Nationwide fitness participation data statistics and personalized service recommendation system

Abstract

The invention relates to the technical field of intelligent fitness equipment control, in particular to a nationwide fitness participation data statistics and personalized service recommendation system, which comprises a processor and a memory, and the processor is configured to execute the following modules: a multi-modal sensing interface module, which is used for synchronously collecting target joint angular acceleration and active muscle group surface electromyographic signals, and sending the signals to the memory; extracting an electromyographic linear envelope as a neural driving representation and completing data alignment preprocessing; and the random dynamics mapping module is used for constructing an Ittan type human joint stochastic differential equation in which random time-varying time-lag parameters and Wiener process noise items are introduced so as to represent the uncertainty of nerve conduction. According to the method, the safety boundary is dynamically contracted based on the nerve time-lag variance, and then prior blocking is implemented when the system is uncontrollable, so that the problem that the motor injury is caused by untimely protection intervention due to phase lag caused by neglecting nerve conduction randomness due to the fact that most traditional protection adopts fixed threshold passive triggering is solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent fitness equipment control technology, and in particular to a system for statistical analysis of public fitness participation data and personalized service recommendation. Background Technology

[0002] In both mass fitness and professional strength training, high-load resistance training such as squats and bench presses are core methods for improving muscle strength. To achieve optimal training results, trainees often need to train to exhaustion, a stage that is also prone to sports injuries. Existing intelligent strength training equipment typically uses deterministic kinetic models for control, and its safety mechanisms mainly rely on fixed physical limits or passive triggering logic based on macroscopic kinematic parameters.

[0003] However, traditional protection methods mostly use fixed thresholds for passive triggering. Because they ignore the phase lag caused by the randomness of nerve conduction, the protection intervention is not timely, which can lead to sports injuries. Summary of the Invention

[0004] To address the above shortcomings, this invention provides a data statistics and personalized service recommendation system for national fitness participation. It aims to improve the problem that traditional protection methods mostly rely on fixed threshold passive triggering, which ignores the phase lag caused by the randomness of nerve conduction, resulting in untimely intervention and sports injuries.

[0005] This invention provides the following technical solution: a system for statistical analysis and personalized service recommendation of national fitness participation data, comprising a processor and a memory, wherein the processor is configured to execute the following modules: The multimodal sensing interface module is used to synchronously acquire the target joint angular acceleration and the electromyographic signals on the surface of the active muscle group, extract the electrical envelope of the myocardial wires as a representation of neural drive, and complete the data alignment preprocessing.

[0006] The stochastic dynamics mapping module is used to construct Ito-type stochastic differential equations for human joints that incorporate stochastic time-varying delay parameters and Wiener process noise terms, in order to characterize the uncertainty of nerve conduction.

[0007] The time delay distribution identification module is used to calculate the cross-correlation function between neural drive representation and angular acceleration, and extract lag time samples to update the mean of time delay parameters and the variance representing uncertainty in real time. The robust constraint reconstruction module is used to determine the spatial coverage of the minimum robust positive invariant set based on variance, and to perform set subtraction operation to subtract the spatial coverage from the preset set of physical safety boundaries containing anatomical limits to generate a shrunken nominal safety constraint set. The predictive control solution module is used to take the nominal safety constraint set as hard constraints, construct and solve the quadratic optimal control problem in the finite time domain for the disturbance-free nominal model in the stochastic differential equation, so as to generate the optimal control input sequence containing future continuous time step control variables, and determine the first time step control component in the optimal control input sequence as the nominal control law; The proactive risk defense module is used to monitor the feasibility of the solution. When there is no solution, it determines that the uncertainty of neural conduction exceeds the limit and triggers mechanical damping locking or reverse braking command.

[0008] By adopting the above technical solution, the safety boundary is dynamically contracted based on the neural time delay variance, and then a priori blocking is implemented when the system is uncontrollable. This improves the problem that traditional protection mostly uses fixed threshold passive triggering, which ignores the phase lag caused by the randomness of neural conduction, resulting in untimely protection intervention and sports injuries.

[0009] Preferably, the multimodal sensing interface module includes: The surface electromyography sensor is controlled to acquire raw electromyography signals. Bandpass filtering and full-wave rectification are performed on the raw electromyography signals to remove power frequency interference. The linear envelope is extracted by a low-pass filter with a set cutoff frequency to construct a time-series signal reflecting the driving intensity of the central nervous system. The control angular acceleration sensor collects real-time motion data of the target joint, performs second-order differential processing on the real-time motion data to obtain an angular acceleration sequence, and uses a Kalman filter algorithm to smooth and denoise the angular acceleration sequence; Based on a unified high-precision clock source, the linear envelope and the angular acceleration sequence are resampled, and the two are aligned to the same time axis through an interpolation algorithm.

[0010] Preferably, the stochastic dynamics mapping module includes: Construct joint rigid body dynamics equations that include rotational inertia matrix, Coriolis force term, centrifugal force term and gravity term, and use them as the drift term of the Ito-type stochastic differential equation; An input delay variable is introduced into the control input channel of the drift term, and the input delay variable is modeled as a stochastic process modulated by neural fatigue. A diffusion term matrix related to the system state amplitude and control input amplitude is constructed. The standard Wiener process is weighted using the diffusion term matrix to simulate the signal-dependent multiplicative noise generated during neural signal transmission, thus completing the construction of the Iton-type stochastic differential equation.

