Traditional Chinese medicine guide operation action adaptation generation method for psychological rehabilitation

Abstract

The invention discloses a traditional Chinese medicine guide operation action adaptation generation method for psychological rehabilitation, and belongs to the technical field of traditional Chinese medicine guide operations. S2, constructing a traditional Chinese medicine guidance knowledge graph; s3, generating a personalized action sequence; s4, reinforcement learning safety fine tuning; and S5, carrying out AR interaction and privacy protection. According to the method, through real-time, dynamic and personalized guide action generation, an intervention scheme can always fit the current psychological and physiological state of an individual, so that the effects of emotion regulation, pressure relief and sleep quality improvement are more remarkable and stable; the closed-loop online fine tuning mechanism enables the system to continuously learn in each breath and each motion range of the user, so that the training safety is ensured, and individualized experience is continuously accumulated.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of Chinese medicine daoyin, and particularly relates to a Chinese medicine daoyin action adaptive generation method for psychological rehabilitation. BACKGROUND

[0002] In the field of the cross of digital therapy and Chinese medicine daoyin, a Chinese medicine daoyin action adaptive generation method for psychological rehabilitation is provided. SUMMARY

[0003] The application aims to provide a Chinese medicine daoyin action adaptive generation method for psychological rehabilitation to solve the problems in the background.

[0004] To achieve the above-mentioned purpose, the application provides the following technical scheme: comprising the following steps: S1, multi-modal assessment of psychological state, obtaining first psychological indicators through psychological assessment questionnaire, obtaining second psychological and physiological indicators through wearable physiological sensing device, and inputting the first indicators and the second indicators into a TCM-Former cross-modal fusion network to output a 64-dimensional psychological state vector; S2, construction of a Chinese medicine daoyin knowledge graph, extracting "action-meridian-emotion-seasonal qi-constitution" entities and relationships in ancient Chinese medical books and modern RCT literature, constructing a knowledge graph, and obtaining an action segment through GraphSAGE; S3, personalized action sequence generation, using a conditional diffusion model DG-Diffusion to sample candidate action sequences in a latent space under the conditions of the psychological state vector, constitution, and emotion polarity, and calculating a comprehensive rehabilitation score through a psychological action effect predictor to output an action sequence; S4, reinforcement learning safety fine-tuning, based on a Shielded-PPO algorithm, taking delta HRV, negative delta HAMA, action completion degree, and safety constraints as reward functions, fine-tuning action amplitude, speed, and breathing rhythm online to form a closed loop; S5, AR interaction and privacy protection, completing action guidance and error correction through AR glasses, bone conduction earphones, and digital human coaches, and protecting user data through a three-layer mechanism of federated learning, differential privacy, and homomorphic encryption; As a further optimization of the technical scheme: the TCM-Former cross-modal fusion network comprises a 12-layer Transformer, a text tower inputting questionnaire text tokens, an InceptionTime+Transformer, a physiological tower inputting 10s sliding window physiological sequences, and a cross-attention layer fusing the text tower and the physiological tower to output a 64-dimensional vector; As a further preferred of the technical solution: the knowledge graph includes 1280 action fragments, 20 meridians, 7 types of emotional labels, 24 solar terms, 9 types of traditional Chinese medicine constitution, 365 acupoints and 46 pathogenesis; As a further preferred of the technical solution: the DG-Diffusion model adopts DiffusionTransformer 8 layers and a latent space of 512x8x8, wherein the conditional control signal includes a safety level mask, a joint angular velocity upper limit and a breathing rhythm range; As a further preferred of the technical solution: the Shielded-PPO algorithm includes a state space of real-time physiology 12 dimensions + attitude error 15 dimensions + residual power 1 dimension, a speed coefficient, a breathing ratio, a joint angle limit, a prompt voice type of action space, the reward function R=0.40xΔHRV+0.30x(-ΔHAMA)+0.15x action completion degree+0.10x satisfaction-0.05x power consumption, and uses a safety shield to prohibit joint torque>60Nxm or heart rate>220-age; As a further preferred of the technical solution: the questionnaire in S1 is selected from the five-state personality scale, SDS / SAS, PCL-5, and SPIRE emotional five-dimensional scale, and the second psycho-physiological index in S1 includes electroencephalogram α / β wave ratio, heart rate variability HRV-LF / HF, skin electricity GSR, 9-axis IMU bone data, and voice MFCC emotional features; As a further preferred of the technical solution: further comprising the following modules: a psychological state evaluation module, a traditional Chinese medicine daoyin knowledge graph module, a personalized action sequence generation module, a reinforcement learning safety fine-tuning module, an AR interaction feedback module, and a cloud federated learning server. As a further preferred of the technical solution: the psychological state evaluation module includes a questionnaire submodule, a physiological sensing submodule and a fusion computing submodule, the traditional Chinese medicine daoyin knowledge graph module stores and dynamically updates the knowledge graph, the personalized action sequence generation module includes a DG-Diffusion inference unit, the reinforcement learning safety fine-tuning module is a Shielded-PPO unit, the AR interaction feedback module includes AR glasses, bone conduction earphones and digital human coaches, and the cloud federated learning server is used for model training and privacy protection data aggregation. As a further preferred of the technical solution: the physiological sensing device includes a 3-lead dry electrode EEG headband, a PPG+EDA integrated bracelet, a 9-axis IMU attitude sticker and a far-field microphone array. As a further preferred of the technical solution: the sampling rate of the 3-lead dry electrode EEG headband is 256Hz, the waterproof level of the PPG+EDA integrated bracelet is IP68, the parameters of the 9-axis IMU posture sticker are ±16g acceleration and ±2000dps gyroscope, and the far-field microphone array supports 3m distance voice emotion recognition.

