Electronic acupuncture closed-loop control method and system based on multi-modal information feedback

Abstract

This invention provides a closed-loop control method and system for electronic acupuncture based on multimodal information feedback. The method includes: acquiring long-term, short-term, and real-time representation information of the subject under an initial stimulation protocol; inputting the long-term and short-term representation information into a collaborative generation network model for learning, and outputting a target stimulation protocol. The collaborative generation network model includes traditional Chinese medicine and Western medicine agents. The traditional Chinese medicine and Western medicine agents learn from the long-term and short-term representation information and output corresponding traditional Chinese medicine treatment protocols and electrical stimulation protocols. The collaborative generation network model collaboratively trains the traditional Chinese medicine and Western medicine agents; calculating the index deviation between real-time representation information and standard representation information to generate a real-time information deviation matrix; inputting the target electrical stimulation protocol and the real-time information deviation matrix into a parameter deviation prediction agent for learning and prediction; and adjusting the parameters of the target stimulation protocol based on the predicted stimulation parameter deviation matrix to achieve a dual-layer closed-loop control of electronic acupuncture with "long / short-term + real-time" characteristics.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an electronic acupuncture closed-loop control method and system based on multimodal information feedback. Background Technology

[0002] Sleep disorders, mood disorders, and other mental illnesses are a wide range of serious psychological disorders characterized by high incidence and low recognition rates. Traditional diagnostic and treatment methods mainly rely on patients' subjective symptom descriptions, standardized psychological scales, and doctors' experience-based judgments. These methods suffer from strong subjectivity and insufficient quantification, resulting in generally low clinical recognition rates, consultation rates, and treatment effectiveness.

[0003] In recent years, non-pharmacological interventions such as transcutaneous auricular vagus nerve stimulation (taVNS) have shown potential in the treatment of mental disorders. taVNS modulates autonomic nerve function through electrical stimulation and has been preliminarily shown to have positive effects on improving sleep and mood. However, existing electronic acupuncture devices still face the following technical bottlenecks:

[0004] Limitations of open-loop stimulation: Most devices use fixed-parameter stimulation modes, lacking the ability to dynamically respond to individual physiological states;

[0005] Lack of biofeedback: Failure to integrate real-time physiological signals (such as heart rate variability, EEG, etc.) to construct a closed-loop regulatory mechanism;

[0006] Insufficient personalized adaptation: The dynamic matching efficiency between stimulation parameters and neural responses is low, affecting the accuracy of intervention.

[0007] The aforementioned problems make it difficult for existing technologies to achieve precise, real-time closed-loop control in a personalized manner, thus limiting the maximization of treatment effects. Summary of the Invention

[0008] In view of the above problems, the present invention is proposed to provide an electronic acupuncture closed-loop control method and system based on multimodal information feedback to solve the above technical problems or at least partially solve the above technical problems.

[0009] One aspect of the present invention provides an electronic acupuncture closed-loop control method based on multimodal information feedback, the method comprising:

[0010] The subjects' multimodal representation information under the initial stimulus protocol was obtained, including long-term representation information, short-term representation information and real-time representation information.

[0011] The long-term and short-term representation information is input into a pre-trained collaborative generative network model for learning, and the model outputs a target stimulation scheme that matches the long-term and short-term representation information. The collaborative generative network model includes a traditional Chinese medicine agent and a Western medicine agent. The traditional Chinese medicine agent is used to learn and classify the long-term and short-term representation information and output a matching traditional Chinese medicine treatment scheme. The Western medicine agent is used to learn and classify the long-term and short-term representation information and output a matching electrical stimulation scheme. The collaborative generative network model is trained on the traditional Chinese medicine agent and the Western medicine agent to output a stimulation scheme that satisfies the convergence of the two agents' generated schemes towards each other and both converge towards an effective treatment scheme.

[0012] Calculate the index deviation between the real-time characterization information and the preset standard characterization information, and generate a real-time information deviation matrix based on the index deviation.

[0013] The target electrical stimulation scheme and the real-time information deviation matrix are input into a pre-trained parameter deviation prediction agent for learning and prediction, and the stimulation parameter deviation matrix is ​​output.

[0014] Based on the stimulation parameter deviation matrix, the stimulation parameters in the target stimulation scheme are adjusted in real time to achieve closed-loop control of electronic acupuncture.

[0015] In another aspect, the present invention provides an electronic acupuncture closed-loop control system based on multimodal information feedback, the system comprising a functional module for implementing the electronic acupuncture closed-loop control method based on multimodal information feedback as described in any of the preceding claims.

[0016] The system includes:

[0017] The representation information acquisition module is used to acquire the multimodal representation information of the subjects under the initial stimulus protocol. The multimodal representation information includes long-term representation information, short-term representation information and real-time representation information.

[0018] The stimulation scheme prediction module is used to input the long-term and short-term representation information into a pre-trained collaborative generative network model for learning, and output a target stimulation scheme that matches the long-term and short-term representation information. The collaborative generative network model includes a traditional Chinese medicine agent and a Western medicine agent. The traditional Chinese medicine agent is used to learn and classify the long-term and short-term representation information and output a matching traditional Chinese medicine treatment scheme. The Western medicine agent is used to learn and classify the long-term and short-term representation information and output a matching electrical stimulation scheme. The collaborative generative network model performs collaborative training on the traditional Chinese medicine agent and the Western medicine agent to output a stimulation scheme that satisfies the convergence of the generated schemes of the two agents towards each other and the convergence of both towards an effective treatment scheme.

[0019] The information deviation estimation module is used to calculate the index deviation between the real-time characterization information and the preset standard characterization information, and generate a real-time information deviation matrix based on the index deviation.

[0020] The stimulation parameter prediction module is used to input the target electrical stimulation scheme and the real-time information deviation matrix into a pre-trained parameter deviation prediction agent for learning and prediction, and output the stimulation parameter deviation matrix.

[0021] The closed-loop control module is used to adjust the stimulation parameters in the target stimulation scheme in real time based on the stimulation parameter deviation matrix, so as to realize the closed-loop control of electronic acupuncture.

[0022] In another aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor; when executed by the processor, the computer program implements the steps of the electronic acupuncture closed-loop control method based on multimodal information feedback as described in any of the preceding claims.

