A physical therapy instrument control system and method based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based physiotherapy instrument control system and method, relating to the field of intelligent medical technology. The control system comprises real-time acquisition of electromyographic (EMG) signals, bandpass filtering, and TK energy operator processing to output a muscle activation feature vector; based on the muscle activation feature vector, spatiotemporal features are extracted using a convolutional neural network to construct a physiotherapy parameter combination; a PID controller is used to adjust the physiotherapy parameter combination in real time and output optimized physiotherapy parameters; electromagnetic-thermal coupling simulation is performed on the optimized physiotherapy parameters using finite element analysis to obtain physiotherapy instrument control instructions; and a stochastic gradient descent algorithm is used to update the convolutional neural network weights based on a physiotherapy performance evaluation report to output optimized execution instructions. The present invention optimizes PID parameters using a quantum annealing algorithm and dynamically adjusts stimulation parameters in combination with EMG β-band phase locking to synchronize the treatment parameters with the physiological response.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and in particular to a physical therapy instrument control system and method based on artificial intelligence. Background Art

[0002] The control method of physiotherapy instrument driven by electromyographic signals is an important research direction in the intersection of rehabilitation medicine and biomedical engineering, and has high application value in the postoperative rehabilitation of anorectal diseases, such as the surgical treatment of hemorrhoids, anal fissures and anal fistulas. Most of the methods currently used are to respond to changes in the amplitude of electromyographic signals through preset stimulation patterns. Among them, anorectal-specific equipment usually uses electrodes on the surface of the anal sphincter to collect signals, and uses low-frequency electrical stimulation to relieve pain or medium-frequency electrical stimulation to improve muscle function. Common electromyographic signal-driven physiotherapy instrument control methods usually use linear time-invariant filters to extract signal features, combine lookup tables and empirical formulas to generate treatment parameters, and then use analog PID controllers to achieve closed-loop regulation. At present, existing technologies have realized basic electromyographic feedback therapy functions and have been widely used in muscle rehabilitation, pain management, anorectal postoperative physiotherapy and other fields.

[0003] However, traditional methods lack adaptability in spatiotemporal feature extraction, and static convolution kernels have difficulty capturing the non-stationary characteristics of electromyographic signals. For example, they cannot effectively identify the unique circular contraction pattern of the anal sphincter, which limits the matching accuracy of treatment parameters and dynamic muscle activation states. In addition, the synergy between electromagnetic thermal coupling simulation and real-time control is also insufficient. The finite element analysis currently used is mostly used as an offline verification tool, and fails to form a closed-loop optimization with online parameter control, affecting the dynamic guarantee capability of the treatment safety margin. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a physiotherapy instrument control method based on artificial intelligence to solve the problem of insufficient synergy between dynamic feature extraction of electromyographic signals and real-time optimization of physiotherapy parameters.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for controlling a physiotherapy apparatus based on artificial intelligence, which comprises collecting electromyographic signals in real time, performing bandpass filtering and TK energy operator processing, and outputting a muscle activation feature vector;

[0008] Based on the muscle activation feature vector, the spatiotemporal features are extracted through convolutional neural networks to construct a combination of physical therapy parameters;

[0009] The PID controller is used to adjust the therapy parameter combination in real time and output the optimized therapy parameters;

[0010] Conduct electromagnetic-thermal coupling simulation on the optimized therapy parameters through finite element analysis to obtain control instructions for the therapy device;

[0011] Combined with constant current source circuit and pulse width modulation temperature control, it executes the control instructions of the physiotherapy instrument, monitors the physiotherapy effect in real time, and generates a physiotherapy performance evaluation report;

[0012] According to the physiotherapy performance evaluation report, the stochastic gradient descent algorithm is used to update the convolutional neural network weights and output optimized physiotherapy device control instructions.

[0013] As a preferred solution of the artificial intelligence-based physiotherapy instrument control method of the present invention, wherein: the real-time acquisition of electromyographic signals, and bandpass filtering and TK energy operator processing, and output of muscle activation feature vectors are performed, the specific steps are as follows:

[0014] Real-time acquisition of electromyographic signals, combined with a bioelectric amplifier and Butterworth filter, removes noise and motion artifacts, and outputs pure electromyographic signals;

[0015] Combining Hilbert transform with TK energy operator, the signal envelope is extracted to obtain time-varying energy features, and the statistical features of the sliding window are used to extract the output muscle energy features;

[0016] The dynamic window statistics method is used to obtain the energy integral value, and the digital filter bank is used to construct the spatial feature vector and combine it with feature splicing to output the muscle activation feature vector.

[0017] As a preferred solution of the artificial intelligence-based physiotherapy device control method of the present invention, wherein: based on the muscle activation feature vector, the spatiotemporal features are extracted by a convolutional neural network to construct a physiotherapy parameter combination, the specific steps are as follows:

[0018] Use sliding window to split the window, perform fast Fourier transform on the window signal, and output time series feature data;

[0019] Use dilated causal convolution to construct a spatiotemporal feature map and output the fused spatiotemporal feature vector;

[0020] A dual-branch predictor is used to simultaneously perform physical therapy parameter optimization and safety risk control to construct a physical therapy parameter combination.

[0021] As a preferred solution of the artificial intelligence-based physiotherapy instrument control method of the present invention, wherein: the PID controller is used to adjust the physiotherapy parameter combination in real time and output the optimized physiotherapy parameters. The specific steps are as follows:

[0022] Use quantum annealing algorithm to optimize the initial PID parameters under Lyapunov stability constraints and output preliminary optimized parameters;

[0023] Combine the preliminary optimized parameters with the real-time pure electromyographic signal, dynamically adjust the stimulation parameters according to the electromyographic β-band phase, and output the dynamically adjusted preliminary optimized parameters;

[0024] Incremental trial-and-error optimization and parameter regression are used to perform feature comparison and output the optimized therapy parameters.

