A muscle stimulation modulation system and method based on real-time dynamic optimization

Abstract

This invention discloses a muscle stimulation modulation system and method based on real-time dynamic optimization, comprising: a real-time signal acquisition module for continuously acquiring multimodal physiological signals of the target limb, including at least electromyography (EMG) signals, joint kinematic angle signals, and neural electrical signals; a state assessment and prediction module for calculating real-time human state indicators based on the multimodal physiological signals, fusing historical state data, and generating future state trend prediction information through a time series prediction model; and a dynamic parameter optimization module for solving the optimal combination of stimulation parameters for the neuromuscular stimulator in real-time using an embedded optimization algorithm, based on the real-time human state indicators and state trend prediction information, with minimizing the objective function as the core. The stimulation parameter combination includes stimulation pulse intensity, frequency sequence, and waveform duty cycle. This invention enables better muscle stimulation modulation.

Description

Technical Field

[0001] This invention relates to the field of regulation systems, and more specifically to a muscle stimulation regulation system and method based on real-time dynamic optimization. Background Technology

[0002] Current muscle stimulators have significant limitations in real-time dynamic optimization and adjustment: most muscle stimulators use preset fixed stimulation parameters (such as fixed stimulation intensity, frequency, and waveform duty cycle), and cannot dynamically adjust parameters according to the real-time physiological state of the human body. They cannot continuously collect multimodal physiological data such as electromyography signals, joint kinematic angle signals, and nerve electrical signals to sense dynamic changes such as muscle activity intensity, joint movement deviation, and muscle fatigue state, nor can they adjust stimulation parameters based on these real-time status feedbacks. This leads to a mismatch between the stimulation effect and the human body's needs: if the parameters are too strong, it is easy to cause user discomfort or excessive muscle fatigue; if the parameters are insufficient, it is difficult to achieve the expected adjustment target. At the same time, they lack the ability to adapt to changes in muscle state in real time and cannot cope with the dynamic fluctuations of the human body's state during training.

[0003] Existing real-time dynamically optimized muscle stimulation modulation systems still have several shortcomings: State assessment relies on a single physiological signal, failing to integrate multi-dimensional data such as the time-domain energy change rate of electromyography signals, the deviation of joint kinematic angles from standard angles, muscle fatigue index, and previous state indicators. This leads to inaccurate judgment of the body's state and an inability to comprehensively reflect the combined state of muscle activity, joint movement, and fatigue. Dynamic parameter optimization lacks foresight and constraint; some systems optimize parameters only based on the current state, failing to predict future state trends using time-series prediction models and lacking a constrained model predictive control framework. This makes it difficult to achieve long-term stable stimulation effects while ensuring parameter smoothness and user comfort, and also fails to strictly limit stimulation parameters within a safe range. Furthermore, the feedback loop mechanism is incomplete. While it can dynamically adjust the parameters of the state assessment model and the weights of the objective function for parameter optimization based on the deviation between real-time physiological feedback signals and expected goals, it lacks integration with users' subjective feelings (such as comfort and effect satisfaction), resulting in a lack of personalization in regulation. Furthermore, it suffers from insufficient safety protection mechanisms, failing to construct multi-dimensional stimulation safety indices (such as combining stimulation intensity thresholds with heart rate variability baselines) and abnormal movement index monitoring, or lacking tiered protection strategies, making it difficult to identify and respond promptly to risks such as stimulation parameter over-limits and abnormal movements. Finally, it lacks cross-user collaborative optimization capabilities, with the local system relying solely on its own data to train the model, resulting in poor generalization ability and an inability to share multi-user optimization experience, leading to low model iteration efficiency and insufficient adaptability to different users. Therefore, this paper proposes a muscle stimulation regulation system and method based on real-time dynamic optimization. Summary of the Invention

[0004] The present invention solves the above-mentioned technical problems through the following technical solutions, the present invention comprising: The real-time signal acquisition module is used to continuously acquire multimodal physiological signals of the target limb. The multimodal physiological signals include at least electromyography signals, joint kinematic angle signals, and nerve electrical signals. The state assessment and prediction module is connected to the real-time signal acquisition module. It is used to calculate real-time human state indicators based on multimodal physiological signals, and integrate historical state data to generate state trend prediction information for future periods through a time series prediction model. The dynamic parameter optimization module, connected to the state assessment and prediction module, is used to solve the optimal combination of stimulation parameters for the neuromuscular stimulator in real time through an embedded optimization algorithm, based on real-time human state indicators and state trend prediction information, with minimizing the objective function as the core. The combination of stimulation parameters includes stimulation pulse intensity, frequency sequence and waveform duty cycle. The stimulus execution and verification module, connected to the dynamic parameter optimization module, is used to load and execute the optimal combination of stimulus parameters and collect real-time physiological feedback signals after stimulus execution to verify the effectiveness of the stimulus. The adaptive feedback closed-loop module, connected to the stimulus execution and verification module and the real-time signal acquisition module, is used to dynamically adjust the model parameters of the state assessment and prediction module and the objective function weights of the dynamic parameter optimization module based on the deviation between the instantaneous physiological feedback signal and the expected target, thus forming a closed-loop control.

[0005] Furthermore, the state assessment and prediction module calculates real-time human state indicators using the following formula. : ; in, This represents the rate of energy change of the electromyography signal acquired by the real-time signal acquisition module in the time domain; This represents the current joint kinematic angle vector obtained by the real-time signal acquisition module. With respect to the preset standard motion angle vector The Euclidean distance between them; The muscle fatigue index is calculated based on frequency domain analysis of electromyography signals. The human body status indicators at the previous sampling time; , , and The dynamic weighting coefficient has an initial value set based on individual user information and is adjusted by the adaptive feedback closed-loop module during system operation, while satisfying preset constraints.

[0006] Furthermore, the dynamic parameter optimization module adopts a constrained model predictive control framework, whose core objective function is... Used to evaluate stimulus parameter vectors (Representing stimulus intensity, frequency, and duty cycle, respectively) in the future prediction time domain Overall effect within, objective function The definition is as follows: ; in, The state assessment and prediction module is based on the current state, historical data, and stimulus parameter vectors. The predicted value of the human body state index σ(t+k) at time k in the future; An ideal state indicator set according to rehabilitation goals; The median value of the preset comfort parameter range for the user; This represents the parameter change within adjacent control cycles; and This is the penalty weighting coefficient; The optimization process satisfies Under hard constraints, online solution enables Minimize the optimal parameter sequence .