[0011] Preferably, the time delay distribution identification module includes: Construct an observation window that slides along the time axis and extract the neural drive representation fragment and angular acceleration sequence fragment within the current window; Calculate the cross-correlation coefficient between the neural drive representation segment and the angular acceleration sequence segment at different lag times, and search for the lag time corresponding to the maximum cross-correlation coefficient as the instantaneous time delay observation value at the current moment; A recursive least squares algorithm with a forgetting factor is used to process the instantaneous time-delay observations at historical moments, and the probability density function parameters of the random time-varying time-delay parameters are updated in real time.

[0012] Preferably, the time delay distribution identification module further includes: The expected value and variance of the time delay are extracted from the parameters of the probability density function. The time-delay variance value is transmitted to the robust constraint reconstruction module as a dynamic indicator for quantifying the uncertainty of neural transmission. The expected time delay value is transmitted to the predictive control solution module for calibrating the input delay in the nominal dynamics model.

[0013] Preferably, the robust constraint reconstruction module includes: A linear state feedback controller is pre-designed to stabilize the error dynamics system of the stochastic differential equation; Based on the stability characteristics and time delay variance of the error dynamics system, a minimum robust positive invariant set that can contain all possible state error trajectories is calculated. A scaling rule is established such that the boundary range of the minimum robust positive invariant set monotonically expands as the time delay variance increases, thereby determining a safe buffer space to accommodate the uncertainty of neural conduction.

[0014] Preferably, the robust constraint reconstruction module further includes: Obtain the set of physical safety boundaries described by a polyhedron, which is enclosed by multiple hyperplanes that define anatomical and mechanical limits; Perform the Pontryagin set difference operation to translate each hyperplane of the physical security boundary set toward the center of the set. The translation distance is determined by the spatial coverage of the minimum robust positive invariant set in the corresponding dimension. The contracted polyhedral space enclosed by the translated hyperplane is defined as the nominal safety constraint set.

[0015] Preferably, the predictive control solution module includes: Construct a quadratic cost function, which includes a penalty term for the system state deviating from the reference trajectory and a penalty term for the energy consumption of the control input; The nominal safety constraint set is set as the hard constraint condition for the constraint state variables and control variables in the optimization problem; Within each control cycle, a quadratic programming algorithm is used to solve for the optimal control input sequence that minimizes the quadratic cost function within the finite prediction time domain, and the first time-step control component in the optimal control input sequence is determined as the nominal control law.

[0016] Preferably, the proactive risk defense module includes: Real-time reading of the return status code of the numerical optimization solver in the predictive control solver module; When the return status code indicates that the optimal solution has been found, a flexible auxiliary instruction containing fine-tuning information on the action rate is generated. When the returned status code indicates that the problem is unsolvable or infeasible, it is determined that the current neural conduction uncertainty has led to the closure of the safe and feasible domain, and a rigid blocking instruction with the highest interrupt priority is immediately generated.

[0017] Preferably, the proactive risk defense module further includes: Obtain the current real-time angular velocity data of the target joint; Based on the principle of viscous damping, the braking torque value that is opposite in direction and proportional in magnitude to the real-time angular velocity is calculated to construct a smooth braking curve. The braking torque value is encapsulated into a digital control signal and output as the rigid stop command to the underlying drive interface to trigger the braking action of the external actuator.

[0018] The present invention has the following beneficial effects: 1. In this invention, by dynamically shrinking the safety boundary based on the neural time delay variance, and then implementing prior blocking when the system is uncontrollable, the problem of traditional protection, which mostly adopts fixed threshold passive triggering, is caused by phase lag due to ignoring the randomness of neural conduction, resulting in untimely protection intervention and sports injuries is improved.

[0019] 2. In this invention, by constructing an Itō-type stochastic differential equation for human joints that incorporates stochastic time-varying delay parameters and Wiener process noise terms, the random fluctuation characteristics of neural signal transmission are realistically reproduced. This improves upon the problem that traditional motion modeling, which mostly uses deterministic rigid body dynamics, suffers from distortion in motion trajectory prediction due to neglecting signal-dependent noise during human fatigue.

[0020] 3. In this invention, by monitoring the feasibility of the predictive control solution module under hard constraints, and then determining that the neural uncertainty exceeds the limit and triggering blocking when there is no solution, the traditional risk defense mostly uses speed threshold judgment, which can only identify macroscopic motion anomalies, thus making it impossible to perform a priori blocking before the system becomes unstable.

[0021] 4. In this invention, by synchronously acquiring the electrical envelope and angular acceleration sequences of myoelectric lines and calculating the cross-correlation function, the lag time samples between neural drive and limb response can be identified in real time. This improves the problem that traditional physiological monitoring mostly uses single sensor data, which lacks multimodal temporal alignment analysis, making it difficult to accurately quantify the degree of neural conduction delay. Attached Figure Description

[0022] Figure 1 This is an architecture diagram of the national fitness participation data statistics and personalized service recommendation system proposed in this invention; Figure 2 This is a flowchart of the multimodal perception interface data processing of the national fitness participation data statistics and personalized service recommendation system proposed in this invention. Figure 3 The robust constraint reconstruction and predictive control logic diagram of the national fitness participation data statistics and personalized service recommendation system proposed in this invention is shown below. Figure 4 This is the execution logic diagram of the proactive risk defense module of the national fitness participation data statistics and personalized service recommendation system proposed in this invention. Detailed Implementation