[0025] As a preferred solution of the artificial intelligence-based physiotherapy device control method of the present invention, wherein: the electromagnetic thermal coupling simulation is performed on the optimized physiotherapy parameters through finite element analysis to obtain the physiotherapy device control instructions, and the specific steps are as follows:

[0026] Convert the optimized therapy parameters into standard units, use middleware to convert the input format, and output a standardized therapy parameter group;

[0027] Combining the standardized physiotherapy parameter group with the real-time pure electromyographic signal, the spatial electromagnetic field distribution is obtained using Maxwell equations, and the instantaneous electromagnetic field distribution is output;

[0028] Based on the instantaneous electromagnetic field distribution, the finite difference method is used to discretize the tissue area, and the real-time temperature value is corrected in combination with weighted fusion to output the corrected temperature field;

[0029] Based on the modified temperature field, threshold control is used to monitor and dynamically adjust the therapy parameters in real time and output safety control instructions;

[0030] Execute safety constraints on safety control instructions and generate executable instructions, outputting physical therapy device control instructions

[0031] As a preferred solution of the artificial intelligence-based physiotherapy instrument control method of the present invention, wherein: the combination of a constant current source circuit and pulse width modulation temperature control, executing physiotherapy instrument control instructions, and real-time monitoring of physiotherapy effects, and generating a physiotherapy performance evaluation report, the specific steps are as follows:

[0032] Combine the constant current source circuit with the microcontroller to receive and send the control instructions of the physiotherapy device and output the electrical stimulation signal;

[0033] Real-time monitoring of the temperature of the therapy area, dynamic adjustment of therapy power using fuzzy PID controller, and output of safe therapy temperature;

[0034] Collect patient subjective feedback and physiological indicator data, integrate therapy parameters and safe therapy temperature, and output efficacy monitoring data set;

[0035] A weighted statistical algorithm was used to analyze the efficacy monitoring data set and generate a therapeutic performance evaluation report.

[0036] The stochastic gradient descent algorithm is used to update the convolutional neural network weights according to the physiotherapy performance evaluation report, and output the optimized physiotherapy instrument control instructions. The specific steps are as follows:

[0037] Using sliding window statistics and standardization processing, physical therapy parameters and physiological index data are extracted from the physical therapy performance evaluation report, and standardized feature vectors and optimized target parameters are output;

[0038] Apply the gradient descent algorithm to obtain the update direction, perform parameter adjustments within a safe range, and output the optimized therapy parameter combination;

[0039] Utilize structured data encapsulation and convert the optimized physiotherapy parameter combination into device execution instructions to output optimized physiotherapy device control instructions.

[0040] In a second aspect, the present invention provides a physical therapy instrument control system based on artificial intelligence, including a signal processing module, a feature extraction module, a parameter control module, a simulation analysis module, an execution feedback module and an optimization iteration module;

[0041] The signal processing module is used to collect electromyographic signals in real time, perform bandpass filtering and TK energy operator processing, and output muscle activation feature vectors;

[0042] Feature extraction module, which is used to extract spatiotemporal features based on muscle activation feature vectors through convolutional neural networks and construct physical therapy parameter combinations;

[0043] Parameter control module, used to use PID controller to adjust the physical therapy parameter combination in real time and output optimized physical therapy parameters;

[0044] The simulation analysis module is used to perform electromagnetic thermal coupling simulation on the optimized physical therapy parameters through finite element analysis to obtain the control instructions of the physical therapy device;

[0045] The execution feedback module is used to combine the constant current source circuit and pulse width modulation temperature control to execute the control instructions of the physical therapy instrument, monitor the physical therapy effect in real time, and generate a physical therapy performance evaluation report;

[0046] The optimization iteration module is used to update the convolutional neural network weights using the stochastic gradient descent algorithm according to the physiotherapy performance evaluation report and output optimized physiotherapy device control instructions.

[0047] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the artificial intelligence-based physiotherapy device control method as described in the first aspect of the present invention is implemented.

[0048] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based physiotherapy device control method as described in the first aspect of the present invention.

[0049] The beneficial effects of the present invention are as follows: optimizing PID parameters through quantum annealing algorithm, dynamically adjusting stimulation parameters in combination with the phase of electromyographic β-band, so that treatment parameters are synchronized with physiological responses; using sliding window segmentation and fast Fourier transform to extract the temporal characteristics of electromyographic signals, combining with void causal convolution to enhance spatiotemporal correlation, so that feature extraction is more in line with the dynamic muscle activation state. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 The figure is a flow chart of a control method of a physiotherapy instrument based on artificial intelligence.

[0052] Figure 2 Schematic diagram of the control system of the physiotherapy instrument based on artificial intelligence.

[0053] Figure 3 This is the flow chart of electromyographic signal processing.

[0054] Figure 4 Flowchart for physiotherapy parameter optimization. DETAILED DESCRIPTION

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0056] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0058] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a method for controlling a physiotherapy device based on artificial intelligence, comprising the following steps:

[0059] S1. Collect electromyographic signals in real time, perform bandpass filtering and TK energy operator processing, and output muscle activation feature vectors.

[0060] Furthermore, the EMG signal is collected in real time, and combined with a bioelectric amplifier and a Butterworth filter, noise and motion artifacts are removed to output pure EMG signals.

[0061] Specifically, a high-precision bioelectric amplifier and a surface electrode array are used for real-time EMG signal acquisition. The original EMG signal is amplified 1000 times by the high-precision bioelectric amplifier, while suppressing common-mode interference. The amplified differential signal is obtained and digitized using a 16-bit analog-to-digital converter at a sampling rate of 2kHz. The Nyquist criterion is used to meet the sampling requirements of the highest frequency of the EMG signal, and the original multi-channel digital EMG signal is output. A 4th-order Butterworth bandpass filter is used to filter out low-frequency noise and suppress high-frequency noise. The low-frequency noise is filtered out through a 20Hz high-pass filter to eliminate baseline drift (such as low-frequency noise caused by breathing or changes in skin electrode contact). Frequency fluctuations), high-frequency noise suppression uses a 450Hz low-pass filter to remove electromagnetic noise outside the EMG signal range (such as radio frequency interference or equipment switching noise). Bidirectional filtering is used for phase compensation to eliminate the phase delay caused by the filter and align the EMG signal time domain, outputting the EMG signal after bandpass filtering. The accelerometer synchronously collects limb movement signals as the reference input for motion artifacts. The least mean square algorithm is used to dynamically adjust the filter and subtract motion-related components from the EMG signal. The sym4 wavelet basis is used to remove the residual high-frequency noise of the filtered signal, and the wavelet soft threshold denoising is performed with a decomposition level of 5 to obtain a smooth EMG signal and output a pure EMG signal.

[0062] Combining Hilbert transform with TK energy operator, the signal envelope is extracted and the time-varying energy features are obtained. Meanwhile, the statistical features of the sliding window are used to extract the muscle energy features.