[0007] Furthermore, the dynamic parameter optimization module embeds a parameter sensitivity analysis unit for online evaluation and updating of dynamic weighting coefficients. , , and For ease of calculation, these dynamic weighting coefficients are... , , and Represented in vector form Its update strategy is based on recursive least squares, and the specific process is as follows: ; ; in, The input feature vector consists of the various calculation elements of the state index σ(t) at time t; The current state prediction error; It is the covariance matrix; It is a forgetting factor.

[0008] Furthermore, it also includes a safety monitoring unit integrated into the adaptive feedback closed-loop module, which calculates the stimulus safety index in real time. : The Stimulation Safety Index The calculation formula is: ; in, For real-time stimulus intensity, The strength safety threshold is set based on user historical data. The heart rate variability is obtained by the real-time signal acquisition module. For the user's resting heart rate variability baseline, and These are the shape adjustment parameters; when When the first threshold is exceeded, the system triggers a level one alarm and automatically reduces the stimulation parameters to a safe range; when When the stimulation exceeds a higher second threshold, the system immediately stops stimulating the system.

[0009] Furthermore, the safety monitoring unit is also used to calculate the motion abnormality index. To detect abnormal movement patterns, its calculation is based on joint angle vectors acquired by a real-time signal acquisition module. : ; in, This is the second derivative of the joint angle vector, i.e., angular acceleration; The weighting function is dynamically calculated based on the amplitude of angular acceleration; T is the integration time window. when When the preset threshold is exceeded, the safety monitoring unit determines that abnormal movement has occurred and instructs the stimulation execution module to pause the current stimulation program and switch to the antispasmodic low-frequency stimulation mode.

[0010] Furthermore, it also includes a user interaction and subjective feedback module, used to receive subjective feeling ratings input by users through their terminal devices. and comfort feedback ; The adaptive feedback closed-loop module utilizes this subjective feeling score and comfort feedback Real-time human condition indicators After making corrections, a comprehensive state index integrating subjective and objective information is obtained. : ; in, This is the upper limit of the scoring range. and The fusion coefficient; the system automatically adjusts by continuously learning user feedback patterns. and This allows the optimization goals to better align with the user's personalized experience. Furthermore, the system connects to multiple user terminals through a cloud-based collaborative learning platform, which performs the following operations: Aggregate local optimization data from each anonymized user, including successful state-parameter mapping relationships and optimization trajectories; Based on aggregated data, a global deep reinforcement learning model is trained periodically, and the reward function of this model is... Defined as: ; Among them, expectations Calculate the average over all users i. To measure the smoothness of changes in user i stimulus parameters, For smoothing weights; The trained global model parameters are distributed to each local system for initialization or fine-tuning of their local dynamic parameter optimization modules, thereby leveraging the global model to improve performance. Improve prediction accuracy and enable collaborative optimization across users.

[0011] A method for muscle stimulation modulation based on real-time dynamic optimization includes the following steps: Real-time signal acquisition steps: Continuously acquire multimodal physiological signals of the target limb, wherein the multimodal physiological signals include at least electromyography signals, joint kinematic angle signals and nerve electrical signals; State assessment and prediction steps: Receive multimodal physiological signals output from the real-time signal acquisition step, calculate real-time human state indicators based on the multimodal physiological signals, and simultaneously integrate historical state data to generate future state trend prediction information through a time series prediction model; Dynamic parameter optimization step: Receive the real-time human state indicators and state trend prediction information output from the state assessment and prediction step, and solve the optimal combination of stimulation parameters of the neuromuscular stimulator in real time through an embedded optimization algorithm with minimizing the objective function as the core. The combination of stimulation parameters includes stimulation pulse intensity, frequency sequence and waveform duty cycle. Stimulation execution and verification steps: Receive the optimal combination of stimulation parameters output from the dynamic parameter optimization step, load and execute the optimal combination of stimulation parameters, and simultaneously collect the immediate physiological feedback signal after stimulation execution, and verify the effectiveness of the stimulation based on the immediate physiological feedback signal. The adaptive feedback closed-loop process involves receiving the immediate physiological feedback signal output from the stimulus execution and verification step, while simultaneously establishing data interaction with the real-time signal acquisition step. Based on the deviation between the immediate physiological feedback signal and the expected target, the model parameters used in the state assessment and prediction step, as well as the objective function weights used in the dynamic parameter optimization step, are dynamically adjusted to form a closed-loop control.

[0012] Compared with existing technologies, this invention has the following advantages: This real-time dynamic optimization-based muscle stimulation regulation system and method continuously collects multimodal physiological signals such as electromyography signals, joint kinematic angle signals, and nerve electrical signals of the target limb, providing comprehensive data support for subsequent processing; it calculates real-time human state indicators based on multimodal physiological signals and integrates historical data to generate future state trend prediction information through a time series prediction model, enabling dynamic understanding and prediction of human state; using a minimization objective function as the core, an embedded optimization algorithm solves for the optimal combination of stimulation parameters such as stimulation pulse intensity, frequency sequence, and waveform duty cycle of the neuromuscular stimulator in real time under parameter constraints, ensuring that the stimulation parameters are adapted to the human state; loading and executing the optimal stimulation parameter combination and collecting immediate physiological feedback signals to verify the effectiveness of the stimulation allows for timely acquisition of relevant data after stimulation execution; and it dynamically adjusts the model parameters of the state assessment and prediction module based on the deviation between the immediate physiological feedback signal and the expected target. The objective function weights of the number and dynamic parameter optimization module form a closed-loop control, which can continuously optimize system operation. The safety monitoring unit calculates the stimulation safety index in real time. When the index exceeds different thresholds, it triggers a first-level alarm to reduce the parameters to a safe range or immediately stop the stimulation. At the same time, it calculates the motion abnormality index to detect abnormal motion patterns. When the index exceeds the threshold, it pauses the current stimulation program and switches to an anti-spasmodic low-frequency stimulation mode to ensure the safety of the stimulation process. The user interaction and subjective feedback module receives the user's subjective feeling score and comfort feedback. The adaptive feedback closed-loop module uses this feedback to correct the real-time human body state index to obtain a comprehensive state index and automatically adjusts the fusion coefficient to meet the user's personalized needs. The cloud-based collaborative learning platform aggregates the local optimization data of each anonymized user to train a global deep reinforcement learning model. The trained model parameters are sent to each local system for initialization or fine-tuning of the dynamic parameter optimization module, which can improve the prediction accuracy of human body state index and realize cross-user collaborative optimization, making the system more worthy of promotion and use. Attached Figure Description