[0005] Compared with the prior art, the beneficial effects of the present application are: 1、The present application can always adapt to the individual's current psychological and physiological state through real-time, dynamic and personalized guiding action generation, so that the effect of emotion regulation, stress relief and sleep quality improvement is more significant and stable, secondly, the closed-loop online fine-tuning mechanism makes the system continuously learn in each breath and each action amplitude of the user, which not only ensures the safety of training, but also continuously accumulates individual experience.

[0006] 2、The edge computing and multi-layer privacy protection design of the present application enable the whole scheme to run long-term in home, office, school or even offline environment, without the need for professional site or personnel escort, which greatly reduces the use threshold and cost, and significantly improves the accuracy and sustainability of psychological rehabilitation. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 The flowchart of the present application is a method for generating adaptive actions of psychological rehabilitation of traditional Chinese medicine daoyin; Figure 2 The structure of the present application is a method for generating adaptive actions of psychological rehabilitation of traditional Chinese medicine daoyin Figure One ; Figure 3 The structure of the present application is a method for generating adaptive actions of psychological rehabilitation of traditional Chinese medicine daoyin Figure Two . DETAILED DESCRIPTION

[0008] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0009] Example 1: Please refer to Figure 1 - Figure 3 The present application provides a technical solution comprising the following steps: S1, multi-modal assessment of psychological state, obtain the first psychological indicators through psychological assessment questionnaire, obtain the second psychological and physiological indicators through wearable physiological sensing device, and input the first indicators and the second indicators into the TCM-Former cross-modal fusion network to output a 64-dimensional psychological state vector; S2, construction of TCM daoyin knowledge graph, extract "action-meridian-emotion-seasonal qi-constitution" entities and relationships in TCM ancient books and modern RCT literature, construct a knowledge graph, and obtain an action segment using GraphSAGE; S3, personalized action sequence generation, using the psychological state vector, constitution, and emotion polarity as conditions, using the conditional diffusion model DG-Diffusion to sample candidate action sequences in the latent space, and calculating the comprehensive rehabilitation score through the psychological action effect predictor, and outputting the action sequence with the highest score; S4, reinforcement learning safety fine-tuning, based on the Shielded-PPO algorithm, using ΔHRV, -ΔHAMA, action completion degree, and safety constraints as reward functions, fine-tuning the action amplitude, speed, and breathing rhythm online, forming a closed loop; S5, AR interaction and privacy protection, through AR glasses, bone conduction earphones, and digital human coaches to complete action guidance and error correction, and using federated learning, differential privacy, and homomorphic encryption three-layer mechanism to protect user data; In this embodiment, specifically: the TCM-Former cross-modal fusion network includes 12 layers of Transformer, a text tower inputting questionnaire text tokens, InceptionTime+Transformer, a physiological tower inputting 10s sliding window physiological sequences, and a cross-attention layer fusing the text tower and the physiological tower to output a 64-dimensional vector; In this embodiment, specifically: the knowledge graph includes 1280 action segments, 20 meridians, 7 emotion labels, 24 solar terms, 9 TCM constitutions, 365 acupoints, and 46 pathogenesis; In this embodiment, specifically: the DG-Diffusion model uses DiffusionTransformer with 8 layers and a latent space of 512x8x8, and the condition control signal includes safety level mask, joint angular velocity upper limit, and breathing rhythm range; In this embodiment, specifically: the Shielded-PPO algorithm includes a state space of 12 physiological dimensions + 15 attitude error dimensions + 1 remaining power dimension, a speed coefficient, a breathing ratio, a joint angle limit, a prompt voice type, a reward function R=0.40xΔHRV+0.30x(-ΔHAMA)+0.15x