Self-adaptive disinfection and sterilization device of wearable medical equipment and control method

Abstract

The invention discloses a wearable medical equipment self-adaptive disinfection and sterilization device and a control method, relates to the technical field of medical equipment disinfection, and solves the problem that precise, efficient and self-adaptive sterilization cannot be realized in the prior art. Comprising a main control module, a multi-mode disinfection execution module, a multi-source environment perception fusion module, a dynamic bionic protection and diversion structure and an energy collaborative management module. A multi-source environment perception fusion module collects biological pollution and microenvironment data, a master control module operates a digital twinborn model based on federal learning to dynamically generate an optimal disinfection strategy, and multi-mode disinfection execution of an ultraviolet light unit, a microcavity plasma unit and an intelligent photocatalytic coating is controlled. The dynamic bionic structure is used for microenvironment management and energy collaborative management. Perceiving, efficient sterilization and self-adaptive regulation and control on microorganisms on the equipment-skin contact interface are realized, and the sterilization efficiency, the safety and the equipment cruising ability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical equipment, and more particularly to an adaptive disinfection and sterilization device for wearable medical equipment and a control method. BACKGROUND

[0002] Wearable medical equipment plays an important role in continuous health monitoring and chronic disease management. The long-term interface with the skin in a microenvironment of temperature, sweat and sebum can provide the possibility for the colonization of pathogenic microorganisms and the formation of biofilms. Excessive proliferation of microorganisms can cause local or systemic infections, and metabolic products can corrode equipment materials, causing sensor signal degradation, performance degradation and even failure.

[0003] Existing disinfection methods mainly rely on daily timed ultraviolet irradiation or broad-spectrum antibacterial coatings to achieve disinfection. Ultraviolet light sources usually act with fixed power and mode, which is not easy to adapt to uneven irradiance distribution due to curved contact, and is easy to cause disinfection dead angles and local exposure. Antibacterial coatings continuously decay in efficacy under complex biological contamination and physical wear. More importantly, existing methods cannot realize real-time quantitative perception of interface microbial load, and cannot judge the types, activity and biofilm state of pollutants. Disinfection operations are blind and inefficient. At the same time, existing methods lack effective energy efficiency management and user adaptability, and long-term high-intensity disinfection greatly reduces the endurance time of the equipment, and the germicidal factor without intelligent control will stimulate sensitive skin. Even if some improved solutions try to use environmental sensors, their control logic is mostly simple threshold, which is difficult to realize multi-modal perception information fusion and predictive decision-making. Centralized cloud training models to realize personalized strategies have serious security risks of user privacy data leakage.

[0004] Therefore, an intelligent disinfection and sterilization technology that can integrate perception, decision-making and execution, precision, adaptability, low power consumption, privacy protection, etc. is needed to solve the technical problem that current wearable medical equipment is difficult to perform long-term biological safety management. SUMMARY

[0005] In view of the existing wearable medical device disinfection technology has the problems of sensing blind area, decision rigid, single execution, privacy hidden danger and other systematic problems, a wearable medical device adaptive disinfection and sterilization device and control method are disclosed. The device forms a closed-loop intelligent disinfection system of perception-decision-execution-evolution through the combination of the main control module, multi-modal disinfection execution module, multi-source environment perception fusion module, dynamic bionic protection and flow guide structure and energy collaborative management module; the multi-source environment perception fusion module relies on the biological pollution degree sensor to realize real-time sensing of the biological membrane and colony state, the main control module runs the digital twin model based on federated learning, and the multi-modal data fusion multi-source environment data generate the optimal disinfection strategy. The multi-modal disinfection execution module works according to the strategy: the programmable ultraviolet light unit adaptively scans the light field according to the microbial distribution thermal map; the microcavity plasma unit switches the discharge mode through the temperature control super-hydrophobic surface electrode, the intelligent photocatalytic coating realizes the catalytic sterilization effect through the near-infrared excitation quantum dot energy transfer system, the dynamic bionic structure optimizes the energy output through the pupil variable aperture, and the micro-channel unit manages the sweat transport. The energy module collects energy through radio frequency and compensates power through super capacitor. The control method involves the processes of environment perception, strategy generation, collaborative execution, real-time optimization and model updating, uses homomorphic encryption and federated learning for model evolution under privacy protection, and finally realizes the perception, adaptive disinfection and system energy efficiency optimization of the microorganism on the device-skin interface.

[0006] In order to realize the above technical effects, the technical scheme adopted by the present application is as follows: The wearable medical device adaptive disinfection and sterilization device comprises a main control module, a multi-modal disinfection execution module, a multi-source environment perception fusion module, a dynamic bionic protection and flow guide structure and an energy collaborative management module. The multi-modal disinfection execution module and the multi-source environment perception fusion module are embedded conformally in the special-shaped curved surface of the device body in contact with the user's skin. The multi-modal disinfection execution module comprises a programmable ultraviolet light unit, a microcavity plasma unit using a super-hydrophobic surface electrode and an intelligent responsive photocatalytic coating based on a composite material of upconversion nanomaterial and graphene quantum dots. The multi-source environment perception fusion module comprises a biological pollution degree sensor and a microenvironment multi-parameter sensor group. The dynamic bionic protection and flow guide structure comprises a pupil variable aperture and a micro-channel sweat management unit. The energy collaborative management module is electrically connected to the device main power supply and the main control module, and is internally provided with a radio frequency energy collection circuit and a super capacitor. The input end of the master control module is electrically connected to the input end of the multi-source environment perception fusion module, a federal learning digital twin model is run inside the master control module, historical disinfection data, environmental data and user physiological state are used to dynamically generate a disinfection scheme; the output end of the master control module is electrically connected to the input end of the programmable ultraviolet light unit, the input end of the microcavity plasma unit, the input end of the variable aperture, the input end of the microchannel unit and the input end of the energy coordination management module.

[0007] The programmable ultraviolet light unit comprises a micro UV-C LED matrix, an adaptive light field scanning algorithm and a multi-channel constant current driving circuit, the micro UV-C LED matrix is a flip chip structure based on aluminum gallium nitride material, and a nano photonic structure is integrated on the light emitting surface; a microchannel radiator is integrated below the micro UV-C LED matrix; The master control module is configured to execute an adaptive light field scanning algorithm, the adaptive light field scanning algorithm dynamically plans the irradiation dose and scanning path of each pixel sub-region based on the microbial distribution thermal map predicted by the digital twin model.

