a multidimensional risk vector; The control module is used to input the multidimensional risk vector into a preset fuzzy inference model to obtain the current risk level of the target object and generate a load control instruction corresponding to the current risk level.

7. The motion load adaptive control system as described in claim 6, characterized in that, The correction module is specifically used for: Wavelet multi-level decomposition was performed on the electrocardiogram signal data to obtain high-frequency detail coefficients and low-frequency approximation coefficients at each scale; The high-frequency detail coefficients and the low-frequency approximation coefficients are processed respectively to obtain denoised high-frequency detail coefficients and drift-free low-frequency approximation coefficients. Wavelet reconstruction is then performed on the denoised high-frequency detail coefficients and the drift-free low-frequency approximation coefficients to obtain the baseline correction signal.

8. The motion load adaptive control system as described in claim 6, characterized in that, The correction module is specifically used for: The gravity component of the triaxial acceleration signal in the motion state data is removed to obtain the triaxial dynamic acceleration signal, and the first instantaneous motion intensity at the current moment is determined based on the triaxial dynamic acceleration signal. A reference input vector is constructed based on the first instantaneous motion intensity and the second instantaneous motion intensity at a historical moment; The baseline correction signal is used as the desired input signal, and the reference input vector is weighted according to the current weight vector of the adaptive filter to obtain the artifact estimation signal. The difference between the desired input signal and the artifact estimation signal is calculated to determine the filtering error signal, which represents the high-fidelity electrical signal after removing non-stationary motion artifacts.

9. The motion load adaptive control system as described in claim 8, characterized in that, The correction module is also used for: Based on the first instantaneous motion intensity and the amplitude of the filtering error signal, the step size factor of the adaptive filter is dynamically updated; wherein, the updated value of the step size factor is positively correlated with the first instantaneous motion intensity and the amplitude of the filtering error signal. Based on the step size factor, the filtering error signal, and the reference input vector, the current weight vector of the adaptive filter is iteratively updated to obtain the target weight vector at the next time step, wherein the target weight vector is used to generate the artifact estimation signal at the next time step.

10. The motion load adaptive control system as described in claim 6, characterized in that, The control module is specifically used for: The multidimensional risk vector is normalized, and the features of each dimension in the normalized multidimensional risk vector are mapped to the corresponding fuzzy set through the Gaussian membership function to obtain the fuzzy input vector. Based on a preset fuzzy rule base, rule reasoning is performed on the fuzzified input vector to determine the trigger strength of each fuzzy rule, and the conclusion fuzzy set output by all fuzzy rules is aggregated based on the trigger strength to obtain an aggregated risk fuzzy set. The aggregated risk fuzzy set is defuzzified to obtain continuous risk scores; The continuous risk score is mapped to a preset numerical range to determine the current risk level of the target object.

Images