action completion degree+0.10x satisfaction-0.05x power consumption, and a safety shield is used to prohibit joint torque>60Nxm or heart rate>220-age; In this embodiment, specifically: the questionnaire in S1 is selected from the Five-Factor Personality Scale, SDS / SAS, PCL-5, and the SPIRE emotion five-dimensional scale, and one or a combination thereof is selected during testing, the second psychophysiological index in S1 includes the EEG alpha / beta wave ratio, heart rate variability HRV-LF / HF, skin electricity GSR, 9-axis IMU bone data, and voice MFCC emotional features, and three or more data need to be tested; In this embodiment, specifically: the following modules are also included: a psychological state assessment module, a traditional Chinese medicine daoyin knowledge graph module, a personalized action sequence generation module, a reinforcement learning safety fine-tuning module, an AR interaction feedback module, and a cloud federated learning server. In this embodiment, specifically: the psychological state assessment module includes a questionnaire submodule, a physiological sensing submodule, and a fusion computing submodule, the traditional Chinese medicine daoyin knowledge graph module stores and dynamically updates the knowledge graph, the personalized action sequence generation module includes a DG-Diffusion inference unit, the reinforcement learning safety fine-tuning module is a Shielded-PPO unit, the AR interaction feedback module includes AR glasses, bone conduction earphones, and a digital human coach, and the cloud federated learning server is used for model training and privacy-protected data aggregation. In this embodiment, specifically: the physiological sensing device includes a 3-lead dry electrode EEG headband, a PPG+EDA integrated bracelet, a 9-axis IMU posture sticker, and a far-field microphone array. In this embodiment, specifically: the sampling rate of the 3-lead dry electrode EEG headband is 256 Hz, the waterproof level of the PPG+EDA integrated bracelet is IP68, the parameters of the 9-axis IMU posture sticker are ±16g acceleration and ±2000dps gyroscope, and the far-field microphone array supports 3m distance voice emotion recognition. When the physiological indicators of the user are monitored to be abnormal and meet any of the following conditions: HRVLF / HF>4, GSR>25us, and EEGtheta / beta>2.5, the system automatically triggers an emergency plan, reduces the action intensity, plays a mindfulness voice, and pushes a remote TCM doctor video intervention.

[0010] Working principle or structural principle: the system first synchronously collects the user's psychological state and physiological signals through questionnaires, EEG, PPG, skin electricity, IMU, microphone array, and fuses them into a 64-dimensional psychological state vector through TCM-Former; then input the 64-dimensional psychological state vector, solar term, constitution, and emotional polarity into the DG-diffusion model to generate a number of candidate action sequences under the constraint of the "action-meridian-emotion-solar term-constitution" knowledge graph; then the psychological action effect predictor calculates the ΔHAMA and ΔHRV comprehensive scores of each sequence, and outputs the optimal sequence; the user performs the action under the guidance of the AR glasses and digital human coach, and the Shielded-PPO adjusts the action amplitude, speed, and breathing rhythm every 30 seconds according to the real-time physiological error and safety threshold, forming a "evaluation-generation-execution-re-evaluation" closed loop; after the training, the local uploads the encrypted gradient through federated learning, the cloud aggregates and updates the global model, and realizes the continuous optimization across users and scenes.

[0011] It is obvious to a person skilled in the art that the application is not limited to the details of the exemplary embodiments described above, but that the application can be implemented in other concrete forms without departing from the spirit or essential characteristics of the application. Therefore, the embodiments should be considered in all respects as illustrative and not restrictive, the scope of the application being defined by the appended claims rather than the above description, and it is therefore intended to embrace all changes and modifications that fall within the meaning and scope of the equivalent elements of the claims. Any reference signs in the claims should not be considered as limiting the claims involved.