[0008] The microcavity plasma unit with a supercrown surface electrode is integrated with a micro temperature control element and a high-frequency high-voltage power supply module, the high-frequency high-voltage power supply module has an impedance matching network built-in; the supercrown surface electrode is a multilayer composite structure based on vanadium dioxide phase change material, the electromagnetic response characteristics are adjustable with temperature, and the master control module can control the microcavity plasma unit to switch between capacitive coupling mode and surface wave plasma mode; The region where the intelligent response type photocatalytic coating based on the composite material of upconversion nanomaterial and graphene quantum dot is provided with a low-power near-infrared LED as an excitation light source; the core-shell structure of the upconversion nanoparticles is doped with a rare earth ion pair, and the graphene quantum dots act as an electronic mediator, together constituting a quantum dot fluorescence resonance energy transfer system.

[0009] The biological contamination sensor comprises a micro electrochemical full system unit, a lock-in amplification circuit and a micro interdigital electrode; The surface of the micro interdigital electrode is modified with a molecularly imprinted polymer layer by in-situ polymerization; the sensor is based on the time domain dielectric spectrum analysis principle, analyzes the complex impedance spectrum relaxation time distribution in a specific frequency range, quantitatively analyzes the formation stage of the biofilm and the activity of the bacterial colony, and distinguishes the pollution characteristics of gram-positive bacteria and gram-negative bacteria.

[0010] The pupil type variable aperture is integrated with a micro photosensitive sensor to detect the aperture diameter in real time and form a closed-loop control; the pupil type variable aperture is driven by a double-electrode layer electrostatic MEMS actuator based on a carbon nanotube aerogel electrode; The micro-channel sweat management unit is provided with a sweat biomarker sensor at the inlet, and the micro-channel sweat management unit comprises a bionic microneedle structure integrated with a pH-responsive hydrogel valve, a hydrophilic and hydrophobic patterned surface with a micro-nano composite structure, and a protein adsorption resistant coating coated on the inner wall of the micro-channel, and the sweat is driven to transport directionally by generating Laplace pressure difference.

[0011] The digital twin model based on federated learning is embedded with a meta-reinforcement learning framework; the meta-reinforcement learning framework models the disinfection process as a partially observable Markov decision process, and continuously optimizes its decision strategy through real-time interaction data with the environment perception fusion module; The master module is configured to fine-tune the model locally using user personalized data to generate personalized disinfection strategies, while only uploading the strategy gradient update amount of the model to the cloud after homomorphic encryption for federated aggregation, and the global model parameters aggregated on the cloud are decrypted and verified by secure multi-party computation before being distributed to each local device for model update.

[0012] An adaptive disinfection control method of a wearable medical device, applied to the device of any one of claims 1-6, comprising the following steps: Step one, system initialization and environment perception: The master module initializes each sensor and actuator, and the multi-source environment perception fusion module collects biological pollution degree data, micro-environment parameters and user physiological state data in real time through the micro-interdigital electrode and the lock-in amplification circuit and transmits them to the master module; Step two, digital twin model construction and strategy generation: The master module fuses historical disinfection data and real-time perception data based on the digital twin model of federated learning to construct a current microorganism distribution thermal map, and calculates and generates an optimal disinfection strategy containing disinfection mode selection, energy distribution parameters and execution timing through the embedded meta-reinforcement learning framework; Step three, multi-modal disinfection collaborative execution: The master module synchronously controls the micro-UV-C LED matrix to execute adaptive light field scanning through the multi-channel constant current driving circuit according to the disinfection strategy, and starts the micro-channel heat sink for thermal management at the same time; controls the micro-temperature control element to adjust the temperature of the super-hydrophobic surface electrode, and excites the plasma of the target mode by using the high-frequency high-voltage power supply module and the impedance matching network; turns on the low-power near-infrared LED to excite the intelligent response type photocatalytic coating to produce a catalytic sterilization effect; Step four, dynamic environment management and real-time optimization: The main control module controls the dynamic biomimetic protection and flow guiding structure to perform the following operations: based on the feedback from the micro photosensitive sensor, it adjusts the opening and closing aperture of the pupil-type variable aperture in a closed loop to optimize the disinfection energy output; according to the sweat biomarker sensor signal, it activates the pH-responsive hydrogel valve of the microfluidic sweat management unit, and utilizes the hydrophilic and hydrophobic patterned surface and the anti-protein adsorption coating to synergistically drive the directional transport of sweat and manage the interface microenvironment; based on real-time feedback data, it dynamically adjusts the disinfection strategy through a meta-reinforcement learning framework. Step 5: Energy Co-management and Energy Efficiency Optimization The energy collaborative management module monitors the system's energy consumption status in real time, activates the supercapacitor for instantaneous energy compensation when disinfection peak power demand is reached, and collects ambient electromagnetic energy through the radio frequency energy acquisition circuit to improve the system's endurance. Step Six: Disinfection Efficacy Assessment and Model Update The intelligent evaluation module records the entire disinfection process and obtains sterilization effect indicators. The main control module receives the disinfection effect indicator data recorded by the biocontamination sensor and analyzes and processes the indicator data. The micro electrochemical cleaning unit cleans and regenerates the micro interdigital electrodes and desensitizes the local digital twin model according to the disinfection results. The model update parameters are uploaded to federated learning through homomorphic encryption.