[0023] The technical solutions in 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1: In the first embodiment of the present invention, the present invention provides a system for statistical analysis of public fitness participation data and personalized service recommendation, such as... Figures 1-4 As shown, it includes a processor and memory, and the processor is configured to execute the following modules: The multimodal sensing interface module is used to synchronously acquire the target joint angular acceleration and the electromyographic signals on the surface of the active muscle group, extract the electrical envelope of the myocardial wires as a representation of neural drive, and complete the data alignment preprocessing. Furthermore, the multimodal sensing interface module includes: The surface electromyography sensor is controlled to acquire raw electromyography signals. Bandpass filtering and full-wave rectification are performed on the raw electromyography signals to remove power frequency interference. The linear envelope is extracted by a low-pass filter with a set cutoff frequency to construct a time-series signal reflecting the driving intensity of the central nervous system. The control angular acceleration sensor collects real-time motion data of the target joint, performs second-order differential processing on the real-time motion data to obtain the angular acceleration sequence, and uses the Kalman filter algorithm to smooth and denoise the angular acceleration sequence; Based on a unified high-precision clock source, the linear envelope and angular acceleration sequences are resampled, and the two are aligned to the same time axis through an interpolation algorithm.

[0025] Specifically, in the surface electromyography (EMG) signal processing flow, the processor controls the surface EMG sensor to be attached to the user's active muscle group location to set the sampling frequency. The raw electromyography (EMG) signal sequence is acquired. After data input, the processor first performs bandpass filtering to remove power frequency interference and motion artifacts. Then, full-wave rectification is performed to convert the bipolar oscillation signal into a unipolar signal. Finally, the linear envelope of the signal is extracted using a digital low-pass filter. The mathematical expression for full-wave rectification in this step is as follows: ; in Indicates the rectified first... One electromyography signal sampling point, This represents the input signal after bandpass filtering.

[0026] The low-pass filter envelope extraction uses a first-order recursive digital filter algorithm, and the calculation formula is as follows: ; in This is the linear envelope value output at the current moment, i.e., the neural drive representation; This is the output value from the previous time step; This is a smoothing coefficient, which is determined by the cutoff frequency. and sampling frequency Decision, fulfilling the relationship Through the above processing, the module outputs a time-series signal reflecting the driving strength of the central nervous system, which serves as the input basis for subsequent neural conduction delay analysis.

[0027] The angular acceleration sequence acquisition and smoothing process involves the processor controlling an angular acceleration sensor or inertial measurement unit to collect real-time motion data of the target joint. When the acquired data is in the form of joint angles... At that time, the processor performs discretization and second-order differential processing to obtain the original sequence of angular accelerations. The formula for calculating the second-order differential is as follows: ; in For the first The estimated angular acceleration at each moment; , , These are the joint angle measurements at the current time, the previous time, and the time before that, respectively. This represents the sampling time interval.

[0028] To eliminate high-frequency noise introduced by differentiation operations, the processor uses the Kalman filter algorithm. The sequence undergoes state estimation and smoothing. State equations are established. and observation equations During the update phase, Kalman gain The calculation formula is as follows: ; in The Kalman gain matrix determines the weighting ratio between the measured and predicted values. The prior error covariance matrix; The observation matrix; To measure the noise covariance matrix, the processor uses this gain to correct the predicted values ​​and outputs a smoothed and denoised angular acceleration sequence.

[0029] In the data alignment and resampling process, due to the independent discrepancies in the sampling clocks of the electromyography (EMG) sensor and the angular accelerometer, the processor uses a unified, high-precision clock source as the reference time axis. For two sets of data sequences with inconsistent sampling rates, the processor uses the higher frequency as the reference and resamples the lower frequency signal.

[0030] For sampling points that are not multiples of an integer, the processor uses a linear interpolation algorithm to calculate the corresponding values ​​at that time. The interpolation formula is as follows: ; in To align to a unified timeline The target signal value at any given time; and For the original data sequence and Two adjacent timestamps; and Each is a timestamp and The corresponding original signal amplitude.

[0031] After the above processing, the module outputs a myoelectric linear envelope sequence and angular acceleration sequence with strictly aligned timestamps, eliminating timing errors caused by hardware asynchrony, and providing a time-domain accurate multimodal dataset for subsequent calculation of cross-correlation functions and identification of neural conduction delays.

[0032] The stochastic dynamics mapping module is used to construct Ito-type stochastic differential equations for human joints that incorporate stochastic time-varying delay parameters and Wiener process noise terms, in order to characterize the uncertainty of nerve conduction. Furthermore, the stochastic dynamics mapping module includes: Construct joint rigid body dynamics equations that include rotational inertia matrix, Coriolis force term, centrifugal force term and gravity term, and use them as drift terms of Iton-type stochastic differential equations; An input delay variable is introduced into the control input channel of the drift term, and the input delay variable is modeled as a stochastic process modulated by neural fatigue. A diffusion term matrix related to the system state amplitude and control input amplitude is constructed. The standard Wiener process is weighted using the diffusion term matrix to simulate the signal-dependent multiplicative noise generated during neural signal transmission, thus completing the construction of the Iton-type stochastic differential equation.

[0033] Specifically, the stochastic dynamics mapping module establishes a mathematical model describing the motion patterns of human joints under neural fatigue based on the alignment data output by the multimodal sensing interface module. The processing flow of this module includes three steps: constructing deterministic drift terms, embedding stochastic time-delay parameters, and constructing stochastic diffusion terms.