[0063] Specifically, a Hilbert transform is performed on the pure EMG signal to obtain the orthogonal components of the pure EMG signal and the analytical signal. The modulus of the analytical signal is the signal envelope, which reflects the trend of the amplitude of the pure EMG signal changing over time. The phase offset of the Hilbert transform is eliminated by group delay compensation, so that the signal envelope is aligned with the original EMG signal in time domain, and a smooth pure EMG signal envelope is output. Combined with the TK energy operator and envelope weighting, the TK energy operator is used for the signal envelope and nonlinear amplification is performed to highlight the transient characteristics of muscle activation and output the time-varying energy characteristics of the pure EMG signal. Based on the time-varying energy characteristics, sliding window division is performed, and the statistical features of the sliding window are extracted to obtain the window energy mean and energy variation coefficient. The energy rising slope and energy peak ratio are obtained by linear fitting, and the characteristic values ​​beyond the physiological range are eliminated. The correlation between the features is checked, the redundant features are removed, and the muscle energy characteristics are output.

[0064] Based on the muscle energy characteristics, the dynamic window statistics method is used to obtain the energy integral value. The digital filter bank is used to construct the spatial feature vector and perform feature splicing to output the muscle activation feature vector.

[0065] Specifically, a dynamic window statistics method was used to divide the window into analysis windows with muscle energy characteristics of 200ms and a step size of 50ms. The energy integral value was obtained in each window to reflect the total work done by muscle activation and the energy fluctuation rate, characterize the stability and rise time of the neural drive signal, measure the time required from baseline to 90% of the peak value and the activation duration, verify the feature reliability through motion sensor data, and output the time domain feature vector; a digital filter group was used to decompose the time-varying energy characteristics of the pure electromyographic signal; a digital filter was used to decompose the time-varying energy characteristics of the pure electromyographic signal into the γ frequency band (30-60Hz), the β frequency band (13-30Hz) and the α frequency band (8-13Hz), and for the γ frequency band, the percentage of energy in the total signal energy was obtained as the voluntary contraction intensity indicator, the peak detection algorithm was used to determine the main frequency position and its offset relative to the center frequency was recorded as the main frequency offset, the bispectral analysis was used to obtain the phase coupling between the γ-β frequency bands, and the harmonic distortion analysis was used to extract the amplitude of the third harmonic component. For the β frequency band, the energy proportion is obtained to reflect the intensity of maintenance activity, the deviation of the main frequency in the range of 13-30Hz is detected, the coherence coefficient with the α frequency band is analyzed as the frequency band coupling degree, and the amplitude ratio of the fundamental wave and the second harmonic is extracted; for the α frequency band, while obtaining the basic energy parameters, the distribution entropy value of the power spectrum density in the 8-13Hz frequency band is monitored to quantify the characteristics of the relaxation state; the energy proportion (3D), main frequency deviation (3D), frequency band coupling (2D for γ-β and β-α) and harmonic components (the fundamental wave and the third harmonic of each frequency band are taken) of the three frequency bands of γ, β and α are calculated. A total of 14 characteristic parameters (6 dimensions in total) form a frequency domain feature vector, which is then normalized to the range before outputting the frequency domain feature vector. The spatial pattern analysis method is used to analyze the activation propagation velocity, synchronization index, spatial entropy value, and dominant conduction direction, and the spatial error is corrected through impedance distribution to output the spatial feature vector. Feature splicing is used to merge the time domain feature vector, frequency domain feature vector, and spatial feature vector for normalization. Logarithmic compression is applied to the time domain feature vector, range normalization is applied to the frequency domain feature vector, and Z value standardization is applied to the spatial feature vector to output the muscle activation feature vector.

[0066] It should be noted that the neural driving signal refers to the electrical activity of motor neurons inferred from the electromyographic signal.

[0067] Better, through high-precision signal acquisition and dynamic feature extraction, the accuracy and real-time performance of electromyographic signal processing are improved, and the robustness of noise suppression and motion artifact elimination is enhanced; through feature fusion and adaptive optimization, the comprehensiveness of muscle activation state representation is improved, and the matching of time-frequency domain features and physiological responses is enhanced; through standardization and dynamic window analysis, the compatibility of feature vectors is improved and the reliability of parameter optimization is enhanced.

[0068] S2. Based on the muscle activation feature vector, the spatiotemporal features are extracted through the convolutional neural network to construct the physical therapy parameter combination.

[0069] Furthermore, Z-score normalization is used to eliminate dimensional differences based on muscle activation feature vectors, and sliding window segmentation is adopted and fast Fourier transform is performed on the window signal to output time series feature data.

[0070] Specifically, Z-score normalization is used to eliminate the dimensional differences between different channels, and missing value interpolation and adjacent window mean filling are adopted to output the standardized feature vector; a dynamic adjustment strategy of muscle energy features is used, such as 100ms for the high energy segment and 300ms for the low energy segment, with a default value of 200ms and an overlap rate of 50%; the standardized feature vector is slidingly intercepted according to the window length and step size to generate n windows, and the mean, variance and peak-to-peak value of the features in each window are obtained. The window signal is then fast Fourier transformed to extract the energy proportion of the 1-500Hz frequency band. The aggregated features of the K windows are arranged in chronological order to form a time series feature matrix, and the time series feature data is output.

[0071] It should be noted that the peak-to-peak value refers to the maximum fluctuation range of the time-varying energy characteristics within the sliding window, reflecting the fluctuation range of the instantaneous intensity of muscle activation.

[0072] The time series feature data is reorganized, and the spatiotemporal feature map is constructed using void causal convolution, and the fused spatiotemporal feature vector is output.

[0073] Specifically, the three-dimensional data of continuous time steps, time domain, frequency domain, inter-channel correlation and multi-electrode channels in the time series feature data are reorganized and organized into a space-time cube to retain the electrode topological relationship; a convolution kernel with interval sampling capability is used, and the initial convolution analysis is performed on three adjacent time points. The next layer of convolution analysis spans five time points to gradually expand the analysis range. The convolution kernel is only allowed to access the current and historical time step data, and future data is not used to ensure real-time performance. The time step feature set is obtained and an enhanced space-time feature map is constructed; each time step feature output by the convolution kernel is converted into three sets of vectors: query vector, key vector and value vector through linear transformation and spatial attention. According to the spatial position of the electrode, the importance weights between channels are automatically obtained. For example, the signal correlation of patch electrodes 1 and 3 obtains a higher weight due to their close anatomical positions. A learnable time series tag is added to each time step, and the fused space-time feature vector is output.

[0074] It should be noted that the spatial position of the electrode can be directly obtained through the physical position of the electrode sheet.

[0075] The fused spatiotemporal feature vectors are preprocessed, and a dual-branch predictor is used to simultaneously perform physiotherapy parameter optimization and safety risk control to construct a physiotherapy parameter combination.