[0013] In step two, the optimal disinfection strategy is generated through the following intelligent decision function, which integrates microbial situational awareness and energy constraints: In formula (1), These are the weighting coefficients; It represents the total number of microorganisms monitored at time t; It is the microbial threshold; It is a risk adjustment factor; It is the risk level of microbial transmission; It is the real-time power required for the disinfection operation at time t; It is the maximum power that the system can provide at time t; It is the power scaling factor; This represents the actual operational capacity of the disinfection system at time t. This is the maximum operating capacity of the disinfection system; These are the disinfection requirements parameters set by the user at time t; It is the baseline value for user needs; k is the temperature sensitivity coefficient; This is the actual temperature of the current environment; This is the temperature at which disinfection is most effective; Introduce a dynamic adjustment mechanism based on real-time environmental data and user behavior: In formula (2), i = 1, 2, 3; is a weight base value; is an adjustment coefficient; is a daily period time constant; the function makes the weight coefficient dynamically adjust according to the microorganism concentration deviation, power budget difference and time period. In step three, the multi-modal disinfection synergy is implemented through the following cross-modal synergy optimization function: ; in formula (3), is the total disinfection effect at the spatial position and time t; are weight coefficients of ultraviolet, plasma and photocatalysis modes respectively; N is the number of ultraviolet disinfection devices; is the radiation intensity of the i-th ultraviolet device; is a spatial position vector; is the spatial position vector of the i-th ultraviolet device; is the spatial diffusion coefficient of ultraviolet radiation; is the correction coefficient of local microorganism concentration to ultraviolet effect; erf is the error function; is the local microorganism concentration at position ; is the microorganism concentration base value; represents the Laplacian operator of plasma potential , t is the time variable; is the characteristic time constant of plasma action; is the absorption coefficient of photocatalytic material; is the ambient light intensity at time t; is the photocatalytic light period. In step six, the model update is implemented through the following privacy-protected federated evolutionary learning function: ; in formula (4), is the updated local model parameter; is the local model parameter before updating; is the learning rate; represents the gradient of the local loss function with respect to the parameter ; represents the gradient mask matrix; represents the Hadamard product; is the global consistency weight coefficient; is the global shared model parameter in federated learning; is the KL divergence of the local model and the global model; is the base threshold of the KL divergence; is the individualization weight coefficient; is the individualization objective function; represents the sample size of the local data size; is a data volume threshold; k is a data volume sensitivity coefficient; represents an encryption weight matrix; a personalized objective function is introduced, and is defined as a regularization term based on user preferences and local environment data: ; in formula (5), M is the dimension of the model parameters, is the parameter weight, is the actual value of the jth model parameter, is the parameter value learned from historical data according to user preferences; is a balance coefficient; is the optimal disinfection effect; is the actual disinfection effect; is the disinfection effect scale factor; is the target satisfaction of user demand; is the actual user demand satisfaction; is the user demand scale factor. The application has the following beneficial technical effects: Through the multi-source environment perception fusion module and the federated learning digital twin model, the device-skin interface microbial pollution perception and decision are completed. The multi-modal disinfection execution module completes the irradiation uniformity of traditional single disinfection, large dead angle and low efficiency through programmable ultraviolet light unit adaptive light field scanning, microcavity plasma unit switching and intelligent photocatalytic coating cooperation. The dynamic biomimetic protection and flow guide structure complete the disinfection energy and interface microenvironment management through the pupil type variable aperture and microchannel sweat management unit. The energy synergy management module completes the energy consumption of continuous high-intensity disinfection through radio frequency energy harvesting and super capacitor. The homomorphic encryption federated learning realizes the update of personalized disinfection strategy, protects user privacy, and finally establishes a closed-loop disinfection system of precise perception, intelligent decision, collaborative execution and autonomous evolution, improves the biological safety reliability and endurance of wearable medical devices. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor. Among them: Figure 1 It is a whole structure diagram of the adaptive disinfection and sterilization device and control method of the wearable medical device of the present application; Figure 2 It is a schematic diagram of the programmable ultraviolet light unit of the adaptive disinfection and sterilization device and control method of the wearable medical device of the present application; Figure 3Schematic diagram of microcavity plasma unit of adaptive disinfection and sterilization device and control method for wearable medical equipment of the present application; Figure 4 Schematic diagram of multi-source environment perception fusion module of adaptive disinfection and sterilization device and control method for wearable medical equipment of the present application; Figure 5 Schematic diagram of dynamic biomimetic protection and flow guiding structure of adaptive disinfection and sterilization device and control method for wearable medical equipment of the present application; Figure 6 Flow chart of adaptive disinfection and sterilization device and control method for wearable medical equipment of the present application; Figure 7 Flow chart of disinfection strategy generation based on digital twinning of adaptive disinfection and sterilization device and control method for wearable medical equipment of the present application; Figure 8 Flow chart of multi-modal disinfection collaborative execution and dynamic management of adaptive disinfection and sterilization device and control method for wearable medical equipment of the present application; Figure 9 Flow chart of energy collaboration and model privacy protection update of adaptive disinfection and sterilization device and control method for wearable medical equipment of the present application; Figure 10 Application scenario diagram of adaptive disinfection and sterilization device and control method for wearable medical equipment of the present application.

[0015] In the figure: main control module 1, multi-modal disinfection execution module 2, multi-source environment perception fusion module 3, dynamic biomimetic protection and flow guiding structure 4, energy collaboration management module 5, programmable ultraviolet light unit 21, microcavity plasma unit 22 using super-hydrophobic surface electrodes, intelligent response type photocatalytic coating 23, biological contamination level sensor 31, microenvironment multi-parameter sensor group 32, pupil type variable aperture 41, microfluidic sweat management unit 42, radio frequency energy harvesting circuit 51, super capacitor 52, micro UV-C LED matrix 211, adaptive light field scanning algorithm 212, multi-channel constant current driving circuit 213, microfluidic heat sink 2111, micro temperature control element 221, high frequency high voltage power supply module 222, impedance matching network 2221, low power near-infrared LED 231, micro electrochemical full-system unit 311, phase-locked amplification circuit 312, micro interdigital electrode 313, micro photosensitive sensor 411, sweat biomarker sensor 421, biomimetic microneedle structure 421, hydrophilic and hydrophobic patterned surface 422, anti-protein adsorption coating 423. DETAILED DESCRIPTION

[0016] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, it should be understood that the embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application; Embodiment 1: An adaptive disinfection device of a wearable medical device, comprising a main control module 1, a multi-modal disinfection execution module 2, a multi-source environment perception fusion module 3, a dynamic bionic protection and flow guide structure 4, and an energy collaborative management module 5. The multi-modal disinfection execution module 2 and the multi-source environment perception fusion module 3 are embedded and conformally integrated in a special-shaped curved surface in contact with the device body and the user's skin. The multi-modal disinfection execution module 2 includes a programmable ultraviolet light unit 21, a microcavity plasma unit 22 using a super-hydrophobic surface electrode, and an intelligent response type photocatalytic coating 23 based on a composite material of upconversion nanomaterials and graphene quantum dots. The multi-source environment perception fusion module 3 includes a biological contamination sensor 31 and a microenvironment multi-parameter sensor group 32. The dynamic bionic protection and flow guide structure 4 includes a pupil type variable aperture 41 and a micro-channel sweat management unit 42. The energy collaborative management module 5 is electrically connected to the device main power supply and the main control module 1, and internally has a radio frequency energy harvesting circuit 51 and a super capacitor 52. The input end of the main control module 1 is electrically connected to the input end of the multi-source environment perception fusion module 3, a federal learning digital twin model is run inside the main control module 1, and a disinfection scheme is dynamically generated using historical disinfection data, environmental data, and user physiological state; the output end of the main control module 1 is electrically connected to the input end of the programmable ultraviolet light unit 21, the input end of the microcavity plasma unit 22, the input end of the variable aperture 41, the input end of the micro-channel unit 42, and the input end of the energy collaborative management module 5. Embodiment 2: The programmable ultraviolet light unit 21 includes a micro UV-C LED matrix 211, an adaptive light field scanning algorithm 212, and a multi-channel constant current driving circuit 213. The micro UV-C LED matrix 211 is based on an inverted chip structure of aluminum gallium nitride material, and the light emitting surface is integrated with a nano photonic structure; a micro-channel heat sink 2111 is integrated below the micro UV-C LED matrix 211. The main control module 1 is configured to execute an adaptive light field scanning algorithm based on the microbial distribution thermodynamic map predicted by the digital twin model to dynamically plan the irradiation dose and scanning path of each pixel sub-region.