[0034] The construction of the drift term and the state space transformation are achieved by first establishing the joint rigid body dynamics equation based on the Lagrange principle to describe the physical evolution of the system under an ideal noise-free state. This part constitutes the drift term of the stochastic differential equation.

[0035] For single-degree-of-freedom or multi-degree-of-freedom joints, the dynamic equations are expressed as follows: ; in Represents the joint angle vector; Represents the joint angular velocity vector; Represents the joint angular acceleration vector; It is a symmetric positive definite moment of inertia matrix, representing the inertial distribution of the limbs; The matrix represents the Coriolis force and centrifugal force, characterizing the coupling effect between multiple joints; This is the gravity term vector, representing the torque generated by the weight of the limbs themselves; This is the input of the control torque generated by the active muscle groups.

[0036] To construct Itō-type equations, the processor transforms the above second-order differential equations into a first-order state-space form. The system state vector is defined. Then the drift term function Expressed as: ; in That is, control input The drift item It describes the deterministic trend of the system state over time.

[0037] The embedding of random time-varying delay parameters addresses the delay in the arrival of neural commands at muscles to generate torque under conditions of neural fatigue, and this delay exhibits random fluctuations. The processor does not directly use the current-moment control input. Instead, time-delay variables are introduced. Correct the control input to .

[0038] Time-delay variables Modeled as subject to neural fatigue parameters Modulated stochastic processes, whose expected value and variance vary with The rate of decrease is not linearly increased. The corrected drift term dynamically reflects the lag effect of neural conduction delay on the evolution of the system state.

[0039] The diffusion term matrix is ​​constructed and the Itō equation is generated. Signal-dependent noise during neural signal transmission causes the actual motion trajectory to deviate from the theoretical drift path. The processor constructs the diffusion term matrix to simulate this multiplicative noise characteristic. The noise intensity is not constant but proportional to the current control input amplitude and state amplitude. The final constructed Itō-type stochastic differential equation for human joints is as follows: ; in This represents the increment of the system state vector within a small time step; For the drift term function with time-delay input constructed above; The time differential term; The increment of the standard multidimensional Wiener process satisfies the Gaussian distribution characteristics and characterizes an independent white noise source; This is the diffusion term matrix, used to weight the standard Wiener process. The specific format is set as follows: ; in The additive noise figure characterizes the background ambient noise. To control for the dependency noise figure, the physiological characteristic that the greater the muscle exertion, the more intense the trembling; The noise figure is state-dependent. and These are the absolute values ​​or norms of the control input and angular velocity, respectively.

[0040] Through the above steps, the module outputs a parameterized Iton-type stochastic differential equation. This equation serves as the basis for subsequent model predictive control, capable of predicting the evolution path of the system state probability distribution considering neural delays and signal noise under different input torques.

[0041] The time delay distribution identification module is used to calculate the cross-correlation function between neural drive representation and angular acceleration, and extract lag time samples to update the mean of time delay parameters and the variance representing uncertainty in real time. Furthermore, the time delay distribution identification module includes: Construct an observation window that slides along the time axis and extract the neural drive representation fragment and angular acceleration sequence fragment within the current window; Calculate the cross-correlation coefficient between the neural drive representation segment and the angular acceleration sequence segment at different lag times, and search for the lag time corresponding to the maximum cross-correlation coefficient as the instantaneous time delay observation value at the current moment; A recursive least squares algorithm with a forgetting factor is used to process the instantaneous time-delay observations of historical moments, and the probability density function parameters of the random time-varying time-delay parameters are updated in real time.

[0042] The time delay distribution identification module also includes: The expected value and variance of the time delay are extracted from the parameters of the probability density function. The time-delay variance is used as a dynamic indicator to quantify the uncertainty of neural transmission and transmitted to the robust constraint reconstruction module. The expected time delay value is transmitted to the predictive control solver module for calibration of the input delay in the nominal dynamics model.

[0043] Specifically, the time-delay distribution identification module receives the aligned data sequence output by the multimodal perception interface module, namely the neural drive representation signal and the joint angular acceleration signal. This module extracts instantaneous features through cross-correlation analysis and uses a recursive algorithm to solve for statistical features.

[0044] Cross-correlation analysis and instantaneous time-delay observation: the processor first constructs a length of... The sliding observation window at the current sampling time Extract the neural drive representation sequence within the window. With angular acceleration sequence The processor calculates the two sets of sequences at different lag times. Cross-correlation function under The discrete calculation formula for the cross-correlation function is as follows: ; in This indicates the first neural drive representation sequence within the window. One historical data point; Indicates the angular acceleration sequence after the applied hysteresis offset The corresponding data points after; The window length is defined, and its value covers a preset range of physiological delay fluctuations. The search variable takes an integer step size within the interval from the minimum physiological delay to the maximum physiological delay.

[0045] The processor traverses all data within the preset range. Value, search for cross-correlation function The hysteresis offset that reaches its maximum value is defined as the instantaneous time-delay observation at the current moment. This value represents the actual physical time difference between the issuance of a neural command and the muscle's mechanical response at the current moment.

[0046] The recursive update of the probability density function parameters, due to the time-varying characteristics of delay changes caused by neural fatigue, involves the processor employing a recursive least squares algorithm with a forgetting factor or a recursive weighted statistical algorithm to process the historical observation sequence and estimate the probability distribution characteristics of the time-delay parameters in real time. The processor establishes the time-delay mean. With time delay variance The recursive update equation is as follows: Mean update formula: ; Variance update formula: ; in and These are the time-delay mean estimate and variance estimate, respectively, updated at the current time. and This is the corresponding estimated value at the previous sampling time. The current instantaneous time-delay observation value calculated in step one; This is the forgetting factor, with a value between 0 and 1. This determines the rate at which the algorithm's weights decay over historical data. The smaller the value, the more sensitive the system is to new observations at the current moment, and the faster it can track delays caused by fatigue.