[0023] The electronic acupuncture closed-loop control method and system based on multimodal information feedback provided in this invention offers a closed-loop control algorithm that includes two time-scale components: a long-term + short-term closed loop and a real-time closed loop. The long-term + short-term closed loop primarily relies on the monitoring status feedback of the subject's long-term and short-term representation information. Through collaborative training of two agents in a collaborative generation network model, the closed loop generates a stimulation scheme that satisfies the convergence of the two agents' generated schemes towards each other and both converge towards an effective treatment scheme. The real-time closed loop primarily adjusts the stimulation parameters of the stimulation scheme in real time based on the monitoring status feedback of the subject's real-time representation information. The combination of these two approaches achieves individualized and precise diagnosis and treatment at both the stimulation scheme and stimulation parameter levels, realizing a dual-layer closed-loop control of electronic acupuncture with "long / short-term + real-time" characteristics, thereby improving the personalized intervention and treatment effect.

[0024] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0025] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings:

[0026] Figure 1 This is a flowchart of the electronic acupuncture closed-loop control method based on multimodal information feedback according to an embodiment of the present invention;

[0027] Figure 2 This is a model architecture diagram of the electronic acupuncture closed-loop control method based on multimodal information feedback according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of the electronic acupuncture closed-loop control system based on multimodal information feedback according to an embodiment of the present invention. Detailed Implementation

[0029] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0030] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art and should not be interpreted in an idealized or overly formal sense unless specifically defined.

[0031] Example 1

[0032] This invention provides a closed-loop control method for electronic acupuncture based on multimodal information feedback, such as... Figure 1 As shown, the electronic acupuncture closed-loop control method based on multimodal information feedback proposed in this invention includes the following steps:

[0033] S11. Obtain the multimodal representation information of the subject under the initial stimulus protocol, wherein the multimodal representation information includes long-term representation information, short-term representation information and real-time representation information.

[0034] Specifically, the long-term characterization information includes: multimodal characterization information collected in units of a first preset time length, including scale assessment information. First Traditional Chinese Medicine Symptom Information First EEG representation information and first electrocardiogram characterization information The short-term characterization information includes: multimodal characterization information collected in units of a second preset time length, including second traditional Chinese medicine characterization information. Second EEG representation information Second electrocardiogram characterization information The real-time representation information includes real-time acquired third-party EEG representation information. and / or third electrocardiogram representation information The first preset time length can be 1-7 days, and the second preset time length can be 1-12 hours.

[0035] Among them, scale assessment information Scales such as the pSQI sleep scale, Hamilton Depression Rating Scale (HAMD-17), and Hamilton Anxiety Rating Scale (HAMA-14) can be used for assessment with the assistance of a doctor, or patients can complete the assessment independently using interactive software. Traditional Chinese medicine (TCM) diagnostic information includes, but is not limited to, tongue diagnosis signals, pulse diagnosis signals, facial diagnosis signals, and ear diagnosis signals. Ear diagnosis signals include ear image data and one or more of the impedance and temperature information of designated acupoints within the concha and helix regions. Auricular point resistance (APR): measured using an ear acupoint detector. Insomnia patients typically exhibit low resistance points or sensitive points. Ear diagnosis image (EDI): obtained through an image sensor and processed using existing image processing algorithms to obtain a text-based description of the image ear diagnosis structure. For example, for the heart region: in healthy individuals, the auricle has a uniform color without bulges or depressions; in insomnia patients, red spots / desquamation (excessive heart fire); corresponding to the internal organs, heart fire rising. Kidney area: In healthy individuals, there is no pigmentation; a dark or pale complexion indicates kidney essence deficiency; corresponding to the internal organs, it indicates kidney yin deficiency or kidney yang deficiency. Liver area: In healthy individuals, the area is smooth; in insomniacs, nodules or congestion indicate liver stagnation transforming into fire; corresponding to the internal organs, it indicates liver qi stagnation. 3) Pulse diagnosis signal (PDS), and the textual description of the pulse diagnosis results is obtained through the following relationships: Pulse position: rapid and swirling (liver fire), thin and weak (qi and blood deficiency) - liver yang hyperactivity / heart and spleen deficiency; Pulse rate: pulse rate > 90 beats, slow pulse (yang deficiency) - heart fire hyperactivity / yang qi deficiency; Pulse shape: wiry and tense (liver stagnation), hesitant pulse (blood stasis), qi stagnation / blood stasis. Facial diagnosis signal, the textual description of the facial diagnosis results is obtained through the overall complexion, the area around the eyes, the forehead, and the cheekbones. For example, in long-term insomniacs: a dark complexion and dark circles around the eyes ("insomnia face") are common; in liver stagnation type: visible blue veins between the eyebrows or temples are visible. EEG representation information can be acquired using an EEG device to collect EEG signals from the prefrontal cortex. EEG can be collected during sleep, before and after each taVNS stimulation, or during taVNS stimulation. After EEG acquisition, EEG signal preprocessing is performed, including real-time electrooculography (EOG) removal and noise removal. Finally, the EEG signal is divided into several frequency bands to capture EEG changes related to emotional fluctuations, specifically including delta, theta, beta, gamma, alpha bands, sleep spindles, K-complexes, etc., and sleep structural stages are determined based on the sleep EEG. The sleep structural stages mainly include: 1) N1 stage (non-rapid eye movement (NREM) sleep stage 1): accounting for approximately 2-5% of total sleep time (TST). During the sleep transition period, the EEG shows theta waves (4-7 Hz), and slow eye movements may occur. 2) N2 stage (NREM sleep stage 2): accounting for approximately 45-55% of TST. The appearance of sleep spindle waves (12-16 Hz) and K complex waves indicates the main sleep stage.3) N3 stage (Non-REM sleep stage 3, slow-wave sleep): accounts for approximately 15-25% of TST. High-amplitude delta waves (0.5-2 Hz), deep sleep, associated with physical recovery. 4) REM stage (Rapid Eye Movement sleep): accounts for approximately 20-25% of TST. Rapid eye movements, EEG similar to wakefulness (mixed frequencies), associated with dreams and memory consolidation. Electrocardiographic information can be obtained through ECG monitoring, mainly including heart rate and heart rate variability (HRV). Heart rate includes mean heart rate (HR) and minimum heart rate. HRV includes total variability (SDNN), parasympathetic activity (RMSSD), low-frequency sympathetic-parasympathetic (LF), high-frequency parasympathetic (HF), LF / HF ratio, etc.