[0076] Specifically, based on the fused spatiotemporal feature vector, Z-score normalization is performed to eliminate dimensional differences, and the time segment is divided into time segments at fixed time intervals (such as 50ms) to adapt to real-time control and output the preprocessed spatiotemporal feature vector; a dual-branch predictor is used, including a physical therapy parameter prediction branch and a safety constraint evaluation branch. The physical therapy parameter prediction branch inputs the preprocessed spatiotemporal feature vector into a three-layer fully connected neural network, using 256 nodes as the first layer, 128 nodes as the second layer, and 64 nodes as the third layer. Each layer uses the ReLU activation function for processing, and the output includes three indicators: stimulation frequency, current intensity, and pulse width. The current intensity is constrained to be within the range of 5 to 100 mA through the activation function, and the pulse width is limited to between 50 and 300 microseconds, and a preliminary physical therapy parameter combination is output; the safety constraint evaluation branch uses the same The system uses the characteristic vector of the patient's heart as input, but uses a different network structure for processing, including a two-layer fully connected network and a long short-term memory network, which is used to analyze the temporal risk characteristics and output a safety probability score between 0 and 1. It also gives a specific risk type judgment and compares and verifies it with the preset clinical rules. The risk rules include various safety restrictions. For example, when the energy of the electromyographic signal in a specific frequency band is detected to be too high, the maximum stimulation frequency is automatically limited. If any parameter is found to violate the safety rules, it is corrected and the corresponding risk type is marked. Dynamic weighting is used to mix the original parameters and the default safety parameters according to the safety probability score (for example, when the safety score is 0.7, 70% uses the predicted value and 30% uses the default value). The physical therapy parameters are forced not to exceed the physiological safety boundary, such as the current is less than or equal to 100mA, and the parameters are fine-tuned based on real-time physiological indicators to construct a physical therapy parameter combination.

[0077] It should be noted that safety constraints refer to a mechanism that dynamically limits and assesses the risks of physical therapy parameters through physiological safety boundaries and clinical rules, so that the physical therapy process is in line with the physiological tolerance range of the human body and reduces tissue damage, increased pain or other adverse reactions caused by inappropriate parameters; physiological indicators include skin impedance, local temperature and electromyographic signal characteristics; preset clinical rules include physiological safety boundaries: current intensity, pulse width and temperature control; signal feature triggering rules: electromyographic signal abnormalities and impedance monitoring; patient status adaptation rules: subjective feedback (pain level score, comfort level selection and local sensation description) and physiological indicator blood oxygen saturation value; risk rules include electrical safety rules, physiological response rules and patient status.

[0078] The better one is to improve the resolution of the temporal characteristics of the electromyographic signal and enhance the adaptability of non-stationary signal processing through sliding window segmentation and fast Fourier transform; through void causal convolution and spatiotemporal feature fusion, improve the ability to capture the correlation of multiple electrode channels and enhance the accuracy of characterization of dynamic muscle activation state; through dual-branch predictor and safety constraint evaluation, improve the safety of physical therapy parameter generation and enhance the matching of physical therapy parameters with physiological needs.

[0079] S3. Use a PID controller to adjust the therapy parameter combination in real time and output the optimized therapy parameters.

[0080] Furthermore, the quantum annealing algorithm is used to optimize the combination of therapy parameters under Lyapunov stability constraints and output preliminary optimized parameters.

[0081] Specifically, by combining parameter range setting and quantum annealing hardware initialization, by defining a reasonable search space for PID parameters and connecting the quantum processor, the initialized parameter space and the ready quantum computing environment are output; using quantum bit encoding and Hamiltonian construction, the continuous PID parameters are converted into discrete quantum optimization, an 8-bit binary encoding scheme is used to encode each PID parameter, and a composite cost function including the time multiplied absolute error integral criterion and the Lyapunov function is defined, and the PID parameter candidate set represented in the form of quantum bits is output; using Lyapunov stability analysis combined with the PID parameter candidate set, the quadratic Lyapunov function and its derivatives are calculated in real time to eliminate parameter combinations that do not meet the stability conditions, and output the PID parameter subset that passes the stability test; using the parallel search of the quantum annealing algorithm, 512 sets of parameter combinations are synchronously evaluated within the set 20ms annealing time window, breaking through the local optimal limit through the quantum tunneling effect, selecting the parameters that minimize the composite cost function, and outputting the preliminary optimized parameters.

[0082] It should be noted that the formula of the composite cost function is: ;

[0083] in, represents a composite cost function that comprehensively measures control error and stability; It is a time-weighted error integral that penalizes persistent control errors (such as current and temperature deviations). ; Represents the Lyapunov function, which ensures stability and prevents loss of control due to parameter adjustment. , The error weight coefficient controls the contribution weight of the error term to adjust the response speed. It is adjusted according to the efficacy evaluation report and can be set to a default value of 0.7. Represents the stability weight coefficient. The contribution weight of the stability term prevents overshoot. It automatically increases when an abnormal EMG signal is detected. The default value can be set to 0.3.

[0084] Combining the preliminary optimized parameters with the real-time pure electromyographic signal, the stimulation parameters are dynamically adjusted according to the electromyographic β-band phase, and the dynamically adjusted preliminary optimized parameters are output.

[0085] Specifically, the preliminary optimization parameters are converted into physical therapy instructions, and digital filtering technology is used to extract the 13-30Hz characteristic frequency band in the real-time pure electromyographic signal. Signal envelope detection is used to identify the neural discharge characteristics, and the β-band characteristic signal reflecting the muscle activation state is output; the electromyographic signal fluctuation period is monitored in real time, and time domain feature matching is used to dynamically adjust the stimulation timing according to the neural discharge characteristics of the rising or falling edge of the waveform. The synchronization of stimulation and neural activity is maintained through delay compensation, and a stimulation timing plan synchronized with the muscle contraction rhythm is output; the upper and lower safety limits of the stimulation intensity are set, such as 10-50mA, the output intensity is automatically adjusted according to the change of signal amplitude, and the skin impedance change is monitored in real time. Protection is activated in case of abnormality, and the preliminary optimization parameters of dynamic adjustment are output.

[0086] It should be noted that abnormal changes in skin impedance refer to changes in the skin's resistance to electric current that are inconsistent with normal physiological laws or expected ranges, including sudden changes in impedance values, imbalanced polarity responses, abnormal frequency responses, left-right asymmetry, and violations of safety thresholds. Impedance values ​​below 300Ω pose a risk of electric shock, and values ​​above 5000Ω result in electrode contact failure, which are all considered abnormal.