[0047] The output flow of statistical characteristic parameters, after the above recursive calculation, the module parses the optimal time-delay statistical parameters for the current moment and outputs them according to the following paths: The first path involves the processor extracting the updated mean time delay. This mean value is then transmitted to the predictive control solver module. It is used to correct the input delay parameter in the nominal dynamics model, ensuring that the predictive model reflects the average lag level of current neural conduction and eliminating system steady-state errors.

[0048] The second path involves the processor extracting the updated time-delay variance. This variance value, as a dynamic indicator of the uncertainty of neural transmission, directly determines the size of the spatial coverage of the subsequent robust positive invariant set. That is, the higher the uncertainty, the larger the variance, and the wider the safety buffer space reserved by the system.

[0049] The robust constraint reconstruction module is used to determine the spatial coverage of the minimum robust positive invariant set based on variance, and to perform set subtraction operation to subtract the spatial coverage from the preset set of physical safety boundaries containing anatomical limits, thereby generating the shrunken nominal safety constraint set. Furthermore, the robust constraint reconstruction module includes: A linear state feedback controller is pre-designed to stabilize the error dynamics system of the stochastic differential equation; Based on the stability characteristics and time delay variance of the error dynamics system, a minimum robust positive invariant set that can contain all possible state error trajectories is calculated. A scaling rule is established such that the boundary range of the minimum robust positive invariant set monotonically expands as the lag variance increases, thereby determining a safe buffer space to accommodate the uncertainty of neural conduction.

[0050] The robust constraint reconstruction module also includes: Obtain the set of physically safe boundaries described by a polyhedron, which is bounded by multiple hyperplanes that define anatomical and mechanical limits; Perform the Pontryagin set difference operation to translate each hyperplane of the physical security boundary set toward the center of the set. The translation distance is determined by the spatial coverage of the minimum robust positive invariant set in the corresponding dimension. The contracted polyhedral space enclosed by the translated hyperplane is defined as the nominal safety constraint set.

[0051] Specifically, the robust constraint reconstruction module receives time-delay variance data from the time-delay distribution identification module and system matrix parameters from the stochastic dynamics mapping module. This module transforms uncertain random disturbances into deterministic geometric constraints through error feedback control design and ensemble geometric operations.

[0052] Error dynamics system construction and stabilization: The processor first defines the nominal system state. With respect to the actual system state State error vector between Its defining formula is To suppress state divergence caused by uncertainties in neural conduction, the processor is pre-configured with a linear state feedback controller. ,in Here is the feedback gain matrix. Based on this controller, the processor constructs the closed-loop error dynamics equation as follows: ; in This is the system state matrix; The input matrix; To make the matrix The eigenvalues ​​of all are located in the stable gain matrix of the left half of the complex plane; The perturbation weighting matrix has a norm whose magnitude is related to the time delay variance of the input. Directly related; This is the normalized random disturbance term. This equation describes the evolution and boundedness of the system error over time under feedback control.

[0053] The spatial coverage of the minimum robust positive invariant set is calculated based on the convergence characteristics of the error dynamics system. The processor calculates a value that can enclose the error state. The minimum robust positive invariant set of all possible trajectories The spatial coverage of this set is determined by the statistical properties of the disturbance terms.

[0054] The processor establishes the boundary scaling rule for the minimum robust positive invariant set, and its geometric dimensions are calculated using the following formula: ; in Representative set The characteristic size or radius in the state space; It is the square root of the time-delay variance, i.e., the standard deviation, used to quantify the current level of uncertainty in neural conduction. A constant scaling factor related to the physical properties of the system; Let be the spectral radius of the closed-loop system matrix, which characterizes the convergence rate of the system.

[0055] According to this formula, when the input time delay variance When increasing, A monotonically increasing error indicates that the range of error fluctuations is expanding, and the system needs to reserve a larger safety buffer.

[0056] Pontryagin set difference operation and boundary shrinkage: The processor obtains a preset set of physical security boundaries. This set is a polyhedron defined by a system of linear inequalities describing the anatomical and mechanical limits of joints: ; in To define the constraint matrix for the normal vectors of the polyhedron, Representation matrix The OK; To define the threshold vector for the boundary position of the polyhedron, Representing vectors The One component; Let be the system state dimension.

[0057] The processor performs the Pontryagin set difference operation. Generate the shrunken nominal safety constraint set This operation is achieved by translating the physical boundary towards the geometric center, resulting in a new threshold vector. The The formula for calculating each component is as follows: ; in For the first contraction One constraint boundary threshold; The original physical boundary threshold; In the error set Internal state error In the The maximum projected component along the direction of the constraint normal vector.

[0058] This calculation process will determine the physical security boundary. Subtracted the maximum possible error boundary caused by neural uncertainty The processor output is determined by the new threshold vector. Defined set of nominal security constraints To the predictive control solver module. This process ensures that only the nominal state... lie in Internally, regardless of how the actual neural delay fluctuates, the actual state after adding errors... Always within the physical security boundary Inside.