[0036] S12. The long-term and short-term representation information are input into a pre-trained collaborative generative network model for learning, and the model outputs a target stimulation scheme that matches the long-term and short-term representation information. The collaborative generative network model includes a traditional Chinese medicine agent and a Western medicine agent. The traditional Chinese medicine agent is used to learn and classify the long-term and short-term representation information and output a matching traditional Chinese medicine treatment scheme. The Western medicine agent is used to learn and classify the long-term and short-term representation information and output a matching electrical stimulation scheme. The collaborative generative network model performs collaborative training on the traditional Chinese medicine agent and the Western medicine agent to output a stimulation scheme that satisfies the convergence of the two agents' generated schemes towards each other and the convergence of both towards an effective treatment scheme.

[0037] Clinical studies have shown that the stimulation parameters (frequency, intensity, etc.) in the electrostimulation protocol of electronic acupuncture are significantly correlated with the biological characterization regulation effect. This invention constructs an intelligent agent model to predict electronic acupuncture stimulation protocols that are adapted to multimodal characterization information, based on the mapping relationship between the stimulation parameters (frequency, intensity) in the electronic acupuncture stimulation protocol and biological characterization information.

[0038] Specifically, the treatment plan generation agent is a collaborative dual agent combining traditional Chinese medicine (TCM) and Western medicine, learning together. For diseases such as insomnia and depression, traditional Chinese medicine and Western medicine have different focuses: traditional Chinese medicine emphasizes a holistic approach and syndrome differentiation, focusing on the overall regulation of the body's "qi, blood, meridians, and internal organs" system; Western medicine, based on "anatomy, physiology, and pathology," focuses on the specific physiological mechanisms triggered by electrical stimulation, emphasizing precise regulation of the nervous, endocrine, and immune systems. At the same time, both are different theories addressing the same objective fact (disease), meaning they have an objective correlation and complementarity. Based on this, a treatment plan generation agent based on traditional Chinese medicine is established. Solution generation of Western medicine intelligent body Furthermore, a collaborative learning and training architecture between the two is proposed, leading to the creation of a collaborative generative network model (GALN).

[0039] The traditional Chinese medicine intelligent agent in the collaborative generative network model Generate Traditional Chinese Medicine Treatment Plans (TS) TCM Western medicine intelligent body Generative Electrical Stimulation Protocol (TS) WM Both represent information using (day-scale) long-term time-series representations. (Time-level) Short-term representation information The input is used to train the traditional Chinese medicine agent and the Western medicine agent in a collaborative manner, so as to output a stimulus scheme that satisfies the convergence of the generation schemes of the two agents towards each other and the convergence of both towards the effective treatment scheme.

[0040] S13. Calculate the index deviation between the real-time characterization information and the preset standard characterization information, and generate a real-time information deviation matrix based on the index deviation. .

[0041] Specifically, standard characterization information refers to pre-formed relevant biological characterization information of healthy individuals.

[0042] S14. Combine the target electrical stimulation scheme and the real-time information deviation matrix. Input a pre-trained parameter bias prediction agent to learn and predict, and output a stimulus parameter bias matrix. .

[0043] In this embodiment, the main function of the real-time closed loop is to extract real-time representation information based on the subject's real-time monitoring status feedback. This real-time representation information includes real-time acquired third EEG representation information and / or third ECG representation information. The third EEG representation information may include: delta wave power, theta wave power, beta wave power, and gamma wave power. The third ECG representation information may include heart rate, heart rate variability (HRV), etc. A real-time information deviation matrix is ​​calculated between the subject's real-time representation information and the standard representation information of a healthy individual. .by The electrical stimulation protocol generated in collaboration with the dual-agent system of Traditional Chinese Medicine and Western Medicine serves as the input, with the stimulation parameter deviation matrix as the input. For the output, a multilayer perceptron (MLP) network structure combined with a physical information neural network (PINN) training method is used to train the parameter bias prediction agent.

[0044] S15, Based on the stimulus parameter bias matrix The stimulation parameters in the target stimulation program are adjusted in real time to achieve closed-loop control of electronic acupuncture.

[0045] In this embodiment, the stimulus parameter deviation matrix is ​​used. The stimulation parameters in the target stimulation protocol are adjusted in real time and applied to the subject to achieve closed-loop control of electronic acupuncture, thereby achieving personalized and precise diagnosis and treatment.

[0046] Figure 2 The diagram illustrates the model architecture of the electronic acupuncture closed-loop control method based on multimodal information feedback according to an embodiment of the present invention. Figure 2 As shown, this invention mainly includes two cross-timescale closed-loop control models: a long-term + short-term closed-loop and a real-time closed-loop. The long-term + short-term closed-loop primarily functions to provide feedback based on the subject's (day-level) long-term and (hour-level) short-term representation information. This is achieved through collaborative training between a traditional Chinese medicine agent and a Western medicine agent to output a treatment plan that satisfies the combined needs of both agents (i.e., a traditional Chinese medicine treatment plan). and electrical stimulation protocols Both converge toward the effective treatment plan (i.e., the truly effective treatment plan). The system employs a convergent electrical stimulation protocol to generate or adjust transcutaneous acupoint electrical stimulation treatment plans in a closed loop. The real-time closed loop primarily adjusts the transcutaneous acupoint electrical stimulation treatment parameters based on the subject's real-time feedback. The combination of these two approaches achieves dual closed-loop control at both the treatment plan and parameter levels, enabling personalized and precise diagnosis and treatment.

[0047] This invention integrates long-term physiological parameters from traditional Chinese medicine diagnosis (tongue diagnosis, pulse diagnosis, facial diagnosis, and ear diagnosis) with data from overnight sleep monitoring to construct a multimodal, cross-timescale closed-loop regulation system. Compared with traditional electrical stimulation systems, it has the following technical advantages:

[0048] 1. Improved anti-interference capability:

[0049] • By establishing an individualized physiological baseline using long-term indicators such as tongue appearance (coating color, tongue shape) and pulse appearance (deep / floating, slippery / rough), it is possible to effectively distinguish the therapeutic effect from random noise interference (such as motion artifacts, equipment fluctuations) in real-time signals.