[0087] Based on the dynamically adjusted preliminary optimization parameters, feature comparison is performed using incremental trial-and-error optimization and parameter rollback, and the parameters are corrected in combination with skin impedance and muscle tremor analysis to output the optimized therapy parameters.

[0088] Specifically, the bioadaptive parameters of the physiotherapy equipment are executed and the waveform of the electromyographic signal during the physiotherapy process is continuously compared with the effective waveform of the physiotherapy, and the characteristic differences such as signal amplitude and fluctuation period are obtained. Parameter adjustment suggestions are generated according to the degree of difference, and the recommended adjustment direction is output. Incremental trial and error optimization is used. Based on the current parameters, a small range of ±5% parameter trial is performed, and the response change trend such as muscle contraction force and fatigue is monitored in real time. Parameter changes that improve the response are retained, invalid adjustments are revoked, and an effective parameter fine-tuning plan is output; indicators such as skin impedance and muscle tremor are tracked simultaneously. When any indicator exceeds the normal fluctuation range, the physiotherapy is automatically suspended, and the safe state is restored through the parameter fallback mechanism. The current physiotherapy data is compared with historical successful cases, and the effective parameter combination in similar cases is adopted to output the optimized physiotherapy parameters.

[0089] It should be noted that the definition standard of the normal fluctuation range is: skin impedance: the baseline range is 800-3000Ω, the variation during physical therapy is less than or equal to 25%, a sudden drop of more than 30% may be due to electrode detachment or skin damage, and a continuous increase of more than 40% may be due to poor contact or too thick stratum corneum; local temperature: the skin surface temperature in the physical therapy area should be 32-37℃, the temperature rise of a single physical therapy is less than or equal to 3℃, the temperature rise per minute is less than or equal to 0.5℃, the absolute temperature is greater than 40℃, the device will be shut down immediately, and the local temperature difference is greater than 2℃ / cm², which will trigger an alarm; muscle tremor: 50-500μV in the resting state, the normal physical therapy response increase is less than or equal to 300%, and abnormal high-frequency tremor greater than 50Hz accounts for more than 15%.

[0090] The better one is to improve the global convergence of PID parameter optimization and enhance stability through quantum annealing algorithm and Lyapunov stability constraint; improve the synergy between stimulation parameters and physiological rhythms and enhance the bioadaptability of physical therapy programs through β-band synchronous adjustment; improve the accuracy of dynamic adjustment of physical therapy parameters and enhance safety fault tolerance under abnormal conditions through incremental trial and error optimization and parameter fallback mechanism.

[0091] S4. Perform electromagnetic thermal coupling simulation on the optimized therapy parameters through finite element analysis to obtain control instructions of the therapy device.

[0092] Furthermore, the optimized physiotherapy parameters are converted into standard units, and the middleware is used to convert the input format and output a standardized physiotherapy parameter group.

[0093] Specifically, regular expression matching is used to identify variables from the optimized physical therapy parameters and convert them into standard units, such as current is converted to amperes, frequency is converted to hertz, and duty cycle is converted to decimals, and a parameter set with unified dimension is output; Z-score standardization is suitable to eliminate dimensional differences between parameters, linear transformation is performed according to the normal fluctuation range, and frequency and duty cycle are standardized and normalized to output dimensionless parameter vectors; tensor filling is used to construct spatiotemporal correlation data, and the dimensionless parameter vectors are arranged into a time series matrix according to the time window. Interpolation method is used to supplement missing values, and a two-dimensional parameter matrix in a standard format is constructed; middleware conversion is used to convert the two-dimensional parameter matrix in the standard format into an input format supported by finite element analysis, the boundary conditions of the electrode contact impedance are embedded, and a standardized physical therapy parameter group is output.

[0094] Combining the standardized physiotherapy parameter group with the real-time pure electromyographic signal, the spatial electromagnetic field distribution is obtained using Maxwell's equations, and the instantaneous electromagnetic field distribution is output.

[0095] Specifically, structured data extraction and physical quantity conversion are used to directly analyze the standardized physical therapy parameter group and the pure electromyographic signal amplitude, convert the current intensity into the electrode surface current density, adjust the tissue conductivity according to the electromyographic signal amplitude, obtain the boundary conditions for electromagnetic field calculation, use Maxwell's equations in combination with the finite difference method to directly solve, output the three-dimensional electric field distribution and magnetic field distribution, set the electrode surface current density, and set the remaining boundaries as insulation or impedance matching conditions to obtain the maximum electric field intensity and maximum magnetic field intensity, use the local gradient method to identify the field strength spatial area and detect the field strength concentration area, identify the field strength spatial area, and obtain the field strength distribution hot spot map in the area where the rate of change suddenly changes, and output the instantaneous electromagnetic field distribution.

[0096] Based on the instantaneous electromagnetic field distribution, the finite difference method is used to discretize the tissue area, and the real-time temperature value is corrected in combination with weighted fusion to output the corrected temperature field.

[0097] Specifically, based on the instantaneous electromagnetic field distribution, the heat production per unit volume is obtained according to the electric field strength and tissue conductivity (dynamically adjusted by the amplitude of the electromyographic signal), and the three-dimensional heat source distribution is output. The three-dimensional heat source distribution is combined with real-time temperature monitoring data (such as body surface temperature). The finite difference method is used to discretize the tissue area, obtain the temperature change of each grid point, and consider the heat dissipation of blood flow (blood flow carries away 0.5W / m³ of heat) to obtain the temperature distribution, and output the predicted temperature field. Based on the predicted temperature field and real-time temperature monitoring data, the temperature value near the monitoring point is directly corrected by weighted fusion, and the corrected temperature field is output.

[0098] It should be noted that the weight coefficient can be adjusted according to the sensor accuracy and can be constructed as 30% for the predicted temperature field and 70% for the real-time temperature monitoring data; the tissue area refers to the three-dimensional biological tissue space where the physiotherapy device acts, which includes layered structures such as skin, fat and muscle, has electrical conductivity, and needs to be discretized by the finite difference method.

[0099] Based on the corrected temperature field, threshold control is used to monitor and dynamically adjust standardized physiotherapy parameters in real time, and output safety control instructions.