[0059] The predictive control solution module is used to construct and solve the quadratic optimal control problem in the finite time domain for the perturbation-free nominal model in the stochastic differential equation, using the nominal safety constraint set as hard constraints, in order to generate the optimal control input sequence containing future continuous time-step control variables, and to determine the first time-step control component in the optimal control input sequence as the nominal control law; Furthermore, the predictive control solution module includes: Construct a quadratic cost function, which includes a penalty term for the system state deviating from the reference trajectory and a penalty term for the energy consumption of the control input; The nominal safety constraint set is set as the hard constraint condition for the constraint state variables and control variables in the optimization problem; Within each control cycle, a quadratic programming algorithm is used to solve for the optimal control input sequence that minimizes the quadratic cost function within the finite prediction time domain, and the first time-step control component in the optimal control input sequence is determined as the nominal control law.

[0060] Specifically, the predictive control solver module receives the nominal safety constraint set output by the robust constraint reconstruction module, as well as the nominal dynamic model parameters calibrated by the time delay distribution identification module. This module solves the finite-time optimal control problem online at each sampling time, calculating the optimal control sequence.

[0061] To construct the prediction model and cost function, the processor first extracts the nominal part of the stochastic dynamics mapping module that does not contain random noise and time delay perturbations, and then constructs a discrete-time nominal prediction model: ; in For the first The nominal state vector at time t; For the first The nominal control input vector at time step; and These are the state transition matrix and input matrix obtained after discretizing the continuous-time system matrix, respectively.

[0062] To achieve accurate tracking of the reference trajectory and suppress control energy consumption, the processor constructs a quadratic cost function. This function is defined in a space of length . In the prediction time domain, its mathematical expression is as follows: ; in The predicted time domain length is the number of steps to predict forward. Indicates at time For future moments The nominal state prediction value; Indicates at time Planning for the future nominally controlled input; For the preset ideal reference trajectory; Representing vectors Regarding the weight matrix The weighted Euclidean norm, i.e. ; This is the state error weighting matrix, used to penalize the deviation between the system state and the reference trajectory; The input weighting matrix is ​​used to penalize excessive control force to prevent actuator saturation; This is the terminal cost weight matrix, used to ensure the stability of the closed-loop system.

[0063] When hard constraints are applied, the processor will use the nominal safety constraint set generated by the robust constraint reconstruction module. and These are set as hard constraints in the optimization process. At each prediction step within the prediction time domain... The system state and control inputs satisfy the following inequality constraints: ; ; Among the constraint sets It is a contracted set that has already eliminated the error buffer caused by uncertainties in neural transmission. This step mathematically mandates that the nominal predicted trajectory must run within the contracted safety tube, thereby ensuring that the actual trajectory does not exceed the physical safety boundary.

[0064] The processor applies the aforementioned cost function to the quadratic programming solution and rolling time-domain control. And the hard constraints are transformed into the standard quadratic programming mathematical form: ; ; in The optimal control input sequence to be solved ; It is a positive definite Hessian matrix; The gradient vector; and These are matrices and vectors used to describe linear inequality constraints.

[0065] The processor invokes either the interior-point method or the effective set method solver to numerically solve the quadratic programming problem. If the solution is successful, the output includes the future... Optimal control input sequence of step control quantity If the solution fails, an infeasibility status code is generated and transmitted to the proactive risk defense module.

[0066] If the solution is successful, the processor, based on the rolling time-domain control principle, extracts only the first component of the optimal control input sequence. This serves as the nominal control law for the current moment, and is then distributed to the underlying actuators or output as an auxiliary signal. In the next moment, the processor repeats the above construction and solution process based on the new state measurements.

[0067] The proactive risk defense module is used to monitor the feasibility of the solution. When there is no solution, it determines that the uncertainty of neural conduction exceeds the limit and triggers mechanical damping locking or reverse braking command. Furthermore, the proactive risk defense module includes: Real-time reading of the return status codes of the numerical optimization solver in the predictive control solver module; When the return status code indicates that the optimal solution has been found, a flexible auxiliary instruction containing fine-tuning information on the action rate is generated. When the returned status code indicates that the problem is unsolvable or infeasible, it is determined that the current neural conduction uncertainty has led to the closure of the safe and feasible domain, and a rigid blocking instruction with the highest interrupt priority is immediately generated.

[0068] The proactive risk defense module also includes: Obtain the current real-time angular velocity data of the target joint; Based on the principle of viscous damping, the braking torque value that is opposite to the real-time angular velocity and proportional to its amplitude is calculated in order to construct a smooth braking curve. The braking torque value is encapsulated as a digital control signal and output as a rigid stop command to the underlying drive interface to trigger the braking action of the external actuator.

[0069] Specifically, the proactive risk defense module is located at the execution end of the control loop, responsible for converting the mathematical calculation results of the predictive control solver into control commands for the physical actuators. This module switches between flexible adjustment and rigid blocking modes based on the solver's feasibility status.

[0070] To perform feasibility monitoring and pattern determination, the processor obtains the return status codes of the quadratic programming solver in the predictive control solver module in real time via memory sharing or register reads. This status code indicates whether there exists a solution that satisfies the constraints within the current nominal safety constraint set.

[0071] Processor execution logic decision: when status code The presence of an optimal solution indicates that the current neural conduction uncertainty is within the system's tolerable range, and the tubular constraint is not closed. The processor extracts the first component of the optimal control sequence, converts it into an action rate fine-tuning signal, generates flexible auxiliary commands, and sends them to the human-machine interface. When the status code... When the indication is unsolvable or infeasible, it means that the robust positive invariant set determined by the time-delay variance is too large, causing the shrunk nominal safety constraint set to become an empty set. At this point, the system is determined to face an uncontrollable risk of going out of bounds. The processor immediately sets the system interrupt flag, suspends other tasks, and generates a rigid blocking instruction with the highest priority.