[0017] In an embodiment, the programmable ultraviolet light unit is applied to the automatic disinfection of the inner cavity surface of a precision medical instrument. When it is predicted through the digital twin model that a certain wrinkle area in the lumen of a certain batch of endoscopes has a high risk of microbial contamination, the main control module starts the micro-UV-CLED matrix composed of aluminum gallium nitride material flip-chip, and the nano-photonic structure on the light-emitting surface of the LED matrix improves the light-emitting performance and directivity of ultraviolet light. During the disinfection process, the integrated micro-channel radiator maintains the temperature of the chip at a high power density by circulating the cooling liquid. At the same time, the main control module runs the adaptive light field scanning algorithm, analyzes the microbial distribution probability heat map generated by the digital twin model in real time, dynamically calculates the optimal scanning path covering the inner cavity surface, divides the surface into pixelized sub-regions, and then assigns appropriate ultraviolet irradiation dose and residence time to each sub-region according to the predicted contamination concentration. The multi-channel constant current driving circuit automatically controls the switching time and driving current of each pixel unit in the LED matrix according to the plan, achieves the adaptive light field coverage effect of key irradiation in high-pollution areas and rapid scanning in low-pollution areas, and reduces the total irradiation time and total energy consumption while ensuring the disinfection performance. The experimental results are shown in Table 1.

[0018] Table 1 Comparison of adaptive ultraviolet disinfection performance

[0019] The results show that the disinfection system equipped with a programmable ultraviolet light unit and an intelligent light field planning algorithm can achieve accurate, efficient and adaptive disinfection based on digital twin prediction, realize the transition from "surface disinfection" to "point disinfection" of the instrument surface, and improve the reliability and efficiency of disinfection. Embodiment 3: The microcavity plasma unit 22 with a super-hydrophobic electrode is integrated with a micro-temperature control element 221 and a high-frequency high-voltage power supply module 222, and the high-frequency high-voltage power supply module 222 is built-in impedance matching network 2221; the super-hydrophobic electrode is a multilayer composite structure based on vanadium dioxide phase change material, and the electromagnetic response characteristics are adjustable with temperature, and the main control module 1 can control the microcavity plasma unit 22 to switch between capacitive coupling mode and surface wave plasma mode; The region where the intelligent response type photocatalytic coating 23 based on the composite material of the up-conversion nanomaterial and graphene quantum dots is located is provided with a low-power near-infrared LED 231 as an excitation light source; the core-shell structure of the up-conversion nanoparticles is doped with a rare earth ion pair, and the graphene quantum dots serve as an electronic mediator, which together constitute a quantum dot fluorescence resonance energy transfer system. In the embodiment, high-end biosafety laboratory complex pollutants are efficiently purified as an experimental case, and a microcavity plasma unit and an intelligent photocatalytic coating are used. In the composite purification mode, the main control module first activates the low-power near-infrared LED of the intelligent photocatalytic coating, so that the up-conversion nanoparticles doped with a rare earth ion pair are excited to convert low-energy near-infrared light into high-energy ultraviolet / visible light through the quantum dot fluorescence resonance energy transfer system, and then activate the photocatalytic degradation reaction of organic pollutants by exciting the graphene quantum dots and other catalytic materials on the coating. When the concentration of microbial aerosols is detected to be high, the main control module starts the microcavity plasma unit. The high-frequency high-voltage power supply module drives the super-hemispherical surface electrode, the built-in impedance matching network ensures effective energy coupling, the temperature of the vanadium dioxide electrode is controlled by using a micro temperature control element, the electromagnetic properties are changed by adjusting the temperature, and the main control module can dynamically switch the plasma discharge mode: the capacitive coupling mode can generate a large-area uniform plasma for broad-spectrum rapid inactivation; the surface wave plasma mode can excite higher active particles to intensify the removal of drug-resistant pathogens. The two work together or at different times to achieve deep collaborative purification of gaseous complex pollutants. The system can realize the synergistic effect of photocatalysis and intelligent plasma, adapt to different complex pollutants through mode switching, and the results are shown in Table 2.

[0020] Table 2 Comparison of air complex pollutant collaborative purification performance

[0021] It is shown that this method improves the purification rate and energy utilization rate in a complex air pollution scene. Embodiment 4: The biological pollution degree sensor 31 includes a micro electrochemical full system unit 311, a lock-in amplification circuit 312, and a micro interdigital electrode 313. The surface of the micro interdigital electrode 313 is modified with a molecularly imprinted polymer layer by in-situ polymerization; the sensor 31 is based on the time domain dielectric spectrum analysis principle, analyzes the relaxation time distribution of the complex impedance spectrum in a specific frequency range, quantitatively analyzes the formation stage of the biofilm and the activity of the bacterial colony, and distinguishes the pollution characteristics of gram-positive bacteria and gram-negative bacteria.

[0022] The embodiment takes the online monitoring of biofilm pollution inside the medical endoscope lumen as the experimental scene, adopts the bio-pollution sensor of the scheme, and contacts the sensor with the measured liquid or wet surface. The molecularly imprinted polymer layer in-situ polymerized on the surface of the micro-interdigital electrode selectively adsorbs target bacteria or bacterial metabolites. The micro electrochemical full system unit inputs an alternating excitation signal containing a specific frequency range to the micro-interdigital electrode. The lock-in amplifier circuit accurately measures the complex impedance response of the electrode system, obtains the complete dielectric spectrum thereof in the frequency range, and the host control module decouples the collected complex impedance data through the time-domain dielectric spectrum analysis method, fits the relaxation time distribution of the complex impedance spectrum, and quantitatively analyzes the important parameters of bio-pollution. The state of the biofilm is discriminated, whether it is in the initial adhesion, mature development or shedding diffusion state, and the pollution type dominated by gram-positive bacteria or gram-negative bacteria can be further discriminated through different characteristic relaxation peaks, so that the synchronous online discrimination of the pollution degree and type is realized. The experimental results are shown in Table 3.

[0023] Table 3 Comparison of bio-pollution monitoring performance

[0024] The experiment shows that the sensor realizes high-sensitivity and specific online monitoring of the complex biofilm pollution state by fusing the molecular imprinting and dielectric spectrum analysis technology, and provides real-time data support for precise cleaning and infection control. The pupil type variable aperture 41 is integrated with a micro light sensitive sensor 411, which detects the aperture aperture in real time and forms a closed loop control; the pupil type variable aperture 41 is driven by a double electric layer electrostatic MEMS actuator based on carbon nanotube aerogel electrode; The entrance of the micro-channel sweat management unit 42 is provided with a sweat biomarker sensor 421, the micro-channel sweat management unit 42 includes a biomimetic microneedle structure 421 integrated with a pH responsive hydrogel valve, a hydrophilic and hydrophobic patterned surface 422 with micro-nano composite structure, and a protein adsorption resistant coating 423 coated on the inner wall of the micro-channel, and the sweat is driven to transport directionally by generating Laplace pressure difference.