[0100] Specifically, the corrected temperature field data and the instantaneous electromagnetic field distribution data are combined to directly check the extreme values, check the temperature values ​​of all positions in the temperature field data, and record the maximum value. If the maximum temperature exceeds 41 degrees Celsius, mark it as needing processing, check the electric field strength values ​​of all positions in the electromagnetic field distribution data, and record the maximum value. If the maximum electric field strength exceeds 100 volts / meter, mark it as needing processing, obtain the temperature difference between adjacent positions, if the temperature change rate exceeds 3 degrees Celsius / cm, mark it as needing processing, and output a safety status report; combine the safety status report with the current standardized physical therapy parameters (current value, frequency value, duty cycle value), and directly modify the standardized physical therapy parameters according to the abnormal situation in the safety status report. If the temperature is too high and needs to be processed, the new current value is 90% of the current current value, and the new frequency value is 105% of the current frequency value. If the electric field strength is too large and needs to be processed, the new duty cycle value is 85% of the current duty cycle, and each parameter is forced to be within a safe range (such as the current is kept between 5-100 mA), and a safety control instruction is output.

[0101] It should be noted that the safety range refers to the range of parameter values ​​that has been clinically verified and can ensure a balance between efficacy and safety during the operation of physical therapy equipment.

[0102] Execute safety constraints on safety control instructions and generate executable instructions, and output physical therapy device control instructions.

[0103] Specifically, based on the safety control instructions, parameter standardization is used to convert all safety control instruction parameters into relative values ​​in the range of 0-1, and weights are set according to the physical therapy stage. For example, the weights for the acute phase are 50% for current, 30% for frequency, and 20% for duty cycle, and the weights for the recovery phase are 30% for current, 40% for frequency, and 50% for duty cycle, and the fused comprehensive control parameters are output; the fused comprehensive control parameters are combined with the equipment safety specifications to perform electrical safety verification, check whether the current is within the range of 5-100mA, verify whether the frequency is within the safe range of 10-150Hz, confirm that the power does not exceed the rated maximum value, the single duration is less than 30 minutes, and the intermittent time is greater than 5 minutes, and output the parameters that have passed the safety verification; use the medical equipment communication protocol conversion to perform standard format encapsulation and output the safety control instructions.

[0104] It should be noted that the weights set according to the physical therapy stages are based on medical rehabilitation principles, clinical evidence data and equipment safety requirements; electrical safety verification is derived from medical device safety specifications, bioelectromagnetic research results and clinical practice evidence.

[0105] Better, through regular expression matching and standardized conversion, the uniformity of physical therapy parameter formats is improved and the compatibility of data interaction is enhanced; through Maxwell equations and finite element analysis, the calculation accuracy of electromagnetic field distribution is improved and the prediction reliability of tissue thermal effects is enhanced; through threshold control and dynamic parameter adjustment, the safety of the physical therapy process is improved and the real-time response capability of abnormal situations is enhanced.

[0106] S5. Combine the constant current source circuit and pulse width modulation temperature control to execute the control instructions of the physiotherapy instrument, monitor the physiotherapy effect in real time, and generate a physiotherapy performance evaluation report.

[0107] Furthermore, the constant current source circuit is combined with the microcontroller to receive and execute the control instructions of the physiotherapy device and output the electrical stimulation signal.

[0108] Specifically, an operational amplifier and a power field-effect transistor are used to form a feedback control loop to monitor the output current in real time and adjust the driving voltage. A precision current sampling resistor is used to detect the actual current, and the current is fed back to the control end through a differential amplifier to output a constant current source circuit. The physiotherapy device control instructions are received by a microcontroller to generate a corresponding PWM signal. An isolated drive circuit is used to enhance safety and prevent high-voltage backflow from damaging the control circuit. A low-pass filter is used to smooth the PWM signal and reduce the interference of high-frequency noise on the nerves. By executing the physiotherapy device control instructions, the microcontroller parses the instructions and sets the PWM duty cycle and frequency parameters. The microcontroller outputs a PWM signal to the field-effect transistor drive circuit to control the current on and off. The current sampling resistor monitors the output in real time and the gate voltage is adjusted by feedback to maintain the target current. If the current exceeds the limit, the output is immediately shut down and an alarm is issued, and an electrical stimulation signal is output.

[0109] It should be noted that the electrical stimulation signal is an adjustable pulse waveform that meets medical standards, such as low-frequency analgesia mode: 1-10Hz for relieving chronic pain, medium-frequency muscle strength training mode: 20-50Hz for muscle activation, and high-frequency nerve inhibition mode: greater than 100Hz for spasm control.

[0110] The temperature of the therapy area is monitored in real time, and the power of the therapy device is dynamically adjusted using a fuzzy PID controller based on the electrical stimulation signal to output a safe therapy temperature.

[0111] Specifically, an infrared temperature sensor is used to measure the skin surface temperature contactlessly, which is collected through an analog-to-digital converter and transmitted to the main control unit of the physiotherapy device. A sliding average filter is used to eliminate instantaneous noise to stabilize the skin surface temperature data. The main control unit compares the real-time temperature with the target temperature, obtains the error, and dynamically adjusts the PWM duty cycle according to the error. If the temperature is too high and the error is less than -1°C, the duty cycle is reduced, for example, by 5%. If the temperature is insufficient and the error is greater than 1°C, the duty cycle is increased, for example, by 5%. The on and off of the heating element of the physiotherapy device is controlled by the field-effect transistor drive circuit. If the temperature exceeds the safety threshold (such as 42°C), the power output is immediately cut off and an audible and visual alarm is triggered. , and control the temperature rise rate to be less than or equal to 1℃ / min to reduce discomfort caused by thermal shock; the infrared sensor and thermocouple synchronously collect the temperature of the therapy area, convert it into a digital converter and input it into the main control unit. Based on the electrical stimulation signal, the fuzzy PID controller outputs the PWM duty cycle adjustment instruction according to the temperature error and change trend. Low temperature state: rapid temperature increase, increase the duty cycle (such as 80%); close to the target: fine-tune the duty cycle (such as 45%-55%), over-temperature risk: forced cooling, reduce the duty cycle to 0% and start the cooling fan, the PWM signal drives the heating element, and continuously monitors the temperature to form a closed-loop control and output a safe therapy temperature.

[0112] It should be noted that the setting of the safety threshold (such as 42°C) is a combination of the critical value of thermal damage to biological tissue, clinical safety standards and equipment reliability requirements.

[0113] Collect patients' subjective feedback and physiological indicator data, integrate physiotherapy parameters and safe physiotherapy temperature, and output efficacy monitoring data set.