[0072] Based on the calculation of braking torque using viscous damping, after triggering the rigid stop command, the processor immediately reads the current real-time angular velocity data of the target joint through the multimodal sensing interface. To avoid impact damage to the user's joints from a sudden, hard stop, the module constructs a smooth braking curve based on the principle of viscous damping and dynamically calculates the required reverse braking torque. Braking torque The calculation formula is as follows: ; in The target braking torque output at the current moment, in Newton-meters; The real-time angular velocity of the joint is expressed in radians per second. This is the preset viscous damping coefficient, measured in Newton-meter-seconds per radian. The negative sign indicates that the direction of the torque is opposite to the direction of the angular velocity, thus creating an effect that hinders motion.

[0073] This calculation logic ensures that the magnitude of the braking torque is proportional to the speed of movement. When the user's speed is high, the system outputs a large braking torque to quickly curb potential energy; as the speed decreases, the braking torque automatically and smoothly decreases until the speed is zero and the torque disappears, thus achieving a flexible yet decisive braking effect.

[0074] Signal encapsulation and low-level output: the processor will calculate the floating-point braking torque value. The signal is quantized and encoded, and then encapsulated into a digital control signal that can be recognized by the underlying driver. This signal includes torque amplitude, direction bit, and enable bit.

[0075] The processor sends the digital control signal to the servo drive unit of the external motor via a CAN bus or PWM interface. The servo drive unit adjusts the stator current accordingly, applying a physical reverse torque to the output shaft within milliseconds to forcibly lock the mechanical structure or provide high-damping protection, thereby physically blocking the risk at the moment it is determined at the algorithm level.

[0076] Example 2: In high-load resistance training applications using intelligent strength training equipment, as users approach their exhaustion threshold, the transmission of control commands from the central nervous system to the muscles to produce a mechanical response is no longer instantaneous or constant, but exhibits significant random time-varying delays and signal-dependent noise. However, existing technologies are all based on deterministic dynamic models, relying solely on fixed physical positions or velocity thresholds for passive protection, neglecting the destructive effect of increased neural delay variance on the closed-loop stability of the human-machine coupling system. The core technical problem arising from this is: how to overcome the deficiency of traditional fixed safety boundaries that cannot dynamically contract with decreasing neural efficiency under conditions where neural conduction delays drift randomly and are not directly observable, thereby solving the hidden motion instability risk caused by control loop phase lag and achieving prior risk blocking before macroscopic kinematic failures occur. To solve the above problems, this invention provides a national fitness participation data statistics and personalized service recommendation system, the structure of which is as follows: Figure 1 As shown. The specific implementation process of this system is as follows: By eliminating sensor asynchronous errors through a multimodal sensing interface, a temporal benchmark for neural commands and limb responses is established. The stochastic dynamics mapping module utilizes Itō-type stochastic differential equations to introduce time-varying delays and Wiener process noise, restoring the random fluctuation characteristics of neural signals under fatigue conditions. The delay distribution identification module analyzes the cross-correlation function in real time, quantifying unobservable neural fatigue into a statistical index of delay variance. The robust constraint reconstruction module dynamically adjusts the coverage of the robust positive invariant set based on this variance and shrinks the preset physical safety boundary through set subtraction operations, automatically reserving a safety buffer space when neural tremors intensify. The predictive control solution module solves for the optimal control sequence within the shrunken nominal constraints. Once the solver indicates no solution, meaning the current uncertainty exceeds the system's controllable domain, the active risk defense module immediately triggers mechanical damping locking or reverse braking. This technical solution overcomes the shortcomings of traditional fixed boundaries in dealing with random drift due to neural delays, upgrading passive post-event protection to a priori risk blocking based on mathematical model feasibility analysis, ensuring that the human motion trajectory is still limited to an absolutely safe range even under the maximum probability delay error.

[0077] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for statistical analysis of public fitness participation and personalized service recommendation, characterized in that: Includes a processor and memory, the processor being configured to execute the following modules: The multimodal sensing interface module is used to synchronously acquire the target joint angular acceleration and the electromyographic signals on the surface of the active muscle group, extract the electrical envelope of the myocardial wires as a representation of neural drive, and complete the data alignment preprocessing. The stochastic dynamics mapping module is used to construct Ito-type stochastic differential equations for human joints that incorporate stochastic time-varying delay parameters and Wiener process noise terms, in order to characterize the uncertainty of nerve conduction. The time delay distribution identification module is used to calculate the cross-correlation function between neural drive representation and angular acceleration, and extract lag time samples to update the mean of time delay parameters and the variance representing uncertainty in real time. The robust constraint reconstruction module is used to determine the spatial coverage of the minimum robust positive invariant set based on variance, and to perform set subtraction operation to subtract the spatial coverage from the preset set of physical safety boundaries containing anatomical limits to generate a shrunken nominal safety constraint set. The predictive control solution module is used to take the nominal safety constraint set as hard constraints, construct and solve the quadratic optimal control problem in the finite time domain for the disturbance-free nominal model in the stochastic differential equation, so as to generate the optimal control input sequence containing future continuous time step control variables, and determine the first time step control component in the optimal control input sequence as the nominal control law; The proactive risk defense module is used to monitor the feasibility of the solution. When there is no solution, it determines that the uncertainty of neural conduction exceeds the limit and triggers mechanical damping locking or reverse braking command.