[0013] Figure 1 This is an overall structural diagram of the present invention. Detailed Implementation

[0014] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0015] like Figure 1 As shown, this embodiment provides a technical solution: a muscle stimulation modulation system based on real-time dynamic optimization, comprising: The real-time signal acquisition module is used to continuously acquire multimodal physiological signals of the target limb. The multimodal physiological signals include at least electromyography signals, joint kinematic angle signals, and nerve electrical signals. The state assessment and prediction module is connected to the real-time signal acquisition module. It is used to calculate real-time human state indicators based on multimodal physiological signals, and integrate historical state data to generate state trend prediction information for future periods through a time series prediction model. The dynamic parameter optimization module, connected to the state assessment and prediction module, is used to solve the optimal combination of stimulation parameters for the neuromuscular stimulator in real time through an embedded optimization algorithm, based on real-time human state indicators and state trend prediction information, with minimizing the objective function as the core. The combination of stimulation parameters includes stimulation pulse intensity, frequency sequence and waveform duty cycle. The stimulus execution and verification module, connected to the dynamic parameter optimization module, is used to load and execute the optimal combination of stimulus parameters and collect real-time physiological feedback signals after stimulus execution to verify the effectiveness of the stimulus. The adaptive feedback closed-loop module, connected to the stimulus execution and verification module and the real-time signal acquisition module, is used to dynamically adjust the model parameters of the state assessment and prediction module and the objective function weights of the dynamic parameter optimization module based on the deviation between the instantaneous physiological feedback signal and the expected target, thus forming a closed-loop control.

[0016] The state assessment and prediction module calculates real-time human state indicators using the following formula. : ; in, This represents the rate of energy change of the electromyography signal acquired by the real-time signal acquisition module in the time domain; This represents the current joint kinematic angle vector obtained by the real-time signal acquisition module. With respect to the preset standard motion angle vector The Euclidean distance between them; The muscle fatigue index is calculated based on frequency domain analysis of electromyography signals. The human body status indicators at the previous sampling time; , , and The dynamic weighting coefficients are initially set based on individual user information and are adjusted by the adaptive feedback closed-loop module during system operation, while satisfying preset constraints. By integrating the time-domain energy change rate of electromyography (EMG) signals, the deviation of joint kinematic angles from standard angles, muscle fatigue index, and state indicators from the previous sampling time into a formula, rather than relying on a single physiological signal, this approach comprehensively reflects the current intensity of muscle activity, the degree of joint movement deviation, and muscle fatigue state. The initial values ​​of the dynamic weights α, β, γ, and δ are set based on individual user information (such as age, physical condition, and rehabilitation stage) and can be adjusted by the adaptive feedback closed-loop module, while also satisfying the constraint α+β+γ+δ=1. This allows for flexible adjustment of the influence weights of each assessment dimension according to the core needs of different users or different training periods for the same user. The formula incorporates the state indicator σ(t-1) from the previous sampling time, avoiding assessment bias caused by fluctuations in physiological signals at a single moment (such as transient interference from EMG signals or instantaneous deviations in joint angles), making the real-time human state indicator σ(t) more closely reflect the continuous process of changes in human state.

[0017] Taking the upper limb dumbbell lifting training scenario of an ordinary adult as an example, the trainee is a 25-year-old adult with a fitness foundation of six months. The core goal is to maintain the standard of the lifting movement while paying attention to the state of muscle fatigue. The initial dynamic weights are set as α=0.2, β=0.4, γ=0.2, δ=0.2 (satisfying α+β+γ+δ=1). At this time, the focus is on the evaluation of the standard of the movement, so the weight of β is relatively high.

[0018] When the training reached t=15min, the real-time signal acquisition module obtained the following data: the time-domain energy change rate of the electromyography signal. Current shoulder joint angle vector Angle vector of standard lifting motion Euclidean distance Muscle fatigue index obtained based on electromyography frequency domain analysis Status indicators at the previous sampling time .

[0019] Substitute the above data into the real-time human condition index calculation formula: ; Calculate each item step by step: ; ; ; ; The final calculation is .

[0020] As training continued (t=25min), muscle fatigue intensified in the trainees. The adaptive feedback closed-loop module adjusts its weights to α=0.2, β=0.3, γ=0.3, and δ=0.2 (still satisfying a sum of 1). At this point, the weight of γ is increased to strengthen the assessment of muscle fatigue. The data is then re-introduced (the time-domain energy change rate of the electromyography signal remains 0.5mV). 2 / s, the joint angle deviation is still 3°, and the previous σ value was 14.4074). Calculate:

[0021] ; Through dynamic adjustment of weights It can accurately match the current joint movement deviation and muscle fatigue status, demonstrating the advantages of this case in multi-dimensional and dynamic adaptation assessment.

[0022] Through dynamic adjustment of weights It can accurately match the current combined state of joint deviation and muscle fatigue, demonstrating the advantages of this case in multi-dimensional and dynamic adaptation assessment.