[0114] Specifically, during the physical therapy process, the pain level score (0-10 points), comfort level selection (1-5 levels) and local sensation description are obtained through patient prompts, and formatted subjective evaluation data is output; 8-channel differential input is performed through surface electromyography signal acquisition, blood oxygen saturation measurement is performed through infrared light detection, and skin impedance measurement is obtained, and physiological parameter time series data and characteristic parameters (e.g., electromyography intensity, blood oxygen value); physical therapy parameters are obtained and recorded in real time, including electrical stimulation intensity, temperature control parameters and spatial position of the treatment head, and time synchronization is performed to output physical therapy parameter data and equipment operation logs; periodic environmental monitoring is performed using an environmental sensor group, including temperature and humidity measurement, atmospheric pressure detection and electromagnetic field intensity scanning, and noise filtering and abnormal data elimination are performed, and environmental parameter records and electromagnetic environment reports are output; combined with formatted subjective evaluation data, physiological parameter time series data, characteristic parameters, physical therapy parameter database and equipment operation logs, data alignment and standardized format conversion are used to establish a unified time reference, and numerical normalization processing and unified format conversion are performed to output an efficacy monitoring data set.

[0115] It should be noted that the patient's subjective feedback includes pain intensity score, comfort level selection and local sensation description; physiological indicator data include surface electromyography signal, blood oxygen saturation measurement and skin impedance measurement.

[0116] A weighted statistical algorithm was used to analyze the efficacy monitoring data set and generate a therapeutic performance evaluation report.

[0117] Specifically, data cleaning and standardization calculations are performed based on the efficacy monitoring data set, and data validity checks are performed on the efficacy monitoring data set. Blank records are deleted and values ​​outside the reasonable range are corrected, such as setting values ​​of blood oxygen saturation greater than 100% to 100. Standard deviations are calculated for continuous variables, categorical variables are numerically encoded, and normalized data sets and data quality descriptions are output. The coefficient of variation method is used to obtain the coefficient of variation, and the indicator weight distribution table is output in combination with the preset basic weights. The linear weighted sum is used to obtain the original score, which is converted into a percentage system and the individual efficacy score is output. Basic statistical functions are used to select indicators with a change range greater than 15%, and the mean difference before and after physical therapy is obtained, and a report template is used to generate a physical therapy performance evaluation report.

[0118] It should be noted that the coefficient of variation is the ratio of the standard deviation to the mean, and the larger the coefficient of variation, the higher the weight; the original score formula is the standardized value of each indicator multiplied by the corresponding weight and then added together.

[0119] The better one is to improve the stability of electrical stimulation output and enhance the precise controllability of physical therapy dosage through constant current source circuit and PWM temperature control technology; improve the dynamic response speed of temperature regulation and enhance the safety of thermal management of physical therapy area through fuzzy PID control and multi-sensor monitoring; improve the objectivity of efficacy evaluation and enhance the comprehensiveness and scientificity of physical therapy feedback through multimodal data acquisition and weighted statistical algorithm.

[0120] S6. Based on the physiotherapy performance evaluation report, the stochastic gradient descent algorithm is used to update the convolutional neural network weights and output optimized physiotherapy device control instructions.

[0121] Furthermore, sliding window statistics and standardization processing are used to extract physical therapy parameters and physiological indicator data from the physical therapy performance evaluation report, and output standardized feature vectors and optimized target parameters.

[0122] Specifically, based on the physiotherapy performance evaluation report, overlapping time windows are used to apply sliding windows to each data channel (such as current and temperature), extract statistical features such as mean, variance and slope within the window, use clustering algorithms to detect and eliminate outlier windows, and output segmented time series data sets; extract time domain features and frequency domain features from electromyographic signals, align temperature data with physiotherapy parameters by timestamp, calculate cross-correlations (such as the phase difference between temperature lagging behind current), use principal component analysis to compress high-dimensional features into low-dimensional space, output a fused feature matrix, and perform Z-value standardization on the fused feature matrix. Use SHAP value analysis to obtain the contribution of features to efficacy (such as pain relief rate), generate parameter adjustment coefficients (such as current increase), output standardized feature vectors and optimized target parameters.

[0123] It should be noted that each window of the segmented time series data set corresponds to a set of statistical features; each row of the fused feature matrix represents the feature vector of a time window; the standardized feature vector is a numerical matrix with a mean of 0 and a variance of 1; the optimized target parameters are such as current +5mA and frequency -2Hz.

[0124] Combining the standardized feature vector and the optimized target parameters, the gradient descent algorithm is applied to obtain the update direction, and parameter adjustments within a safe range are performed to output the optimized physiotherapy parameter combination.

[0125] Specifically, by combining the standardized eigenvectors and the optimized target parameters, a slight adjustment is made to each therapy parameter, such as the current ±1mA, and the difference in therapeutic effect before and after the adjustment is compared to determine the optimal adjustment direction of each parameter, and the adjustment direction and amplitude of each parameter are output; adjustments are made according to the adjustment direction and amplitude of each parameter, and it is checked whether the adjusted parameters are within the safe range. If they exceed the safe range, they are set to boundary values, and the preliminary optimized parameter combination is output; the preliminary optimized parameter combination is combined with the real-time monitoring data to check whether the temperature is close to the safe range. If the temperature is higher than the safe range, the stimulation intensity is reduced proportionally, and the electromyographic signal is checked for abnormality. The frequency is fine-tuned according to the electromyographic signal, and the optimized therapy parameter combination is output.

[0126] Utilize structured data encapsulation and convert the optimized physiotherapy parameter combination into device execution instructions to output optimized physiotherapy device control instructions.

[0127] Specifically, the optimized physiotherapy parameter combination is standardized, the parameter values ​​are rounded according to the minimum resolution of the device, and all time parameters are uniformly converted into seconds so that the physiotherapy device can directly identify it, and check whether each parameter value is within the allowable range of the physiotherapy device. If it exceeds the range, it is automatically corrected to the closest allowable value, and a set of standardized parameter values ​​is output; according to the type of connected device, the corresponding communication protocol is selected, and each parameter value is filled in the corresponding instruction field according to the format requirements of the selected protocol, and a check code is added at the end of the instruction to output the optimized physiotherapy device control instruction.

[0128] Better, through sliding window statistics and feature standardization processing, the effectiveness of efficacy data extraction is improved and the targetedness of parameter adjustment is enhanced; through gradient descent algorithm and safety range constraints, the adaptability of neural network weight update is improved and the stability of physical therapy parameter optimization is enhanced; through structured data encapsulation and protocol conversion, the standardization of instruction generation is improved and the reliability and compatibility of equipment execution are enhanced.