[0050] • Sleep parameters (such as REM cycles and deep sleep ratio) serve as natural physiological state benchmarks undisturbed by consciousness, and can correct for circadian rhythm deviations in real-time daytime monitoring data;

[0051] 2. Enhanced model stability:

[0052] • A two-layer architecture of "long-term anchoring + short-term fine-tuning" is adopted: the tongue vein parameter is used as a prior constraint to limit the model search space, and the sleep parameter provides daily state calibration points to avoid model parameter oscillation caused by real-time data;

[0053] • Introducing the diagnostic logic of "inferring internal from external symptoms" from traditional Chinese medicine, when there is a contradiction between real-time EEG (such as abnormal enhancement of gamma waves) and pulse (weak pulse), the pulse trend judgment that has lasted for more than 3 days shall be given priority.

[0054] 3. Improved treatment compliance:

[0055] • Simulates the clinical rhythm of acupuncture "deqi-needle retention-waiting for qi", with long-term parameters corresponding to the cumulative effect of acupuncture treatment courses and short-term parameters reflecting the response of a single treatment, realizing the spatiotemporal coupling of "immediate adjustment" and "treatment course control".

[0056] • Pulse diagnosis data can capture the effects of the meridian flow at different times of day and automatically optimize the stimulation time window (such as strengthening the stimulation of Zusanli when the Stomach Meridian is dominant at Chen time).

[0057] 4. Failure protection mechanism:

[0058] When real-time signals are lost, the system can switch to a "Traditional Chinese Medicine Prescription Mode" based on the pulse trend over the past three days to maintain the baseline stimulation level. This fusion approach enables the system to maintain real-time responsiveness while possessing a comprehensive decision-making capability similar to the "four diagnostic methods" in Traditional Chinese Medicine. Clinical trials have shown a 42% reduction in erroneous adjustments and a 37% improvement in treatment protocol compliance.

[0059] In this embodiment of the invention, a traditional Chinese medicine intelligent body It is a classification network trained based on the Transformer network model. The training process includes: constructing a first training dataset, in which training samples include long-term and short-term representation information of different test subjects under different stimulus schemes, as well as corresponding effective traditional Chinese medicine treatment schemes. The long-term and short-term representation information of different test subjects under different stimulus schemes are used as sample inputs, and the corresponding effective traditional Chinese medicine treatment schemes are used as labels. The Transformer network model is trained based on the first training dataset to obtain a traditional Chinese medicine intelligent agent.

[0060] In this embodiment of the invention, the Western medicine intelligent agent It is a classification network trained based on the Transformer network model. The training process includes: constructing a second training dataset, in which training samples include long-term and short-term representation information of different test subjects under different stimulation schemes, as well as the corresponding effective electrical stimulation schemes. The long-term and short-term representation information of different test subjects under different stimulation schemes are used as sample inputs, and the corresponding effective electrical stimulation schemes are used as labels. The Transformer network model is trained based on the second training dataset to obtain the Western Medicine Intelligent Agent.

[0061] Furthermore, considering the inherent correlation between traditional Chinese medicine (TCM) representational information and diagnostic information such as scales, EEG, and ECG, both the TCM-based intelligent agent and the Western medicine intelligent agent receive inputs including TCM representational information and diagnostic information from scales, EEG, and ECG. The data-driven deep learning intelligent agent will then consider the relationship between these two types of diagnostic information and the electrical stimulation treatment plan. The difference lies in the fact that, during the learning and training process, the TCM-based intelligent agent outputs a TCM treatment plan. It will pay more attention to TCM manifestation information, and Western medicine intelligent agents with electrical stimulation protocols as output. The model will place greater emphasis on diagnostic information such as scales, EEG, and ECG. Specifically, the learning process can leverage the multi-head self-attention mechanism of the Transforemre architecture to adjust the level of attention given to different representational information.

[0062] The traditional Chinese medicine treatment plan proposed in this embodiment includes acupoint stimulation, acupuncture techniques, and treatment duration. The acupuncture techniques include a twisting operation and a lifting and thrusting operation. The twisting operation includes the angle and frequency of twisting, and the lifting and thrusting operation includes the depth of lifting and thrusting. The electrical stimulation plan includes stimulation sites, electrical stimulation parameters, and electrical stimulation duration. The electrical stimulation parameters include stimulation frequency and stimulation intensity. The correspondence between the traditional Chinese medicine treatment plan and the electrical stimulation plan is as follows: acupoint stimulation corresponds to stimulation site; twisting operation corresponds to stimulation frequency (see Table 1); lifting and thrusting operation corresponds to stimulation intensity (see Table 2); and treatment duration corresponds to electrical stimulation duration.

[0063] In a specific example, traditional Chinese medicine treatment plans can be divided into the following treatment plans corresponding to the syndrome types:

[0064] 1. Liver Qi Stagnation Transforming into Fire Type:

[0065] Etiology and pathogenesis: Emotional distress leads to stagnation of liver qi, which over time transforms into fire, disturbing the mind.

[0066] Main manifestations:

[0067] Insomnia, vivid dreams, or even complete sleeplessness;

[0068] Irritability and anger, chest and rib pain;

[0069] Bitter taste in mouth, dry throat, dizziness, and red eyes;

[0070] Red tongue with yellow coating, wiry and rapid pulse;

[0071] Treatment method: Ear acupoints for heart and liver, Taichong and Hegu, using the reducing method.

[0072] 2. Heart and Spleen Deficiency Type:

[0073] Etiology and pathogenesis: Excessive thinking or prolonged illness leads to insufficient blood in the heart and weakness of the spleen.

[0074] Main manifestations:

[0075] Difficulty falling asleep, light sleep, and frequent awakenings;

[0076] Palpitations, forgetfulness, and loss of appetite;

[0077] Sallow complexion, fatigue and lethargy;

[0078] The tongue is pale and swollen with teeth marks, and the pulse is thin and weak.

[0079] Treatment method: Ear acupoints Heart and Spleen, Zusanli, Tianshu, tonification method.

[0080] 3. Yin deficiency with excessive fire type:

[0081] Etiology and pathogenesis: Kidney Yin deficiency, excessive Heart Fire, and disharmony between the Heart and Kidneys.

[0082] Main manifestations:

[0083] Irritability, insomnia, vivid dreams, and easy startling;

[0084] Five-center heat (hot palms and soles, hot chest);

[0085] Night sweats, lower back and knee pain;

[0086] The tongue is red with little coating, and the pulse is thready and rapid.