2. The national fitness participation data statistics and personalized service recommendation system according to claim 1, characterized in that, The multimodal sensing interface module includes: The surface electromyography sensor is controlled to acquire raw electromyography signals. Bandpass filtering and full-wave rectification are performed on the raw electromyography signals to remove power frequency interference. The linear envelope is extracted by a low-pass filter with a set cutoff frequency to construct a time-series signal reflecting the driving intensity of the central nervous system. The control angular acceleration sensor collects real-time motion data of the target joint, performs second-order differential processing on the real-time motion data to obtain an angular acceleration sequence, and uses a Kalman filter algorithm to smooth and denoise the angular acceleration sequence; Based on a unified high-precision clock source, the linear envelope and the angular acceleration sequence are resampled, and the two are aligned to the same time axis through an interpolation algorithm.

3. The national fitness participation data statistics and personalized service recommendation system according to claim 1, characterized in that, The stochastic dynamics mapping module includes: Construct joint rigid body dynamics equations that include rotational inertia matrix, Coriolis force term, centrifugal force term and gravity term, and use them as the drift term of the Ito-type stochastic differential equation; An input delay variable is introduced into the control input channel of the drift term, and the input delay variable is modeled as a stochastic process modulated by neural fatigue. A diffusion term matrix related to the system state amplitude and control input amplitude is constructed. The standard Wiener process is weighted using the diffusion term matrix to simulate the signal-dependent multiplicative noise generated during neural signal transmission, thus completing the construction of the Iton-type stochastic differential equation.

4. The national fitness participation data statistics and personalized service recommendation system according to claim 1, characterized in that, The time delay distribution identification module includes: Construct an observation window that slides along the time axis and extract the neural drive representation fragment and angular acceleration sequence fragment within the current window; Calculate the cross-correlation coefficient between the neural drive representation segment and the angular acceleration sequence segment at different lag times, and search for the lag time corresponding to the maximum cross-correlation coefficient as the instantaneous time delay observation value at the current moment; A recursive least squares algorithm with a forgetting factor is used to process the instantaneous time-delay observations at historical moments, and the probability density function parameters of the random time-varying time-delay parameters are updated in real time.

5. The national fitness participation data statistics and personalized service recommendation system according to claim 4, characterized in that, The time delay distribution identification module also includes: The expected value and variance of the time delay are extracted from the parameters of the probability density function. The time-delay variance value is transmitted to the robust constraint reconstruction module as a dynamic indicator for quantifying the uncertainty of neural transmission. The expected time delay value is transmitted to the predictive control solution module for calibrating the input delay in the nominal dynamics model.

6. The national fitness participation data statistics and personalized service recommendation system according to claim 1, characterized in that, The robust constraint reconstruction module includes: A linear state feedback controller is pre-designed to stabilize the error dynamics system of the stochastic differential equation; Based on the stability characteristics and time delay variance of the error dynamics system, a minimum robust positive invariant set that can contain all possible state error trajectories is calculated. A scaling rule is established such that the boundary range of the minimum robust positive invariant set monotonically expands as the time delay variance increases, thereby determining a safe buffer space to accommodate the uncertainty of neural conduction.

7. The national fitness participation data statistics and personalized service recommendation system according to claim 1, characterized in that, The robust constraint reconstruction module also includes: Obtain the set of physical safety boundaries described by a polyhedron, which is enclosed by multiple hyperplanes that define anatomical and mechanical limits; Perform the Pontryagin set difference operation to translate each hyperplane of the physical security boundary set toward the center of the set. The translation distance is determined by the spatial coverage of the minimum robust positive invariant set in the corresponding dimension. The contracted polyhedral space enclosed by the translated hyperplane is defined as the nominal safety constraint set.

8. The national fitness participation data statistics and personalized service recommendation system according to claim 1, characterized in that, The predictive control solution module includes: Construct a quadratic cost function, which includes a penalty term for the system state deviating from the reference trajectory and a penalty term for the energy consumption of the control input; The nominal safety constraint set is set as the hard constraint condition for the constraint state variables and control variables in the optimization problem; Within each control cycle, a quadratic programming algorithm is used to solve for the optimal control input sequence that minimizes the quadratic cost function within the finite prediction time domain, and the first time-step control component in the optimal control input sequence is determined as the nominal control law.

9. The national fitness participation data statistics and personalized service recommendation system according to claim 1, characterized in that, The proactive risk defense module includes: Real-time reading of the return status code of the numerical optimization solver in the predictive control solver module; When the return status code indicates that the optimal solution has been found, a flexible auxiliary instruction containing fine-tuning information on the action rate is generated. When the returned status code indicates that the problem is unsolvable or infeasible, it is determined that the current neural conduction uncertainty has led to the closure of the safe and feasible domain, and a rigid blocking instruction with the highest interrupt priority is immediately generated.

10. The national fitness participation data statistics and personalized service recommendation system according to claim 9, characterized in that, The proactive risk defense module also includes: Obtain the current real-time angular velocity data of the target joint; Based on the principle of viscous damping, the braking torque value that is opposite in direction and proportional in magnitude to the real-time angular velocity is calculated to construct a smooth braking curve. The braking torque value is encapsulated into a digital control signal and output as the rigid stop command to the underlying drive interface to trigger the braking action of the external actuator.

Images