[0129] This embodiment further provides an artificial intelligence-based physiotherapy instrument control system, comprising: a signal processing module for real-time acquisition of electromyographic signals, performing bandpass filtering and TK energy operator processing, and outputting a muscle activation feature vector; a feature extraction module for extracting spatiotemporal features based on the muscle activation feature vector through a convolutional neural network to construct a physiotherapy parameter combination; a parameter control module for using a PID controller to adjust the physiotherapy parameter combination in real time and output optimized physiotherapy parameters; a simulation analysis module for performing electromagnetic-thermal coupling simulation on the optimized physiotherapy parameters through finite element analysis to obtain physiotherapy instrument control instructions;

[0130] The execution feedback module is used to combine the constant current source circuit and pulse width modulation temperature control to execute the physical therapy device control instructions, monitor the physical therapy effect in real time, and generate a physical therapy performance evaluation report; the optimization iteration module is used to update the convolutional neural network weights using the stochastic gradient descent algorithm according to the physical therapy performance evaluation report, and output the optimized physical therapy device control instructions.

[0131] This embodiment also provides a computer device, which is suitable for the case of an artificial intelligence-based physiotherapy device control method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the artificial intelligence-based physiotherapy device control method proposed in the above embodiment.

[0132] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0133] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the artificial intelligence-based physiotherapy device control method proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0134] In summary, the present invention achieves this by: optimizing PID parameters using a quantum annealing algorithm, dynamically adjusting stimulation parameters based on the myoelectric β-band phase, and synchronizing treatment parameters with physiological responses; extracting temporal features of electromyographic signals using sliding window segmentation and fast Fourier transform, and enhancing spatiotemporal correlations based on dilated causal convolution, making feature extraction more tailored to dynamic muscle activation states.

[0135] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An artificial intelligence physiotherapy instrument control system, characterized by: include, The signal processing module collects electromyographic signals in real time, performs bandpass filtering and TK energy operator processing, and outputs muscle activation feature vectors; The feature extraction module extracts spatiotemporal features based on muscle activation feature vectors through convolutional neural networks and constructs a combination of physical therapy parameters. The specific steps are as follows: Use sliding window to split the window, perform fast Fourier transform on the window signal, and output time series feature data; Use dilated causal convolution to construct a spatiotemporal feature map and output the fused spatiotemporal feature vector; Use a dual-branch predictor to simultaneously optimize therapy parameters and control safety risks, and construct a therapy parameter combination; The parameter control module uses a PID controller to adjust the therapy parameter combination in real time and output the optimized therapy parameters. The specific steps are as follows: Use quantum annealing algorithm to optimize the initial PID parameters under Lyapunov stability constraints and output preliminary optimized parameters; Combine the preliminary optimized parameters with the real-time pure electromyographic signal, dynamically adjust the stimulation parameters according to the electromyographic β-band phase, and output the dynamically adjusted preliminary optimized parameters; Incremental trial-and-error optimization and parameter regression are used to compare features and output optimized therapy parameters. The simulation analysis module performs electromagnetic thermal coupling simulation on the optimized therapy parameters through finite element analysis to obtain the control instructions of the therapy device; The execution feedback module combines the constant current source circuit and pulse width modulation temperature control to execute the control instructions of the physiotherapy instrument, monitor the physiotherapy effect in real time, and generate a physiotherapy performance evaluation report; The optimization iteration module uses the stochastic gradient descent algorithm to update the convolutional neural network weights according to the physiotherapy performance evaluation report and outputs the optimized physiotherapy device control instructions.

2. The artificial intelligence physiotherapy device control system according to claim 1, characterized in that: The real-time acquisition of electromyographic signals, bandpass filtering and TK energy operator processing, and output of muscle activation feature vectors are as follows: Real-time acquisition of electromyographic signals, combined with a bioelectric amplifier and Butterworth filter, removes noise and motion artifacts, and outputs pure electromyographic signals; Combining Hilbert transform with TK energy operator, the signal envelope is extracted to obtain time-varying energy features, and the statistical features of the sliding window are used to extract the output muscle energy features; The dynamic window statistics method is used to obtain the energy integral value, and the digital filter bank is used to construct the spatial feature vector and combine it with feature splicing to output the muscle activation feature vector.

3. The artificial intelligence physiotherapy device control system according to claim 2, characterized in that: The electromagnetic thermal coupling simulation is performed on the optimized physical therapy parameters through finite element analysis to obtain the physical therapy device control instructions. The specific steps are as follows: Convert the optimized therapy parameters into standard units, use middleware to convert the input format, and output a standardized therapy parameter group; Combining the standardized physiotherapy parameter group with the real-time pure electromyographic signal, the spatial electromagnetic field distribution is obtained using Maxwell equations, and the instantaneous electromagnetic field distribution is output; Based on the instantaneous electromagnetic field distribution, the finite difference method is used to discretize the tissue area, and the real-time temperature value is corrected in combination with weighted fusion to output the corrected temperature field; Based on the modified temperature field, threshold control is used to monitor and dynamically adjust the therapy parameters in real time and output safety control instructions; Execute safety constraints on safety control instructions and generate executable instructions, and output physical therapy device control instructions.

4. The artificial intelligence physiotherapy device control system according to claim 3, characterized in that: The combination of constant current source circuit and pulse width modulation temperature control executes the control instructions of the physiotherapy instrument, monitors the physiotherapy effect in real time, and generates a physiotherapy performance evaluation report. The specific steps are as follows: Combine the constant current source circuit with the microcontroller to receive and send the control instructions of the physiotherapy device and output the electrical stimulation signal; Real-time monitoring of the temperature of the therapy area, dynamic adjustment of therapy power using fuzzy PID controller, and output of safe therapy temperature; Collect patient subjective feedback and physiological indicator data, integrate therapy parameters and safe therapy temperature, and output efficacy monitoring data set; A weighted statistical algorithm was used to analyze the efficacy monitoring data set and generate a therapeutic performance evaluation report.

5. The artificial intelligence physiotherapy device control system according to claim 4, characterized in that: The stochastic gradient descent algorithm is used to update the convolutional neural network weights according to the physiotherapy performance evaluation report, and output the optimized physiotherapy instrument control instructions. The specific steps are as follows: Using sliding window statistics and standardization processing, physical therapy parameters and physiological index data are extracted from the physical therapy performance evaluation report, and standardized feature vectors and optimized target parameters are output; Apply the gradient descent algorithm to obtain the update direction, perform parameter adjustments within a safe range, and output the optimized therapy parameter combination; Utilize structured data encapsulation and convert the optimized physiotherapy parameter combination into device execution instructions to output optimized physiotherapy device control instructions.

Images