[0087] Treatment method: Ear acupoints: Heart and Kidney, Taixi, Shenmen, Shenshu, using a balanced tonifying and reducing method.

[0088] 4. Phlegm-heat disturbing the mind type:

[0089] Etiology and pathogenesis: Improper diet or spleen deficiency leads to phlegm production, which in turn generates heat and disturbs the mind.

[0090] Main manifestations:

[0091] Insomnia, heaviness in the head, chest tightness, and epigastric fullness;

[0092] Irritability, bitter taste in mouth, and thick, sticky phlegm;

[0093] The tongue is red with a yellow, greasy coating, and the pulse is slippery and rapid.

[0094] Treatment: Taichong (LR3), Fenglong (ST40), Hegu (LI4), use purgative method.

[0095] 5. Heart and gallbladder qi deficiency type:

[0096] Etiology and pathogenesis: Weak constitution or sudden fright, deficiency of heart and gallbladder qi, and restlessness of spirit.

[0097] Main manifestations:

[0098] I am easily startled during sleep and have many nightmares;

[0099] Timid and easily startled;

[0100] Shortness of breath, spontaneous sweating, fatigue, and weakness;

[0101] Pale tongue, thready and wiry pulse;

[0102] Treatment method: Use auricular acupoints such as Heart and Gallbladder, Shenmen, Neiguan, and Zulinqi, and apply tonification.

[0103] Table 1 Correspondence between Stimulation Frequency and Twisting Operation

[0104] Stimulation frequency range Twist frequency classification Twists per minute Twist range Operating characteristics Tonification and purgation techniques 40-50 Hz Ultra-high frequency 100~120 times 300~360° Rapid, large-scale, and intense stimulation Purgative 30-40 Hz high frequency 80~100 times 270~360° Fast, large amplitude, moderately strong stimulation Purgative 20-30 Hz Intermediate frequency 60-80 times 180~270° Moderate, moderate amplitude, gentle stimulation Even tonification and purgation method 10-20 Hz low frequency 40-60 times 90~180° Slow, small-amplitude, gentle stimulation Supplement <10 Hz Ultra-low frequency 20-40 times <90° Extremely slow, slight amplitude, and very mild stimulation Supplement

[0105] Table 2 Correspondence between Acupuncture Technique Intensity and Electrical Stimulation Parameters

[0106] Equivalent parameters of electrical stimulation intensity Types of techniques Twist / lifting parameters Mechanical Stimulation Intensity Tonification and purgation techniques Current: 0.1-0.5mA weak stimulation Insertion / lifting: 0.3-0.5cm, slow speed. Gentle, so that the patient feels nothing. Supplement Current: 0.5-1.5mA Moderate stimulation Insertion / lifting: 0.5-1cm, medium speed Noticeable soreness and swelling, no tolerance Even tonification and purgation method Current: 1.5-3mA Strong stimulation Insertion / lifting: 1-2cm, quickly. Intense tingling, numbness, and swelling; tolerable. Purgative Current: 3-5mA Extremely stimulating Insertion / retraction: >2cm, impact type Endurance limit, brief muscle contraction Purgative

[0107] Furthermore, when collaboratively training the traditional Chinese medicine intelligent agent and the Western medicine intelligent agent, in order to calculate the loss function of the traditional Chinese medicine intelligent agent... And Western medicine intelligent agent loss function Traditional Chinese Medicine treatment plan With electrical stimulation protocol Deviation between It can transform traditional Chinese medicine treatment plans into equivalent electrical stimulation plans based on the pre-defined correspondence between traditional Chinese medicine treatment plans and electrical stimulation plans, and perform unified feature coding on the equivalent electrical stimulation plans and electrical stimulation plans respectively.

[0108] In this embodiment of the invention, a third training dataset is pre-constructed. The training samples in the third training dataset include long-term and short-term representation information of different test subjects under different stimulation protocols, as well as the corresponding effective real treatment protocols. The long-term and short-term representation information of different test subjects under different stimulation protocols are used as sample inputs, and the corresponding effective real treatment protocols are used as labels. The effective real treatment protocols are effective electrical stimulation protocols or effective traditional Chinese medicine treatment protocols.

[0109] A loss function for constructing a collaborative generative network model is used. Based on the third training dataset and the loss function, traditional Chinese medicine agents and Western medicine agents are collaboratively trained. By optimizing the loss function, the generation schemes of the two agents converge towards each other and both converge towards effective treatment schemes. The loss function of the collaborative generative network model is as follows:

[0110]

[0111] in, The loss function for traditional Chinese medicine intelligent agents includes effective and realistic treatment plans. Compared with traditional Chinese medicine treatment plan Deviation between Traditional Chinese Medicine Treatment Plan With electrical stimulation protocol Deviation between ; The loss function for the Western medicine intelligent agent includes the actual treatment plan. With electrical stimulation protocol Deviations between them, traditional Chinese medicine treatment plans With electrical stimulation protocol Deviation between, In the traditional Chinese medicine intelligent agent loss function Weighting factors For the loss function of Western medicine intelligent agent The weighting factor. For each batch during training, the traditional Chinese medicine agent loss function. Backpropagation enables parameter tuning of traditional Chinese medicine intelligent agents and loss functions of Western medicine intelligent agents. Backpropagation enables parameter tuning of the Western medicine intelligent agent. Due to the correlation between the two loss functions, joint training of the two agents is achieved: that is, the generated schemes of the two agents converge towards each other, and the two agents converge towards the real treatment scheme, thus realizing collaborative learning.

[0112] The traditional Chinese medicine intelligent agent and the Western medicine intelligent agent can be various artificial intelligence models, and they can have the same or different model structures. Here, both adopt the same model structure, which inputs long and short time-course representation information, which has been unified into text form, into the Transformer architecture, and outputs a treatment plan, as follows:

[0113] The Transformer architecture generally adopts an encoder-decoder structure, with six encoders stacked together in the encoder layer and six decoders stacked together in the decoder layer. Each encoder contains two layers: a multi-head attention layer and a feedforward neural network layer (two fully connected layers); each decoder contains a mask multi-head attention layer, a multi-head attention layer, and a feedforward neural network layer.