[0023] The dynamic parameter optimization module adopts a constrained model predictive control framework, whose core objective function is... Used to evaluate stimulus parameter vectors (Representing stimulus intensity, frequency, and duty cycle, respectively) in the future prediction time domain Overall effect within, objective function The definition is as follows: ; in, The state assessment and prediction module is based on the current state, historical data, and stimulus parameter vectors. The predicted value of the human body state index σ(t+k) at time k in the future; An ideal state indicator set according to rehabilitation goals; The median value of the preset comfort parameter range for the user; This represents the parameter change within adjacent control cycles; and This is the penalty weighting coefficient; The optimization process satisfies Under hard constraints, online solution enables Minimize the optimal parameter sequence ; The above process employs a constrained model predictive control framework, with the objective function based on the future prediction time domain. Evaluating the effects of stimulus parameters, rather than focusing solely on the current state, can proactively avoid subsequent deviations in the human body's state caused by inappropriate current parameter selection, ensuring the long-term stability of the stimulus effect; the objective function should incorporate the median of the preset comfort parameter range. The deviation penalty, and the change in parameters between adjacent control cycles. The penalty can reduce user discomfort caused by stimulation parameters deviating from the comfort range, and avoid abrupt stimulation to the human body caused by sudden parameter changes; the optimization process needs to meet the following requirements. The hard constraints can strictly limit the intensity, frequency, and duty cycle of the stimulus pulse within a safe range, preventing risks caused by parameter over-limitation; the objective function integrates three core indicators: human state deviation, comfort deviation, and parameter change smoothness. By adjusting the weights, different optimization needs can be balanced, comprehensively improving the overall performance of the system regulation.

[0024] Taking assisted lower limb squat training for a 28-year-old adult male as an example, the core goal is to maintain stable muscle activity during training (setting ideal state indicators). The system parameters are set as follows: Stimulus parameter vector ( Stimulus intensity For frequency, (Duty cycle) Future prediction time domain (Covering the next 5 control cycles, each cycle lasting 10 seconds) Control Time Domain (Optimize parameters for the next 3 control cycles) User preset comfort parameter range median value ; Penalty weight (Comfort deviation penalty) (Penalty for parameter changes) Stimulus parameter hard constraints , ; Input data: Based on the state assessment and prediction module, the predicted values ​​of the human body state at the next 5 time points under the current stimulus parameters are obtained: ; Parameter changes over the next 3 control cycles: ; ; ; Parameters for the next 3 control cycles and deviation: ; ; ; Objective function calculation: Substitute the data into the objective function formula: ; Calculation of state deviation term: = ; Comfort deviation penalty calculation: ; Calculation of penalty term for parameter change: ; Total value of the objective function: ; Solving for optimal parameters: The system satisfies Under hard constraints, adjustments are made through embedded optimization algorithms. The value was calculated repeatedly. Finally found The smallest (approximately 42.3) optimal parameter sequence This parameter ensures that within the next 5 periods With a value close to 80, which is also close to the comfort range and the parameters change gradually, this case demonstrates the advantages of optimization.

[0025] The dynamic parameter optimization module embeds a parameter sensitivity analysis unit for online evaluation and updating of dynamic weighting coefficients. , , and For ease of calculation, these dynamic weighting coefficients are... , , and Represented in vector form Its update strategy is based on recursive least squares, and the specific process is as follows: ; ; in, The input feature vector consists of the various calculation elements of the state index σ(t) at time t; The current state prediction error; It is the covariance matrix; Forgetting factor; By embedding a parameter sensitivity analysis unit and combining recursive least squares method, dynamic weight coefficients are evaluated and adjusted in real time without offline manual adjustment. This adapts to the dynamic changes in human body state over time, avoiding the problem of fixed weights failing to match real-time physiological signal characteristics. The recursive least squares method calculates weights through a recursive approach, eliminating the need to reprocess all historical data. It updates the weights based solely on the current input feature vector φ(t), state prediction error e(t), and the covariance matrix P(t-1) from the previous time step, meeting the system's real-time requirements and making it suitable for dynamic scenarios involving muscle stimulation regulation. The input feature vector φ(t) directly corresponds to the human body... The core calculation elements of the body state index σ(t) (electromyography energy change rate, joint angle deviation, muscle fatigue index, and previous state index) and the state prediction error e(t) reflect the deviation between the current state and the target. Based on these two, the weights are updated, which can accurately adjust the weight ratio for key dimensions that affect state assessment and avoid blind adjustment. Through the recursive update of the covariance matrix P(t), the accuracy and sensitivity of the weight coefficient estimation can be tracked in real time, and the influence of historical data and current data on the weights can be balanced (by adjusting the forgetting factor λ), preventing sudden changes in weights due to signal fluctuations at a single moment and ensuring the continuity of state assessment.

[0026] For example, a scenario involving assisted lower body squat training for a 28-year-old adult male (core objective) The system parameters are set as follows: Dynamic weighting coefficients at the previous sampling time (t-1) (Satisfies α+β+γ+δ=1) Covariance matrix at the previous sampling time (t-1)=I (4×4 identity matrix, with consistent parameter estimation accuracy under initial conditions) Forgetting factor λ = 0.95 (balancing the stability of historical data with the timeliness of current data) Error in state prediction at time t (σ(t) = 76 at time t) Input data: Input feature vector at time t : ; Dynamic weight coefficient w(t) update calculation: Substitute into the update formula for w(t): ; Key terms for calculating the numerator and denominator: :because For the identity matrix, the result is equal to That is, [0.4,2,0.3,78]^T :Right now The sum of squares of all elements is calculated to be 0.4. 2 +2 2 +0.3 2 +78 2 =0.16 + 4 + 0.09 + 6084 = 6088.25 Denominator: =0.95 + 6088.25 = 6089.2 Coefficient terms: ; After the update : ; covariance matrix Update calculation Substitution Updated formula:;

[0027] Calculate key items = (Since P(t-1) is the identity matrix), the result is a 4×4 matrix: ; After dividing this term by the denominator 6089.2, the matrix element values ​​are extremely small (e.g., 6084 / 6089.2≈0.9991). =1 / 0.95≈1.0526 After the update : Calculated Each element is slightly adjusted compared to P(t-1), while the core diagonal elements remain close to 1, reflecting the tracking of parameter estimation accuracy by the covariance matrix and avoiding excessive fluctuations in weight updates.

[0028] Results Explanation: Updated The weight of the previous moment's state indicator δ was increased from 0.2 to approximately 0.2509. Since σ(t) = 76 at time t and the deviation from the target e(t) = 4, it is necessary to enhance the influence of the previous moment's state on the current assessment to improve continuity. Other weights were slightly adjusted to reflect the stable muscle activity and slight increase in fatigue during the current squat training, demonstrating the advantages of this case in updating weights online, accurately, and stably.