[0012] Furthermore, it should be understood that although the present specification is described in terms of embodiments, not every embodiment contains only one independent technical solution, and the description of the specification is only for the sake of clarity, and a person skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for generating a Chinese medicine daoyin action adapted for psychological rehabilitation, characterized in that, The method comprises the following steps: S1, multi-modal psychological state assessment, obtaining first psychological indicators through psychological assessment questionnaire, obtaining second psychological and physiological indicators through wearable physiological sensing device, and inputting the first indicators and the second indicators into the TCM-Former cross-modal fusion network to output a 64-dimensional psychological state vector; S2, construction of a TCM daoyin knowledge graph, extracting "action-meridian-emotion-seasonal qi-constitution" entities and relationships in TCM ancient books and modern RCT literature, constructing a knowledge graph, and obtaining an action segment using GraphSAGE; S3, personalized action sequence generation, using the psychological state vector, constitution, and emotion polarity as conditions to sample candidate action sequences in the latent space using the conditional diffusion model DG-Diffusion, and calculating a comprehensive rehabilitation score through a psychological action effect predictor to output the action sequence; S4, reinforcement learning safety fine-tuning, based on the Shielded-PPO algorithm, using ΔHRV, -ΔHAMA, action completion degree, and safety constraints as reward functions to fine-tune the action amplitude, speed, and breathing rhythm online, forming a closed loop; S5, AR interaction and privacy protection, completing action guidance and error correction through AR glasses, bone conduction earphones, and digital human coaches, and protecting user data using a three-layer mechanism of federated learning, differential privacy, and homomorphic encryption.

2. The method for generating a Qigong movement adaptation for mental rehabilitation according to claim 1, wherein, The TCM-Former cross-modal fusion network comprises a 12-layer Transformer, a text tower inputting questionnaire text tokens, an InceptionTime+Transformer, a physiological tower inputting 10s sliding window physiological sequences, and a cross-attention layer fusing the text tower and the physiological tower to output a 64-dimensional vector.

3. The method of claim 2, wherein the method further comprises: The knowledge graph comprises 1280 action segments, 20 meridians, 7 emotion labels, 24 solar terms, 9 TCM constitutions, 365 acupoints, and 46 pathogenesis.

4. The method of claim 3, wherein the method further comprises: The DG-Diffusion model uses DiffusionTransformer with 8 layers and a latent space of 512x8x8, and the condition control signals include safety level mask, joint angular velocity upper limit, and breathing rate range.

5. The method for generating a Qigong movement adaptation for mental rehabilitation according to claim 4, wherein, The Shielded-PPO algorithm includes a state space of 12 physiological dimensions + 15 attitude error dimensions + 1 remaining power dimension, a speed coefficient, a breathing ratio, a joint angle limit, and a prompt voice type, the reward function R=0.40xΔHRV+0.30x(-ΔHAMA)+0.15x action completion degree+0.10x satisfaction-0.05x power consumption, and uses a safety shield to prohibit joint torque >60Nxm or heart rate >220-age.

6. The method of claim 5, wherein the method further comprises: The questionnaire in S1 is selected from the Five State Personality Scale, SDS / SAS, PCL-5, and SPIRE emotion five-dimensional scale, and the second psychological and physiological indicators in S1 include EEG alpha / beta wave ratio, heart rate variability HRV-LF / HF, skin electricity GSR, 9-axis IMU bone data, and voice MFCC emotion features.

7. The method of claim 6, wherein the method further comprises: The system further comprises the following modules: a psychological state assessment module, a traditional Chinese medicine daoyin knowledge graph module, a personalized action sequence generation module, a reinforcement learning safety fine-tuning module, an AR interaction feedback module, and a cloud federated learning server.

8. The method of claim 7, wherein the method further comprises: The psychological state assessment module comprises a questionnaire sub-module, a physiological sensing sub-module, and a fusion computing sub-module; the traditional Chinese medicine daoyin knowledge graph module stores and dynamically updates a knowledge graph; the personalized action sequence generation module comprises a DG-Diffusion inference unit; the reinforcement learning safety fine-tuning module is a Shielded-PPO unit; the AR interaction feedback module comprises AR glasses, bone conduction earphones, and a digital human coach; and the cloud federated learning server is used for model training and privacy-protected data aggregation.

9. The method of claim 8, wherein the method further comprises: The physiological sensing device comprises a 3-lead dry electrode EEG headband, a PPG+EDA integrated bracelet, a 9-axis IMU posture sticker, and a far-field microphone array.

10. The method of claim 9, wherein the method further comprises: The sampling rate of the 3-lead dry electrode EEG headband is 256 Hz; the waterproof level of the PPG+EDA integrated bracelet is IP68; the parameters of the 9-axis IMU posture sticker are ±16g acceleration and ±2000 dps gyroscope; and the far-field microphone array supports 3m distance voice emotion recognition.

Images