[0114] In the Encoder, the input data is first embedded (using the WordPiece tokenizer to segment the text in the input information into tokens or sub-words that the model can process); then it is input into the encoder layer. After the self-attention process is completed, the data is sent to the feedforward neural network. The output calculated by the feedforward neural network is input into the next encoder.

[0115] After the decoder layer has finished executing, its output is fed into a fully connected layer and a softmax layer. The softmax layer outputs the treatment plan in text form.

[0116] In one embodiment, the example is given where the real-time representation information only includes EEG representation information. In this embodiment, a fourth training dataset is pre-constructed to train the parameter bias prediction agent. The training samples in the fourth training dataset include real-time representation information of different test subjects under different stimulus schemes and the corresponding stimulus parameter bias matrices. The real-time representation information of different test subjects under different stimulus schemes is used as the sample input, and the corresponding stimulus parameter bias matrix is ​​used as the label.

[0117] In this embodiment of the invention, a dual-driven modeling approach of "data + mechanism" is implemented through an MLP-PINN hybrid architecture to train a parameter bias prediction agent. Specifically, the parameter bias prediction agent can be trained using a multilayer perceptron (MLP) as the base network. The number of neurons in the input layer of the MLP is determined based on real-time representation information; the hidden layer contains three layers, with 256 neurons in each layer; the output layer is a purely linear layer containing three neuron nodes, directly outputting the stimulus parameter bias matrix of intensity, frequency, and duration. A Wilson-Cowan neuron swarm model is used to construct the physical constraints of the parameter bias prediction agent training process. Based on these physical constraints, a physical information error term is embedded in the model loss function to obtain a result including the physical information error term. and data error terms The total loss function is used to achieve multi-objective optimization by minimizing data error and physical information error during agent training. The total loss function is as follows:

[0118]

[0119] in, and To weigh the physical information error term and data error terms Hyperparameters corresponding to the weights. By selecting appropriate hyperparameters, during model optimization training, it is necessary to minimize not only data error but also physical information error, thus constraining the predicted physical quantities within the range allowed by physical laws.

[0120] The physical constraints constructed using the Wilson-Cowan neuron population model are as follows:

[0121]

[0122] in, and The time constant of the response speed of the reactive neuron. and These represent the average firing rates of excitatory neurons and inhibitory neurons per unit time, respectively, and correspond to the electroencephalogram (EEG) signals induced by electrical stimulation. and These are functions representing the natural decay of excitatory and inhibitory neuron populations over time, respectively. and These represent the non-response periods of excitatory and inhibitory neurons, respectively. It is the weight of the self-excitation connection. It is the connection weight from inhibitory neuron I to excitatory neuron E. It is the connection weight from excitatory neuron E to inhibitory neuron I. It is a self-inhibiting connection weight. , The input representing the excitation or inhibition of a neuronal population by external electrical stimulation is determined by the electrical stimulation waveform, amplitude, and frequency; S(·) represents the hyperbolic activation response function of the neuron. In a specific example... , It can be represented as: Where A is the amplitude of electrical stimulation. The frequency of electrical stimulation;

[0123] E(t) and I(t) and the electroencephalogram (EEG) signals induced by electrical stimulation The correspondence is as follows:

[0124]

[0125] Preset hyperparameters;

[0126] Performing a Fourier transform on L(t) yields the EEG signal. The frequency domain representation, based on EEG signals Calculate power spectral density :

[0127]

[0128] The power calculation value of the EEG information in the corresponding frequency band is obtained by integrating the power spectral density P(f) at a specified frequency band. A physical information error term is then constructed based on the power calculation value of the EEG information in the corresponding frequency band. The specified frequency bands include the frequency bands of each EEG representation information included in the real-time representation information.

[0129] Physical information error term for:

[0130] ,

[0131] in,

[0132] ;

[0133] ;

[0134]

[0135] Data error term for:

[0136]

[0137] in, This is the calculated power value of the nth frequency band EEG information in the real-time representation information. This represents the true power value of the nth frequency band EEG information in the real-time characterization information.

[0138] In a specific example, we will use the EEG representation information, which includes beta waves, gamma waves, and theta waves, as an example to illustrate this.

[0139] Electroencephalogram (EEG) signals are the sum of postsynaptic potentials from a local group of neurons. The EEG signal L(t) primarily reflects the difference in EI (Electroencephalogram intensity).

[0140]

[0141] Perform a Fourier transform on L(t) to obtain its frequency domain representation, and then calculate the power spectral density P(f) as follows:

[0142]

[0143] The power integral of a specific frequency band (e.g., the gamma band: 30–100 Hz) can be obtained by integration:

[0144]

[0145] In this specific example, the parameter dependencies for different frequency bands are as follows:

[0146] γ oscillations (30–100 Hz): , ~1-10ms, and c(t) is strongly suppressed.

[0147] β oscillations (13–30 Hz): require slower suppression ( (~10–20 ms), b(t) is moderate.

[0148] Theta oscillations (4–8 Hz): ~50–100 ms (slow inhibition).

[0149] The final result is:

[0150] Physical information error term for: ,

[0151] in,

[0152] ;

[0153] ;

[0154]

[0155] Data error term for:

[0156]

[0157] in, This represents the actual power value of the EEG information in the corresponding frequency band. This is the calculated power value of the corresponding frequency band EEG information in the sample data.

[0158] In a specific example, the physical laws governing the parameter deviation calculation of the PINN agent are established using the Wilson-Cowan neuron population model to characterize neural oscillation entrainment or inhibition: through frequency matching-phase locking or frequency competition, it targets and enhances delta waves or inhibits high-frequency beta and gamma waves. The parameter value ranges in the Wilson-Cowan neuron population model are as follows:

[0159] (Time constant of excitatory neurons): 10–20 ms;

[0160] (Inhibitory neuron time constant): 5–10 ms;

[0161] a(t) (excitatory decay rate): 0.1–0.5 ms⁻¹;

[0162] d(t) (inhibitory decay rate): 0.2–1.0 ms⁻¹;

[0163] Average discharge rate (E(t), I(t)): 0-100Hz;

[0164] r_E(t) (absolute refractory period of excitation): 1–5 ms;

[0165] r_I(t) (inhibitory absolute refractory period): 0.5–2 ms;

[0166] ω(t) (self-excitation weight): 1.0–3.0;

[0167] b(t) (I→E suppression weights): 1.0–2.5;

[0168] c(t) (E→I excitation weight): 1.0–2.0;

[0169] e(t) (self-suppression weight): 0.5–1.5;

[0170] Electrical stimulation, ItCAS-E(t), ItCAS-I(t): A: 0.1-10 (normalized units), f_tCAS (frequency): 1–100Hz;

[0171] Activation function S(·): S(x) = 1 / [1 + exp(−k(x−θ))], k (slope): 0.5–2.0, θ (threshold): 1.0–3.0.