[0029] It also includes a safety monitoring unit integrated into the adaptive feedback closed-loop module, which calculates the stimulus safety index in real time. : The Stimulation Safety Index The calculation formula is: ; in, For real-time stimulus intensity, The strength safety threshold is set based on user historical data. The heart rate variability is obtained by the real-time signal acquisition module. For the user's resting heart rate variability baseline, and These are the shape adjustment parameters; when When the first threshold is exceeded, the system triggers a level one alarm and automatically reduces the stimulation parameters to a safe range; when If the stimulation exceeds a higher second threshold, the system immediately stops stimulating the system. By stimulating the safety index The calculation formula integrates two core safety dimensions: stimulus intensity (hardware output parameter) and heart rate variability (human physiological response parameter), rather than relying on a single parameter to determine safety status. This comprehensively reflects the safety of the stimulus operation for the human body, avoiding missed risks due to monitoring only one dimension; the stimulus intensity safety threshold... Based on user historical data, the baseline heart rate variability is set. Using measured values ​​under resting conditions, rather than universal fixed values, allows for precise matching of different users' physiological tolerance and safety boundaries, reducing safety misjudgments due to individual differences. By setting first and second thresholds, a tiered response logic is constructed from warning to intervention to emergency termination, rather than a single trigger termination. This allows for timely avoidance of potential risks by reducing parameters in the early stages, and immediate cessation of stimulation in high-risk situations, balancing safety and training continuity. The safety monitoring unit calculates in real time. It can simultaneously track changes in stimulation parameters and fluctuations in human physiological state. Compared with offline or periodic monitoring, it can more quickly capture safety risks (such as a sudden increase in stimulation intensity or an abnormal decrease in heart rate variability), ensuring safety throughout the stimulation process.

[0030] For example, in a scenario involving assisted lower limb squat training for a 28-year-old adult male, the system safety parameters are set based on the user's historical data and resting test results as follows: Stimulation intensity safety threshold (The maximum intensity in which the user has not experienced any discomfort during previous training) User resting heart rate variability baseline (Measured value under resting state) Shape adjustment parameters (Adapted to the user's physiological signal sensitivity) Safety index grading threshold: First threshold (Triggers Level 1 Alert), Second Threshold (Triggering abort) Input data and Calculation (Level 1 Alarm Scenario) Squat training continued At that time, the real-time signal acquisition module obtains data: Real-time stimulation intensity (near ); Real-time heart rate variability (Due to a slight decrease in training fatigue, it is below) ) Substitute into the formula for calculating the stimulation safety index: ; Stimulus intensity related fraction: ; Heart rate variability related formula: ; Safety Index : ; System response: because This triggers a Level 1 alarm, and the system automatically reduces the stimulus intensity from... (Within a safe range), heart rate variability gradually recovers. , It dropped to 0.78, returning to a safe state.

[0031] Input data and Calculation (Level 2 Abort Scenario): If the equipment parameters suddenly fluctuate during training... Real-time data is as follows: Real-time stimulation intensity (Exceed ); Real-time heart rate variability (significantly lower than) ); Recalculate : Stimulus intensity related fraction: ; Heart rate variability related formula: ; Safety Index : ; System response: because (Second threshold) The system immediately stops the stimulation operation to avoid the risks caused by the superposition of high-intensity stimulation and abnormal physiological state, and waits for the user's heart rate variability to recover. After that, restart the low-intensity stimulation mode.

[0032] The safety monitoring unit is also used to calculate the motion abnormality index. To detect abnormal movement patterns, its calculation is based on joint angle vectors acquired by a real-time signal acquisition module. :

[0033] in, This is the second derivative of the joint angle vector, i.e., angular acceleration; The weighting function is dynamically calculated based on the amplitude of angular acceleration; T is the integration time window. when When the preset threshold is exceeded, the safety monitoring unit determines that abnormal movement has occurred and instructs the stimulation execution module to pause the current stimulation program and switch to the antispasmodic low-frequency stimulation mode. Through the motion abnormality index The calculation uses the second derivative (angular acceleration) of the joint angle vector as the core monitoring indicator, which is more sensitive to identifying sudden and severe joint movement abnormalities (such as twitching and locking) than the first derivative (angular velocity). It also incorporates a dynamic weighting function. This can enhance the identification of high-risk abnormal movements and reduce misjudgments caused by slight fluctuations. When When the threshold is exceeded, the system does not simply stop the stimulation, but instructs the stimulation execution module to pause the current program and switch to an anti-spasmodic low-frequency stimulation mode. This ensures safety while specifically alleviating muscle spasms that may be caused by abnormal movement, better meeting the practical application needs of muscle stimulation regulation; dynamic weighting function The amplitude of angular acceleration is dynamically adjusted, which can flexibly adapt to the abnormal judgment criteria according to different motion stages (such as normal acceleration when squatting and standing up, and sudden abnormal violent acceleration), avoiding misjudgment of normal motion or missed judgment of abnormal motion caused by fixed weights, and improving the detection adaptability.

[0034] If the scenario of assisted lower limb squat training for a 28-year-old adult male is continued, the system's abnormal detection parameters are calibrated and set based on the user's normal squat exercise data: Integral Time Window (Balancing real-time performance with data stability, covering key change periods of a single joint movement) Dynamic weight function The larger the angular acceleration amplitude, the higher the weight, thus strengthening the identification of high-risk anomalies; Motion abnormality index threshold (Based on normal squatting time) The mean plus two standard deviations is set during normal exercise. (usually below 3) Antispasmodic low-frequency stimulation mode parameters: frequency ,strength Duty cycle 0.2 (based on preliminary testing to meet the user's antispasmodic needs) Normal training scenario Below the threshold: Squat training continued (Time window) The knee joint angle vector within ) The angular acceleration sequence was calculated as follows: The change in angular acceleration during a normal squat and stand-up is stable in amplitude. Substitute into the motion abnormality index calculation formula: ; because Substituting, we get: ; Step-by-step calculation (discrete integral approximation, time interval) ): Calculate each time point : ; Integral approximation (summation × time interval): ; calculate : ; System response: because The exercise was determined to be normal, and the system continued to execute the current squatting assisted stimulation protocol (intensity). ,frequency ).