[0025] In an embodiment, a new generation of smart contact lenses is used as an application platform to achieve uninterrupted and comfortable monitoring and adjustment of the ocular surface environment and biochemical indicators. When the user wears the smart contact lens, the micro photosensitive sensor in the entire system continuously monitors the ambient light intensity and feeds it back to the master chip, which derives the appropriate light amount and drives the pupil variable aperture. The aperture can be controlled by a double-layer electrostatic MEMS actuator based on carbon nanotube aerogel electrodes, which can quickly and smoothly change the aperture like the human eye iris to dynamically adjust the incident light intensity and protect the retina and adjust the posterior segment imaging sensor input. At the same time, the PH-responsive hydrogel valve on the surface of the bionic microneedle structure in contact with the cornea can intelligently adjust the valve opening according to the changes in the ocular surface microenvironment pH, thereby entering the microfluidic channel. The anti-protein adsorption coating on the inner wall of the microfluidic channel will hinder the blockage of biological molecules, and the surface with a certain hydrophilic and hydrophobic patterned structure drives the directional pump-free transport of micro-amount of tear fluid to the entrance area of the sweat biomarker sensor by generating Laplace pressure difference, and the sensor analyzes the micro-amount of tear fluid transported thereto to complete in-situ monitoring of biomarkers such as glucose. The experimental results are shown in Table 4.

[0026] Table 4 Performance test results of integrated system

[0027] This embodiment verifies the simultaneous realization of intelligent adaptation to environmental light and management and in-situ monitoring of ocular surface micro-amount of biological fluid on a very small platform, becoming a technical prototype of the next generation of wearable health monitoring devices. Embodiment 6: The digital twin model based on federated learning embeds a meta-reinforcement learning framework; the meta-reinforcement learning framework models the disinfection process as a partially observable Markov decision process, and continuously optimizes its decision strategy through real-time interaction data with the environment perception fusion module 3; The master module 1 is configured to fine-tune the model locally using user personalized data to generate personalized disinfection strategies, while only uploading the strategy gradient update amount of the model to the cloud after homomorphic encryption to participate in federated aggregation, and the decrypted global model parameters after cloud aggregation are verified by secure multi-party computation and then distributed to each local device for model update.

[0028] In practical examples, taking a smart disinfection robot cluster as an example, based on a federated learning digital twin model based on a meta-reinforcement learning framework, each robot models its actual disinfection process as a partially observable Markov decision when performing disinfection in environments such as hospital rooms and operating rooms. The digital twin model continuously receives part of the observable state information in the current state of the environment perception fusion module, including the current space, object surface material, real-time microbial load, and personnel flow. The meta-reinforcement learning framework continuously generates and executes dynamic disinfection paths, ultraviolet light dosage, and disinfectant atomization decisions during the actual disinfection process, and adjusts the decision strategy online through data and actual disinfection results to adapt to environmental uncertainty. Each robot uses local personalized task data to fine-tune the local digital twin model, and homomorphically encrypts the gradient update of the model strategy. Through the cloud server, federated aggregation is performed, the cloud does not decrypt the privacy of each participant, and the encrypted gradient is safely aggregated to generate global model update parameters. After decryption and secure multi-party computing protocol verification, the correct and consistent parameters are distributed to each robot through an encrypted channel. The robot receives and updates the updated digital twin model, which evolves into a digital twin model that can better design dynamic disinfection strategies. The performance comparison of different learning schemes is shown in Table 5.

[0029] Table 5 Performance comparison of different learning schemes

[0030] Table 5 verifies that the digital twin framework combining federated learning and meta-reinforcement learning can realize distributed intelligent co-evolution and personalized strategy generation of disinfection equipment clusters while ensuring data privacy, and improve the dynamic overall disinfection capability and self-adaptability. Embodiment 7: An adaptive disinfection and sterilization control method for a wearable medical device, applied to the device of any one of claims 1-6, comprising the following steps: Step one, system initialization and environment perception: The main control module 1 initializes each sensor and actuator, and the multi-source environment perception fusion module 3 collects biological contamination data, microenvironment parameters, and user physiological state data in real time through the micro-interdigital electrode 313 and the lock-in amplification circuit 312 and transmits them to the main control module 1. Step two, digital twin model construction and strategy generation: The main control module 1 fuses historical disinfection data and real-time perception data based on a federated learning digital twin model to construct a current microbial distribution thermal map, and calculates and generates an optimal disinfection strategy including disinfection mode selection, energy allocation parameters, and execution timing through an embedded meta-reinforcement learning framework; Step three, multi-modal disinfection collaborative execution: The master module 1 synchronously controls the adaptive light field scanning performed by the micro UV-C LED matrix 211 through the multi-channel constant current drive circuit 213 according to the disinfection strategy, while starting the micro-channel heat sink 2111 for heat management; controls the micro temperature control element 221 to adjust the temperature of the super-hydrophobic surface electrode, and uses the high-frequency high-voltage power supply module 222 and the impedance matching network 2221 to excite the target mode of plasma; turns on the low-power near-infrared LED 231 to excite the intelligent response type photocatalytic coating 23, and generates a catalytic sterilization effect; Step four, dynamic environment management and real-time optimization: The master module 1 controls the dynamic biomimetic protection and flow guiding structure 4 to perform operations: based on the feedback of the micro light-sensitive sensor 411, the opening and closing aperture of the pupil type variable aperture 41 is adjusted in a closed loop to optimize the disinfection energy output; according to the signal of the sweat biomarker sensor 421, the pH responsive hydrogel valve of the micro-channel sweat management unit 42 is activated, and the hydrophilic and hydrophobic patterned surface 422 and the anti-protein adsorption coating 423 are used to cooperatively drive the directional transport of sweat to manage the interface microenvironment; according to the real-time feedback data, the disinfection strategy is dynamically adjusted through the meta-reinforcement learning framework; Step five, energy coordination management and energy efficiency optimization: The energy coordination management module 5 monitors the system energy consumption state in real time, enables the super capacitor 52 for instantaneous energy compensation when the disinfection peak power demand is required, and collects environmental electromagnetic energy through the radio frequency energy collection circuit 51 to improve the system endurance; Step six, disinfection efficiency evaluation and model updating: The intelligent evaluation module 24 records the entire disinfection process and obtains the sterilization effect index, the master module 1 receives the disinfection effect index data recorded by the biological pollution degree sensor 31, and analyzes and processes the index data; the micro electrochemical cleaning unit 311 cleans and regenerates the micro interdigital electrode 313, and desensitizes the local digital twin model according to the disinfection result, and the model update parameters are uploaded to the federated learning through homomorphic encryption.