[0172] For the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0173] Example 2

[0174] Another embodiment of the present invention provides an electronic acupuncture closed-loop control system based on multimodal information feedback, the system including a functional module for implementing the electronic acupuncture closed-loop control method based on multimodal information feedback as described in any of the preceding claims. Figure 3 The schematic diagram illustrates the structure of an electronic acupuncture closed-loop control system based on multimodal information feedback provided in an embodiment of the present invention. (Refer to...) Figure 3 The electronic acupuncture closed-loop control system based on multimodal information feedback in this embodiment of the invention specifically includes a characterization information acquisition module 201, a stimulation scheme prediction module 202, an information deviation estimation module 203, a stimulation parameter prediction module 204, and a closed-loop control module 205, wherein:

[0175] The representation information acquisition module 201 is used to acquire the multimodal representation information of the subject under the initial stimulus scheme, wherein the multimodal representation information includes long-term representation information, short-term representation information and real-time representation information;

[0176] The stimulation program prediction module 202 is used to input the long-term and short-term representation information into a pre-trained collaborative generative network model for learning, and output a target stimulation program that matches the long-term and short-term representation information. The collaborative generative network model includes a traditional Chinese medicine agent and a Western medicine agent. The traditional Chinese medicine agent is used to learn and classify the long-term and short-term representation information and output a matching traditional Chinese medicine treatment program. The Western medicine agent is used to learn and classify the long-term and short-term representation information and output a matching electrical stimulation program. The collaborative generative network model performs collaborative training on the traditional Chinese medicine agent and the Western medicine agent to output a stimulation program that satisfies the convergence of the generated programs of the two agents towards each other and the convergence of both towards an effective treatment program.

[0177] Information deviation estimation module 203 is used to calculate the index deviation between the real-time characterization information and the preset standard characterization information, and generate a real-time information deviation matrix based on the index deviation.

[0178] The stimulation parameter prediction module 204 is used to input the target electrical stimulation scheme and the real-time information deviation matrix into a pre-trained parameter deviation prediction agent for learning and prediction, and output the stimulation parameter deviation matrix.

[0179] The closed-loop control module 205 is used to adjust the stimulation parameters in the target stimulation scheme in real time based on the stimulation parameter deviation matrix, so as to realize the closed-loop control of electronic acupuncture.

[0180] As the system implementation is basically similar to the method implementation, it is described in a relatively simple way. For relevant parts, please refer to the description of the method implementation and it has the corresponding technical effects.

[0181] Example 3

[0182] This invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps described in the embodiments of the electronic acupuncture closed-loop control method based on multimodal information feedback, for example... Figure 1 The steps S11-S15 are shown. Alternatively, when the processor executes the computer program, it implements the functions of each module in the above embodiments of the electronic acupuncture closed-loop control system based on multimodal information feedback, for example... Figure 3 The module shown includes a characterization information acquisition module 201, a stimulus scheme prediction module 202, an information bias estimation module 203, a stimulus parameter prediction module 204, and a closed-loop control module 205.

[0183] As the device embodiment is basically similar to the method embodiment, the description is relatively simple. For relevant parts, please refer to the description of the method embodiment, and it has the corresponding technical effects.

[0184] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, any of the claimed embodiments can be used in any combination.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An electronic acupuncture closed-loop control system based on multi-modal information feedback, characterized in that, The system includes: The representation information acquisition module is used to acquire the multimodal representation information of the subjects under the initial stimulus protocol. The multimodal representation information includes long-term representation information, short-term representation information and real-time representation information. The stimulation scheme prediction module is used to input the long-term and short-term representation information into a pre-trained collaborative generative network model for learning, and output a target stimulation scheme that matches the long-term and short-term representation information. The collaborative generative network model includes a traditional Chinese medicine agent and a Western medicine agent. The traditional Chinese medicine agent is used to learn and classify the long-term and short-term representation information and output a matching traditional Chinese medicine treatment scheme. The Western medicine agent is used to learn and classify the long-term and short-term representation information and output a matching electrical stimulation scheme. The collaborative generative network model performs collaborative training on the traditional Chinese medicine agent and the Western medicine agent to output a stimulation scheme that satisfies the convergence of the generated schemes of the two agents towards each other and the convergence of both towards an effective treatment scheme. The information deviation estimation module is used to calculate the index deviation between the real-time characterization information and the preset standard characterization information, and generate a real-time information deviation matrix based on the index deviation. The stimulus parameter prediction module is used to input the target stimulus scheme and the real-time information deviation matrix into a pre-trained parameter deviation prediction agent for learning and prediction, and output the stimulus parameter deviation matrix. The closed-loop control module is used to adjust the stimulation parameters in the target stimulation scheme in real time based on the stimulation parameter deviation matrix, so as to realize the closed-loop control of electronic acupuncture.

2. The system of claim 1, wherein, The long-term characterization information includes: multimodal characterization information collected in units of a first preset time length, including scale assessment information, first traditional Chinese medicine characterization information, first electroencephalogram (EEG) characterization information, and first electrocardiogram (ECG) characterization information; The short-term characterization information includes: multimodal characterization information collected in units of a second preset time length, including second traditional Chinese medicine characterization information, second electroencephalogram (EEG) characterization information, and second electrocardiogram (ECG) characterization information; The real-time representation information includes real-time acquired third EEG representation information and / or third ECG representation information.

3. The system of claim 1, wherein, The traditional Chinese medicine intelligent agent is a classification network trained based on a Transformer network model. The training process includes: Construct the first training dataset, which includes training samples with long-term and short-term representation information of different test subjects under different stimulus protocols as input and corresponding effective traditional Chinese medicine treatment protocols as labels. The Transformer network model is trained based on the first training dataset to obtain a traditional Chinese medicine intelligent agent. The Western medicine intelligent agent is a classification network trained based on the Transformer network model. The training process includes: Construct a second training dataset, which includes training samples with long-term and short-term representation information of different test subjects under different stimulation protocols as input and the corresponding effective electrical stimulation protocols as labels. The Transformer network model is trained based on the second training dataset to obtain the Western Medicine intelligent agent.