[0035] Abnormal motion scenarios, Exceeding the threshold: Training proceeded to At that moment, the user's knee joint suddenly experienced a slight twitch, which was detected by the real-time signal acquisition module. Internal knee joint angular acceleration sequence: (The amplitude of angular acceleration is significantly increased, exceeding the normal range); Calculate using the same formula : Calculate each time point : ; Integral approximation: ; 3. Calculation : ; System response: because The safety monitoring unit detected abnormal movement and immediately instructed the stimulation execution module to pause the current squatting assisted stimulation program and switch to the anti-spasticity low-frequency stimulation mode (frequency...). ,strength After 30 seconds of continuous stimulation, the user's knee joint spasms subsided, and the angular acceleration returned to normal. Down to The system prompts the user to choose between resuming training or taking a break.

[0036] It also includes a user interaction and subjective feedback module, used to receive subjective feeling ratings input by users through their terminal devices. and comfort feedback ; The adaptive feedback closed-loop module utilizes this subjective feeling score and comfort feedback Real-time human condition indicators After making corrections, a comprehensive state index integrating subjective and objective information is obtained. : ; in, This is the upper limit of the scoring range. and The fusion coefficient; the system automatically adjusts by continuously learning user feedback patterns. and This makes the optimization goals more aligned with the user's personalized experience; Incorporating user subjective feelings Comfort feedback This supplements the user's actual physical sensations (such as muscle soreness tolerance and discomfort from stimulation intensity) that cannot be covered by purely objective physiological signals (such as electromyography and joint angles), avoiding optimization biases where objective indicators are normal but the user's physical sensations are poor, and making the state assessment more in line with actual usage scenarios; the objective physiological state indicators are converted into formulas. By integrating subjective feedback data, a comprehensive status index is generated. This approach retains the quantitative basis of objective physiological states while incorporating the dimension of user subjective experience, enabling the state assessment results to simultaneously reflect physiological functional status and user acceptance, thus providing a more comprehensive decision-making basis for subsequent stimulation parameter optimization. The system can continuously learn user feedback patterns and automatically adjust the fusion coefficients. and Instead of using a fixed coefficient, for example, for users with high somatosensory sensitivity, the [function / effort] can be increased. (Comfort feedback weight); for users who are more focused on training results, optimization is possible. (Subjective feeling rating weight) to achieve personalized status assessment logic and avoid a one-size-fits-all approach to integration.

[0037] If we continue with the scenario of assisted lower limb squat training for a 28-year-old adult male, the core objective is to balance muscle activity effectiveness with user comfort. The system parameters are set based on the user's initial feedback. Upper limit of subjective feeling rating range (10 points represents the most satisfactory training effect, and 0 points represents "extremely unsatisfactory") Initial value of fusion coefficient (The system is preset based on similar user data and will be dynamically adjusted based on feedback.) During training, objective real-time human state indicators are calculated in advance. It reflects the comprehensive objective level of muscle activity intensity and joint deviation; Initial fusion calculation (first feedback): Training in progress hour: Calculated objective indicators (Objective physiological state: moderate muscle activity intensity and small joint angle deviation). Users input their subjective feelings rating via the terminal. (Satisfied with the current training results, but not perfect) Comfort feedback (The range is 0-1, where 1 is "most comfortable" and 0.7 indicates a slight tingling or soreness.) Substitute into the comprehensive status index calculation formula: ; Step-by-step calculation: Subjective feeling correction items: ,but ; Comfort feedback correction items: ,but ; Comprehensive status indicators: ; Calculation of fusion coefficient after adjustment (follow-up feedback): Training proceeded to At that time, the user's muscle soreness increased, and the feedback changed as follows: Subjective feeling rating (Decreased satisfaction with training effectiveness, believing the current stimulus intensity is too strong) Comfort feedback (Discomfort significantly increased) The system learns the pattern of feedback scores decreasing in tandem with a decline in user comfort, and automatically adjusts the fusion coefficient accordingly. (Increase the weight of subjective ratings) (Enhancing the weight of comfort feedback) Objective indicators obtained from calculations during the same period (Objective physiological indicators have increased slightly, due to the previous fine-tuning of stimulation intensity to maintain the effect.) Substitute the values ​​into the formula again to calculate: ; Step-by-step calculation: Subjective feeling correction items: ,but

[0038] Comfort feedback correction items: ,but ; Comprehensive status indicators: ; The increase from 16.808 to 18.972 clearly reflects the overall state of "a slight increase in objective physiological indicators but a significant deterioration in subjective experience." Based on this indicator, the adaptive feedback closed-loop module adjusted the objective function weights of the dynamic parameter optimization module, appropriately reducing the weight of "muscle activity intensity" and increasing the weight of "comfort," ultimately reducing the stimulation intensity from 25mA to 22mA while maintaining the frequency at 30Hz. After the adjustment, user feedback... , The core value is to achieve a balance between "performance and comfort" by dropping back to 17.5.

[0039] The system connects to multiple user terminals through a cloud-based collaborative learning platform, which performs the following operations: Aggregate local optimization data from each anonymized user, including successful state-parameter mapping relationships and optimization trajectories; Based on aggregated data, a global deep reinforcement learning model is trained periodically, and the reward function of this model is... Defined as: ; Among them, expectations Calculate the average over all users i. To measure the smoothness of changes in user i stimulus parameters, For smoothing weights; The trained global model parameters are distributed to each local system for initialization or fine-tuning of their local dynamic parameter optimization modules, thereby leveraging the global model to improve performance. Improve prediction accuracy and enable collaborative optimization across users; Overcoming the limitations of single-device data and improving the adaptability of stimulation parameters: The amount of data stored locally by a muscle stimulator for a single user is limited and the scenario coverage is narrow (such as only containing the user's normal training data), which can easily lead to overfitting of the parameter optimization model. By aggregating anonymized data from 500 similar muscle stimulators through a cloud-based collaborative platform, covering the state-parameter mapping relationship of users of different ages and training levels, the trained global model can adapt to more diverse muscle states (such as the muscle weakness period of beginners and the fatigue period of experienced users), avoiding stimulation bias caused by insufficient experience of a single device.

[0040] Reduce local computing power burden and accelerate response speed: Muscle stimulators are mostly portable devices with limited local computing power. By adopting the cloud training-local loading mode, the device does not need to train a deep reinforcement learning model from scratch. It only needs to receive lightweight parameters from the cloud and fine-tune them. The computing power consumption is reduced by more than 100%. The time from startup to achieving optimal stimulation adjustment is shortened from 15 minutes to 3 minutes, meeting the needs of real-time training.