[0031] In the implementation example, an intelligent sports wristband is used to perform adaptive disinfection of the skin contact surface and microenvironment management. After system initialization, the multi-source sensing module collects data on skin biocontamination, microenvironment, and physiology on the wrist. The federated learning digital twin model is integrated and fused, and the mode and parameters are generated by the meta-reinforcement learning model. The main control module coordinates multimodal disinfection: micro UV-CLED matrix adaptive light field scanning, ultra-sensitive surface electrode temperature-controlled emission mode plasma, near-infrared LED activated photocatalytic coating, and microchannel heat sink thermal stability. The dynamic environment management unit optimizes in real time: pupil-type variable aperture closed-loop adjustment of light output, and the sweat management unit guides sweat through pH-responsive hydrogel valves and hydrophilic / hydrophobic surfaces based on sensor signals. The energy management unit uses power peak supercapacitor compensation and extends battery life through radio frequency energy harvesting. After disinfection, the module records trigger electrode self-cleaning and local model update. The desensitized model update parameters are uploaded to the federated learning global model evolution through homomorphic encryption. The results are shown in Table 6.

[0032] Table 6 Performance Comparison of Different Solutions

[0033] This solution enables wearable devices to transform from passive protection to proactive, intelligent, and personalized health management, offering excellent disinfection, dynamic environmental adjustment, and privacy protection. Example 8: In step two, the optimal disinfection strategy is generated through the following intelligent decision function, which integrates microbial situational awareness and energy constraints: In formula (1), These are the weighting coefficients; It represents the total number of microorganisms monitored at time t; It is the microbial threshold; It is a risk adjustment factor; It is the risk level of microbial transmission; It is the real-time power required for the disinfection operation at time t; It is the maximum power that the system can provide at time t; It is the power scaling factor; This represents the actual operational capacity of the disinfection system at time t. This is the maximum operating capacity of the disinfection system; These are the disinfection requirements parameters set by the user at time t; It is the baseline value for user needs; k is the temperature sensitivity coefficient; This is the actual temperature of the current environment; This is the temperature at which disinfection is most effective; Introduce a dynamic adjustment mechanism based on real-time environmental data and user behavior: In formula (2), i = 1, 2, 3; It is the base value of the weights; is an adjustment coefficient; is a daily period time constant; the function makes the weight coefficient dynamically adjust according to the microbial concentration deviation, power budget difference and time period. The experiment is to verify an intelligent decision function and dynamic adjustment mechanism for generating an optimal disinfection strategy. The indoor environment cabin is composed of a control air conditioning system, a programmable microbial aerosol release device, a multi-spectral biological aerosol sensor, a ultraviolet and chemical spray disinfection module, and a real-time power monitoring unit. The software platform generates sensor data streams based on RoS and runs decision algorithms. The experiment lasts for 7 days, simulates common biological pollution scenarios by means of office, concentration of personnel, night confinement, etc., and artificially introduces different concentrations of Bacillus subtilis as model microorganisms to simulate biological pollution.

[0034] Formula (1) is a comprehensive evaluation function of the optimal disinfection scheme , which consists of three weighted items: a microbial threat item, which is a nonlinear amplification of the ratio of real-time microbial concentration to threshold value, and a sinusoidal modulation related to the risk level of transmission to strengthen the response to high-risk situations; an energy and system capacity constraint item, which is based on the negative exponential difference between disinfection demand power and system power budget, multiplied by the current running capacity saturation function of the system, multiplied by the willingness intensity of the user demand parameter arctangent function, multiplied by the inhibition function of the environmental temperature deviation from the optimal temperature. Formula (2) represents the dynamic change of the decision weight. The weight coefficient is transformed by three real-time factors on the basis of : the microbial concentration deviation acts on the weight through the hyperbolic tangent function, the over-standard constraint threat item; the power supply and demand anomaly acts on the weight through the negative exponential function, the power shortage constraint energy consumption item; the daily period cosine modulation is used for time period mode, rapid response to environment, availability and time period change. The experiment lasts for 7 days, simulates various indoor scenarios and simulates microbial pollution, and compares the static weight strategy with the adaptive dynamic strategy. The system collects environmental data every five minutes, substitutes the decision function of the two strategies, and solves the value, which exceeds the threshold value and starts disinfection, and compares the effects of the two strategies in terms of microbial control effect, energy consumption and satisfaction. The comparison of the comprehensive performance of the strategies is shown in Table 7.

[0035] Table 7 Comparison of comprehensive performance of strategies

[0036] The dynamic adaptive method in Table 7 is higher than the static method in terms of shortening the microbial over-standard time, reducing the average energy consumption and user satisfaction.

[0037] Examples of dynamic strategy weight evolution are shown in Table 8.

[0038] Table 8 Examples of dynamic strategy weight evolution

[0039] Table 8 shows that the weight is dynamically changed according to the scene, the greater the threat weight, the higher the power tightness, and the higher the energy weight, the more effective the adaptation.

[0040] Experiments prove the feasibility of the adaptive disinfection decision system. Formula (1) is a multi-objective comprehensive decision function, and formula (2) is an intelligent dynamic adjustment method of weight. The adaptive disinfection decision system can judge the microbial threat, energy constraint and user requirements, and adopt an appropriate scheme for disinfection. In step three, the multi-modal disinfection collaborative execution is realized through the following cross-modal collaborative optimization function: In formula (3), is the total disinfection effect at spatial position and time t; are the weight coefficients of ultraviolet, plasma and photocatalytic modalities respectively; N is the number of ultraviolet disinfection devices; is the radiation intensity of the i-th ultraviolet device; is the spatial position vector; is the spatial position vector of the i-th ultraviolet device; is the spatial diffusion coefficient of ultraviolet radiation; is the correction coefficient of local microbial concentration to ultraviolet effect; erf is the error function; is the local microbial concentration at position ; is the microbial concentration reference value; represents the Laplacian of the plasma potential , and t is the time variable; is the characteristic time constant of plasma action; is the absorption coefficient of photocatalytic material; is the ambient light intensity at time t; is the photocatalytic light illumination period. This experiment is a multi-modal disinfection collaborative execution method based on a cross-modal collaborative optimization function. Ultraviolet, plasma and photocatalytic disinfectors are set up in a controllable chamber, and a grid sensor network is used to monitor microbial concentration, light and electric field intensity.

[0041] The function in formula (3) represents the spatiotemporal contribution value of the three disinfection modalities; the ultraviolet contribution term is a Gaussian diffusion with nonlinear correction of local microbial concentration to severely polluted areas. The plasma contribution term is a potential Laplacian, and the spatial density gradient of active particles is time-saturated; the photocatalytic contribution term is the ambient light intensity and the periodically modulated light, with a weighting coefficient to represent the action of each modality.

[0042] The experiment first independently calibrates the characteristic parameters of each modality, and compares three execution strategies based on formula 3, namely sequential execution, parallel independent execution, and optimized collaborative execution based on formula three. Optimization refers to adjusting the output of each device according to real-time monitoring data to maximize the total disinfection of the space. The experimental results are shown in the average kill rate, spatial uniformity, energy consumption, and time efficiency. The comparison of the effects of different execution strategies is shown in Table 9.