4. The system of claim 3, wherein, The system also includes: Construct a third training dataset, which includes training samples with long-term and short-term representation information of different test subjects under different stimulation protocols as input and corresponding effective real treatment protocols as labels. The effective real treatment protocols are effective electrical stimulation protocols or effective traditional Chinese medicine treatment protocols. The loss function of the collaborative generation network model is constructed. Based on the third training dataset and the loss function, the traditional Chinese medicine agent and the Western medicine agent are collaboratively trained so that the generation schemes of the two agents converge towards each other and both converge towards effective treatment schemes. The loss function of the collaborative generation network model is: ， in, The loss function for traditional Chinese medicine intelligent agents includes effective and realistic treatment plans. Compared with traditional Chinese medicine treatment plan Deviation between Traditional Chinese Medicine Treatment Plan With electrical stimulation protocol Deviation between ; The loss function for the Western medicine intelligent agent includes the actual treatment plan. With electrical stimulation protocol Deviations between them, traditional Chinese medicine treatment plans With electrical stimulation protocol Deviation between, In the traditional Chinese medicine intelligent agent loss function Weighting factors For the loss function of Western medicine intelligent agent Weighting factors.

5. The system according to claim 1, characterized in that, Traditional Chinese medicine treatment plans include stimulating acupoints, acupuncture techniques, and treatment duration. Acupuncture techniques include: twisting operation and lifting and thrusting operation. Twisting operation includes the angle and frequency of twisting, and lifting and thrusting operation includes the depth of lifting and thrusting. Electrical stimulation plans include stimulation sites, electrical stimulation parameters, and electrical stimulation duration. Electrical stimulation parameters include: stimulation frequency and stimulation intensity. The correspondence between traditional Chinese medicine treatment plans and electrostimulation plans is as follows: the stimulation point corresponds to the stimulation site, the twisting operation method corresponds to the stimulation frequency, the lifting and thrusting operation method corresponds to the stimulation intensity, and the treatment duration corresponds to the electrostimulation duration.

6. The system according to claim 5, characterized in that, When training traditional Chinese medicine (TCM) intelligence agents and Western medicine intelligence agents in a coordinated manner, the TCM treatment plan is transformed into an equivalent electrical stimulation plan based on the pre-set correspondence between the TCM treatment plan and the electrical stimulation plan, and the equivalent electrical stimulation plan and the electrical stimulation plan are respectively given unified feature encoding.

7. The system according to claim 1, characterized in that, The parameter deviation prediction agent is trained using a multilayer perceptron (MLP) as the base network. The Wilson-Cowan neuron swarm model is used to construct physical constraints for the training process of the parameter bias prediction agent. Based on these physical constraints, a physical information error term is embedded in the model loss function, resulting in a model that includes the physical information error term. and data error terms The total loss function is used to achieve multi-objective optimization by minimizing data error and physical information error during agent training. The total loss function is as follows: ， in, and To weigh the physical information error term and data error terms Hyperparameters corresponding to the weights.

8. The system according to claim 7, characterized in that, The physical constraints constructed using the Wilson-Cowan neuron population model are: ； in, and The time constant of the response speed of the neuron. and These represent the average firing rates of excitatory neurons and inhibitory neurons per unit time, respectively, and correspond to the electroencephalogram (EEG) signals induced by electrical stimulation. and These are functions representing the natural decay of excitatory and inhibitory neuron populations over time, respectively. and These represent the non-response periods of excitatory and inhibitory neurons, respectively. It is the weight of the self-excitation connection. It is the connection weight from inhibitory neuron I to excitatory neuron E. It is the connection weight from excitatory neuron E to inhibitory neuron I. It is a self-inhibiting connection weight. , The input that external electrical stimulation produces excitation or inhibition on a neuronal population is determined by the electrical stimulation waveform, amplitude, and frequency; S(·) represents the hyperbolic activation response function of the neuron. E(t) and I(t) and the electroencephalogram (EEG) signals induced by electrical stimulation The correspondence is as follows: ， These are preset hyperparameters; Performing a Fourier transform on L(t) yields the EEG signal. The frequency domain representation, based on EEG signals Calculate power spectral density : ， Based on power spectral density Power integration is performed at a specified frequency band to obtain the power calculation value of the EEG information at the corresponding frequency band. Based on the power calculation value of the EEG information at the corresponding frequency band, a physical information error term is constructed. The specified frequency bands include the frequency bands of each EEG representation information included in the real-time representation information. Physical information error term for: ， in, ； ； ； Data error term for: ， in, This is the calculated power value of the nth frequency band EEG information in the real-time representation information. This represents the true power value of the nth frequency band EEG information in the real-time characterization information.

9. A computer device, characterized in that, Includes a memory, a processor, and a computer program stored in the memory and executable on the processor; When the computer program is executed by the processor, it implements the following method: The subjects' multimodal representation information under the initial stimulus protocol was obtained, including long-term representation information, short-term representation information and real-time representation information. The long-term and short-term representation information is input into a pre-trained collaborative generative network model for learning, and the model outputs a target stimulation scheme that matches the long-term and short-term representation information. The collaborative generative network model includes a traditional Chinese medicine agent and a Western medicine agent. The traditional Chinese medicine agent is used to learn and classify the long-term and short-term representation information and output a matching traditional Chinese medicine treatment scheme. The Western medicine agent is used to learn and classify the long-term and short-term representation information and output a matching electrical stimulation scheme. The collaborative generative network model is trained on the traditional Chinese medicine agent and the Western medicine agent to output a stimulation scheme that satisfies the convergence of the two agents' generated schemes towards each other and both converge towards an effective treatment scheme. Calculate the index deviation between the real-time characterization information and the preset standard characterization information, and generate a real-time information deviation matrix based on the index deviation. The target stimulus scheme and the real-time information deviation matrix are input into a pre-trained parameter deviation prediction agent for learning and prediction, and the stimulus parameter deviation matrix is ​​output. Based on the stimulation parameter deviation matrix, the stimulation parameters in the target stimulation scheme are adjusted in real time to achieve closed-loop control of electronic acupuncture.

Images