[0041] Achieving cross-user experience sharing and reducing individual trial-and-error costs: The cloud-based model continuously absorbs muscle stimulation optimization experiences from different users, such as parameter adjustment strategies for muscle soreness during squats in 30-year-old women and low-frequency stimulation programs for high-fatigue states in 45-year-old men, and synchronizes them to each muscle stimulator. When new users use the device, they can directly reuse the optimization logic of similar groups without repeatedly testing stimulation intensity and frequency combinations, thus reducing the rate of subjective discomfort feedback.

[0042] The electromyography and heart rate data collected by the muscle stimulator are sensitive health information. The cloud uses AES-256 encryption for transmission and storage, and differential privacy technology is used for anonymization. Only the mapping relationship between the state index σ and the stimulation parameter Θ is retained, which can realize multi-device data sharing training and avoid leakage of user identity and raw physiological data.

[0043] For example, in the scenario of assisted lower limb squat training for a 28-year-old adult male, the core equipment is a lower limb muscle stimulator, and the parameters of the cloud-based collaborative system are as follows: The cloud connects 500 identical muscle stimulators, aggregates anonymized local optimization data every 7 days, and trains a global deep reinforcement learning model using a federated learning framework. The global model reward function has a smoothing weight ξ=0.4, minimizing the equilibrium state deviation and ensuring smooth parameter adjustment; The male's muscle stimulator was initially equipped with a local model, for future states. The prediction error is about 12%, and the error needs to be reduced to less than 5% after loading the cloud model; Cloud-based data aggregation and model training: Anonymized data collection: The cloud receives desensitized data uploaded by each muscle stimulator, including: State-parameter mapping pairs: For example, the optimal parameters for user A's stimulator when σ=16 (muscle activity intensity). For user B, the optimal Θ is [20mA, 26Hz, 0.25] when σ=14. Optimize trajectory data: such as adjusting user C's stimulator from initial to optimal. During the process, the curve of σ increasing from 13.5 to 15.2 Global reward function calculation: Substitute into the global reward function formula: ; in: E represents the average value of data from 500 stimulators; The ideal state index for user i (approximately 14 for beginners, approximately 18 for experienced users). The smoothness of parameter adjustment is calculated as the average difference between parameters in adjacent adjustment steps; the smaller the value, the smoother the adjustment.

[0044] Calculate the single-round input by selecting three types of typical user data: New user (25 years old) The sum of squared state deviations is The parameter smoothness is 0.8, and the sum of the components is 0.069 + 0.4 × 0.8 ≈ 0.389. Experienced user (35 years old) The sum of squared state deviations is The parameter smoothness is 0.6, and the sum of the components is 0.03 + 0.4 × 0.6 ≈ 0.27; Recovery training user (40 years old, The sum of squared state deviations is With a parameter smoothness of 0.7, the sum of the components is 0.026 + 0.4 × 0.7 ≈ 0.306 After averaging the data from 500 devices, E≈0.32, meaning R_global≈0.32 in the first round; after 10 rounds of iterative training, The error decreased to 0.18, and the prediction error of the model pair stabilized at 4.8%.

[0045] Local adaptation and enhanced effectiveness of muscle stimulators: Model parameter delivery: The cloud sends the global model parameters (including state prediction network weights and activation function coefficients) to the man's muscle stimulator via the MQTT protocol. The device fine-tunes the parameters based on its own squat training data from the past 3 days to complete the adaptation.

[0046] Comparison of effects before and after optimization: Before optimization (local model): The stimulator program adopted ,predict After actual execution, the measured σ=14.5, with an error of approximately 11.7%. Due to the relatively high parameters, users reported muscle soreness and reduced comfort. .

[0047] Optimized (cloud model): Same parameters as planned Model-corrected predictions The measured σ=14.5, with an error of only 1.4%. Improved parameter matching enhances user comfort. Rising to 0.8, subjective score =9.

[0048] Long-term iteration effect: Every 7 days, after the model is updated in the cloud, the stimulator automatically loads the new parameters, and the prediction error is consistently kept within 2%. When the user enters a period of muscle fatigue (fatigue index F rises from 0.3 to 0.6), the device adjusts Θ to [22mA, 27Hz, 0.28] within 1 second based on the optimal strategy for highly fatigued users in the cloud model, without requiring manual adjustments. Compared to stimulators not connected to the cloud, this device reduces the number of parameter optimization trials from 12 to 3, significantly improving training safety and efficiency.

[0049] A method for muscle stimulation modulation based on real-time dynamic optimization includes the following steps: Real-time signal acquisition steps: Continuously acquire multimodal physiological signals of the target limb, wherein the multimodal physiological signals include at least electromyography signals, joint kinematic angle signals and nerve electrical signals; State assessment and prediction steps: Receive multimodal physiological signals output from the real-time signal acquisition step, calculate real-time human state indicators based on the multimodal physiological signals, and simultaneously integrate historical state data to generate future state trend prediction information through a time series prediction model; Dynamic parameter optimization step: Receive the real-time human state indicators and state trend prediction information output from the state assessment and prediction step, and solve the optimal combination of stimulation parameters of the neuromuscular stimulator in real time through an embedded optimization algorithm with minimizing the objective function as the core. The combination of stimulation parameters includes stimulation pulse intensity, frequency sequence and waveform duty cycle. Stimulation execution and verification steps: Receive the optimal combination of stimulation parameters output from the dynamic parameter optimization step, load and execute the optimal combination of stimulation parameters, and simultaneously collect the immediate physiological feedback signal after stimulation execution, and verify the effectiveness of the stimulation based on the immediate physiological feedback signal. The adaptive feedback closed-loop process involves receiving the immediate physiological feedback signal output from the stimulus execution and verification step, while simultaneously establishing data interaction with the real-time signal acquisition step. Based on the deviation between the immediate physiological feedback signal and the expected target, the model parameters used in the state assessment and prediction step, as well as the objective function weights used in the dynamic parameter optimization step, are dynamically adjusted to form a closed-loop control.