[0043] Table 9 Comparison of effects of different execution strategies

[0044] As shown in Table 9, the optimized collaborative strategy has the best disinfection efficiency, uniformity, and time efficiency, and the energy consumption is lower than that of the parallel strategy.

[0045] The dynamic analysis of the contribution of the collaborative strategy modality is shown in Table 10.

[0046] Table 10 Dynamic analysis of the contribution of the collaborative strategy modality

[0047] Table 10 shows that the contribution of each modality is increased according to real-time needs, with increased coverage of ultraviolet light in high concentration areas, increased uniform coverage of plasma in the later period, and increased photocatalysis period, verifying the collaborative ability of the formula.

[0048] The experiment proves the feasibility of the multi-modal collaborative optimization function based on formula (3), dynamically adjusts the collaborative disinfection between each modality, and provides an algorithm basis for intelligent disinfection in complex environments. In step six, the model update is achieved through the following privacy-protected federated evolutionary learning function: In formula (4), is the updated local model parameter; is the local model parameter before updating; is the learning rate; represents the gradient of the local loss function with respect to the parameter ; represents the gradient mask matrix; represents the Hadamard product; is the global consistency weight coefficient; is the global shared model parameter in federated learning; is the KL divergence of the local model and the global model; is the benchmark threshold of the KL divergence; is the individualized weight coefficient; is the individualized objective function; represents the sample size of the local data size; is the data size benchmark threshold; k is the data size sensitivity coefficient; represents the encrypted weight matrix; a personalized objective function is introduced, defined as as a regularization term, based on user preferences and local environment data: ; in formula (5), M is the model parameter dimension, is the parameter weight, is the actual value of the jth model parameter, is the parameter value learned from historical data according to user preferences; is the balance coefficient; is the optimal disinfection effect; is the actual disinfection effect; is the disinfection effect scale factor; is the target satisfaction of user demand; is the actual user demand satisfaction; is the user demand scale factor. The experiment verifies the federated evolutionary learning model update mechanism with privacy protection and personalization optimization. The experiment sets up a simulation distributed disinfection network with one central server and ten edge nodes, each node uses local data and individual demand to verify the update mechanism defined by formula (4) and formula (5).

[0049] Formula (4) defines the model update, which is divided into three evolutionary increments. The basic gradient term uses a mask matrix, the global consistency is adjusted by the exponential decay function of KL divergence according to the direction of parameter difference, balancing local adaptation and global consensus; the direction of personalized enhancement term is determined by the gradient of personalized objective function defined by formula (5), and the strength varies with local data, which activates personalized learning when local data is sufficient and suppresses overfitting when local data is insufficient. The final update vector uses an encrypted matrix transformation.

[0050] Formula (5) defines the personalized objective function, which includes the parameter preference term to make the model closer to the user's historical preferences, the effect adaptation term to make the actual effect deviate from the optimal optimal, and the demand fitting term to make the decision meet the user's set target, providing personalized guidance for model evolution.

[0051] The experiment simulates multiple rounds of federated learning, and compares standard federated averaging, differential privacy federated learning, fixed personalized federated learning and the proposed privacy protection federated evolutionary learning. Randomly select nodes according to their local data and received global model to calculate encrypted parameter updates and upload aggregation according to formula at each round. The evaluation indexes are model versatility, personalized effect, privacy protection and user satisfaction. The performance comparison of different federated learning strategies is shown in Table 11.

[0052] Table 11 Performance comparison of different federated learning strategies

[0053] The data in Table 11 show that the method performs best in both global and local, has a significant privacy protection effect, and has the highest user satisfaction.

[0054] The evolution of the key indicators of the evolutionary learning process is shown in Table 12.

[0055] Table 12 Evolution of key indicators of evolutionary learning process

[0056] The results in Table 12 show that as the learning deepens, the model improves the personalization ability with the increase of data in the connection with the global consensus, from the focus of learning to the importance of fitting with users after learning.

[0057] The experimental verification proves the feasibility of the federated evolutionary learning mechanism. Formula (4) is a general, personalized and secure formula for privacy protection, and formula (5) is an optimization target for personalization. Through formula (4) and formula (5), the system can realize the general intelligent model, the general personalized intelligent model and the safe collaborative evolution of the distributed system.

[0058] Although the specific embodiments of the present application are described above, those skilled in the art should understand that these specific embodiments are only illustrative, and those skilled in the art can make various omissions, substitutions and changes to the details of the above method and system without departing from the principles and essence of the present application; for example, the above method steps are combined, and the substantially same method is used to perform substantially the same function to achieve substantially the same result, which belongs to the scope of the present application; therefore, the scope of the present application is only limited by the appended claims.

Claims

1. An adaptive disinfection device for wearable medical devices, comprising: The device comprises a master control module (1), a multi-modal disinfection execution module (2), a multi-source environment perception fusion module (3), a dynamic biomimetic protection and flow guiding structure (4), and an energy collaborative management module (5). The multi-modal disinfection execution module (2) and the multi-source environment perception fusion module (3) are embedded and conformally integrated in a special-shaped curved surface in contact with the user's skin on the device body. The multi-modal disinfection execution module (2) comprises a programmable ultraviolet light unit (21), a microcavity plasma unit (22) using a super-hydrophobic surface electrode, and an intelligent response type photocatalytic coating (23) based on a composite material of upconversion nanomaterials and graphene quantum dots. The multi-source environment perception fusion module (3) comprises a biological pollution level sensor (31) and a microenvironment multi-parameter sensor group (32). The dynamic biomimetic protection and flow guiding structure (4) comprises a pupil type variable aperture (41) and a micro-channel sweat management unit (42). The energy collaborative management module (5) is electrically connected to the device main power supply and the master control module (1), and internally comprises a radio frequency energy harvesting circuit (51) and a super capacitor (52). The input end of the master control module (1) is electrically connected to the input end of the multi-source environment perception fusion module (3), and a federal learning digital twin model is run inside the master control module (1) to dynamically generate a disinfection scheme using historical disinfection data, environmental data, and user physiological state. The output end of the master control module (1) is electrically connected to the input end of the programmable ultraviolet light unit (21), the input end of the microcavity plasma unit (22), the input end of the variable aperture (41), the input end of the micro-channel unit (42), and the input end of the energy collaborative management module (5).