[0050] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0051] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0052] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A muscle stimulation modulation system based on real-time dynamic optimization, characterized in that, include: The real-time signal acquisition module is used to continuously acquire multimodal physiological signals of the target limb. The multimodal physiological signals include at least electromyography signals, joint kinematic angle signals, and nerve electrical signals. The state assessment and prediction module is connected to the real-time signal acquisition module. It is used to calculate real-time human state indicators based on multimodal physiological signals, and integrate historical state data to generate state trend prediction information for future periods through a time series prediction model. The dynamic parameter optimization module, connected to the state assessment and prediction module, is used to solve the optimal combination of stimulation parameters for the neuromuscular stimulator in real time through an embedded optimization algorithm, based on real-time human state indicators and state trend prediction information, with minimizing the objective function as the core. The combination of stimulation parameters includes stimulation pulse intensity, frequency sequence and waveform duty cycle. The stimulus execution and verification module, connected to the dynamic parameter optimization module, is used to load and execute the optimal combination of stimulus parameters and collect real-time physiological feedback signals after stimulus execution to verify the effectiveness of the stimulus. The adaptive feedback closed-loop module, connected to the stimulus execution and verification module and the real-time signal acquisition module, is used to dynamically adjust the model parameters of the state assessment and prediction module and the objective function weights of the dynamic parameter optimization module based on the deviation between the instantaneous physiological feedback signal and the expected target, thus forming a closed-loop control.

2. The muscle stimulation modulation system based on real-time dynamic optimization according to claim 1, characterized in that: The state assessment and prediction module calculates real-time human state indicators using the following formula. : ； in, This represents the rate of energy change of the electromyography signal acquired by the real-time signal acquisition module in the time domain; This represents the current joint kinematic angle vector obtained by the real-time signal acquisition module. With respect to the preset standard motion angle vector The Euclidean distance between them; The muscle fatigue index is calculated based on frequency domain analysis of electromyography signals. The human body status indicators at the previous sampling time; , , and The dynamic weighting coefficients are initially set based on individual user information and are adjusted by the adaptive feedback closed-loop module during system operation, while also satisfying preset constraints.

3. The muscle stimulation modulation system based on real-time dynamic optimization according to claim 2, characterized in that: The dynamic parameter optimization module adopts a constrained model predictive control framework, whose core objective function is... Used to evaluate stimulus parameter vectors (Representing stimulus intensity, frequency, and duty cycle, respectively) in the future prediction time domain Overall effect within; The optimization process satisfies Under hard constraints, online solution enables Minimize the optimal parameter sequence .

4. The muscle stimulation modulation system based on real-time dynamic optimization according to claim 2, characterized in that: The dynamic parameter optimization module embeds a parameter sensitivity analysis unit for online evaluation and updating of dynamic weighting coefficients. , , and .

5. The muscle stimulation modulation system based on real-time dynamic optimization according to claim 1, characterized in that: It also includes a safety monitoring unit integrated into the adaptive feedback closed-loop module, which calculates the stimulus safety index in real time. ; when When the first threshold is exceeded, the system triggers a level one alarm and automatically reduces the stimulation parameters to a safe range; when When the stimulation exceeds a higher second threshold, the system immediately stops stimulating the system.

6. A muscle stimulation modulation system based on real-time dynamic optimization according to claim 5, characterized in that: The safety monitoring unit is also used to calculate the motion abnormality index. To detect abnormal movement patterns, its calculation is based on joint angle vectors acquired by a real-time signal acquisition module. ; when When the preset threshold is exceeded, the safety monitoring unit determines that abnormal movement has occurred and instructs the stimulation execution module to pause the current stimulation program and switch to the antispasmodic low-frequency stimulation mode.

7. A muscle stimulation modulation system based on real-time dynamic optimization according to claim 2, characterized in that: It also includes a user interaction and subjective feedback module, used to receive subjective feeling ratings input by users through their terminal devices. and comfort feedback ; The adaptive feedback closed-loop module utilizes this subjective feeling score. and comfort feedback Real-time human condition indicators After making corrections, a comprehensive state index integrating subjective and objective information is obtained. Verification.

8. A muscle stimulation modulation system based on real-time dynamic optimization according to claim 3, characterized in that, The system connects to multiple user terminals through a cloud-based collaborative learning platform, which performs the following operations: Aggregate local optimization data from each anonymized user, including successful state-parameter mapping relationships and optimization trajectories; Based on aggregated data, a global deep reinforcement learning model is trained periodically. The trained global model parameters are distributed to each local system for initialization or fine-tuning of their local dynamic parameter optimization modules, thereby leveraging the global model to improve performance. Improve prediction accuracy and enable collaborative optimization across users.

9. A method for regulating muscle stimulation based on real-time dynamic optimization, said method being applied to the regulation system according to any one of claims 1-8, characterized in that, Includes the following steps: Real-time signal acquisition steps: Continuously acquire multimodal physiological signals of the target limb, wherein the multimodal physiological signals include at least electromyography signals, joint kinematic angle signals and nerve electrical signals; State assessment and prediction steps: Receive multimodal physiological signals output from the real-time signal acquisition step, calculate real-time human state indicators based on the multimodal physiological signals, and simultaneously integrate historical state data to generate future state trend prediction information through a time series prediction model; Dynamic parameter optimization step: Receive the real-time human state indicators and state trend prediction information output from the state assessment and prediction step, and solve the optimal combination of stimulation parameters of the neuromuscular stimulator in real time through an embedded optimization algorithm with minimizing the objective function as the core. The combination of stimulation parameters includes stimulation pulse intensity, frequency sequence and waveform duty cycle. Stimulation execution and verification steps: Receive the optimal combination of stimulation parameters output from the dynamic parameter optimization step, load and execute the optimal combination of stimulation parameters, and simultaneously collect the immediate physiological feedback signal after stimulation execution, and verify the effectiveness of the stimulation based on the immediate physiological feedback signal. The adaptive feedback closed-loop process involves receiving the immediate physiological feedback signal output from the stimulus execution and verification step, while simultaneously establishing data interaction with the real-time signal acquisition step. Based on the deviation between the immediate physiological feedback signal and the expected target, the model parameters used in the state assessment and prediction step, as well as the objective function weights used in the dynamic parameter optimization step, are dynamically adjusted to form a closed-loop control.

Images