2. The self-adapting sterilization device of wearable medical equipment according to claim 1, wherein, The programmable ultraviolet light unit (21) comprises a micro UV-C LED matrix (211), an adaptive light field scanning algorithm (212), and a multi-channel constant current driving circuit (213). The micro UV-C LED matrix (211) is a flip-chip structure based on aluminum gallium nitride material, and the light emitting surface is integrated with a nano photonic structure. A micro-channel radiator (2111) is integrated below the micro UV-C LED matrix (211). The master control module (1) is configured to execute an adaptive light field scanning algorithm based on the microbial distribution thermodynamic map predicted by the digital twin model to dynamically plan the irradiation dose and scanning path of each pixel sub-region.

3. The self-adapting sterilization device for wearable medical devices of claim 1, wherein, The microcavity plasma unit (22) using a super-hydrophobic surface electrode is integrated with a micro temperature control element (221) and a high-frequency high-voltage power supply module (222), and the high-frequency high-voltage power supply module (222) is internally provided with an impedance matching network (2221). The super-hydrophobic surface electrode is a multi-layer composite structure based on vanadium dioxide phase change material, and the electromagnetic response characteristics are adjustable with temperature. The master control module (1) can control the microcavity plasma unit (22) to switch between capacitive coupling mode and surface wave plasma mode. The intelligent response type photocatalytic coating (23) based on the composite material of up-conversion nanomaterial and graphene quantum dots is provided with a low-power near-infrared LED (231) as an excitation light source in the region where the intelligent response type photocatalytic coating (23) is located; the core-shell structure of the up-conversion nanoparticles is doped with a rare earth ion pair, and the graphene quantum dots serve as an electronic mediator, which together constitute a quantum dot fluorescence resonance energy transfer system.

4. The self-adapting sterilization device for wearable medical devices of claim 1, wherein, The biological contamination sensor (31) comprises a micro electrochemical full system unit (311), a lock-in amplification circuit (312), and a micro interdigital electrode (313); The surface of the micro interdigital electrode (313) is modified with a molecularly imprinted polymer layer through in-situ polymerization; the sensor (31) is based on the time domain dielectric spectrum analysis principle, and the formation stage of the biofilm and the activity of the bacterial colony are quantitatively analyzed by analyzing the complex impedance spectrum relaxation time distribution in a specific frequency range, and the pollution characteristics of gram-positive bacteria and gram-negative bacteria are distinguished.

5. The self-adapting sterilization device for wearable medical devices of claim 1, wherein, The pupil type variable aperture (41) is integrated with a micro light-sensitive sensor (411) to detect the aperture aperture in real time and form a closed-loop control; the pupil type variable aperture (41) is driven by a double-electrode layer electrostatic MEMS actuator based on a carbon nanotube aerogel electrode; A sweat biomarker sensor (421) is arranged at the inlet of the micro-channel sweat management unit (42), the micro-channel sweat management unit (42) comprises a biomimetic microneedle structure (421) integrated with a pH-responsive hydrogel valve, a hydrophilic and hydrophobic patterned surface (422) with a micro-nano composite structure, and a protein adsorption resistant coating (423) coated on the inner wall of the micro-channel, and the sweat is driven to be transported directionally by generating Laplace pressure difference.

6. The self-adapting sterilization device for wearable medical devices of claim 1, wherein, The digital twin model based on federated learning is embedded with a meta-reinforcement learning framework; the meta-reinforcement learning framework models the disinfection process as a partially observable Markov decision process, and continuously optimizes its decision strategy through real-time interaction data of the environment perception fusion module (3); The main control module (1) is configured to fine-tune the model locally using user personalized data to generate personalized disinfection strategies, while only uploading the strategy gradient update amount of the model to the cloud after homomorphic encryption for federated aggregation, and the global model parameters aggregated on the cloud are decrypted and verified by secure multi-party computation, and then distributed to each local device for model updating.

7. An adaptive disinfection control method for a wearable medical device, applied to the apparatus of any one of claims 1-6, characterized in that, The method comprises the following steps: Step one, system initialization and environment perception: The main control module (1) initializes each sensor and actuator, and the multi-source environment perception fusion module (3) collects biological contamination data, micro-environment parameters and user physiological state data in real time through the micro interdigital electrode (313) and the lock-in amplification circuit (312), and transmits them to the main control module (1); Step two, digital twin model construction and strategy generation: The main control module (1) fuses historical disinfection data and real-time perception data based on the digital twin model of federated learning to construct a current microorganism distribution thermal map, and calculates and generates an optimal disinfection strategy including disinfection mode selection, energy distribution parameters and execution timing through the embedded meta-reinforcement learning framework; Step three, multi-modal disinfection collaborative execution: The master module (1) synchronously controls the micro-UV-C LED matrix (211) to perform adaptive light field scanning according to the disinfection strategy, and starts the micro-channel radiator (2111) for heat management; controls the micro-temperature control element (221) to adjust the temperature of the super-hydrophobic surface electrode, and uses the high-frequency high-voltage power supply module (222) and the impedance matching network (2221) to excite the target mode of plasma; turns on the low-power near-infrared LED (231) to excite the intelligent response type photocatalytic coating (23), and generates a catalytic sterilization effect; Step four, dynamic environment management and real-time optimization: The master module (1) controls the dynamic biomimetic protection and flow guiding structure (4) to perform operations: based on the feedback of the micro light-sensitive sensor (411), the opening and closing aperture of the pupil type variable aperture (41) is adjusted in a closed loop to optimize the disinfection energy output; according to the signal of the sweat biomarker sensor (421), the pH-responsive hydrogel valve of the micro-channel sweat management unit (42) is activated, and the hydrophilic and hydrophobic patterned surface (422) and the anti-protein adsorption coating (423) are used to cooperatively drive the directional transport of sweat, and the interface microenvironment is managed; according to the real-time feedback data, the disinfection strategy is dynamically adjusted through the meta-reinforcement learning framework; Step five, energy coordination management and energy efficiency optimization: The energy coordination management module (5) monitors the system energy consumption state in real time, enables the super capacitor (52) for instantaneous energy compensation when the disinfection peak power demand is required, and collects environmental electromagnetic energy through the radio frequency energy harvesting circuit (51) to improve the system endurance; Step six, disinfection efficiency evaluation and model updating: The intelligent evaluation module (24) records the entire disinfection process and obtains the sterilization effect index, the master module (1) receives the disinfection effect index data recorded by the biological pollution degree sensor (31), and analyzes and processes the index data; the micro electrochemical cleaning unit (311) cleans and regenerates the micro interdigital electrode (313), and desensitizes the local digital twin model according to the disinfection result, and the model update parameters are uploaded to the federated learning through homomorphic encryption.

8. The control method according to claim 7, characterized by In step two, the optimal disinfection strategy is generated by the following intelligent decision function, which integrates microbial situation awareness and energy constraints: ; in formula (1), is the weight coefficient; is the total amount of microorganisms monitored at time t; is the microbial threshold value; is the risk correction coefficient; is the risk level of microbial transmission; is the real-time power required for disinfection at time t; is the maximum power that the system can provide at time t; is the power scale coefficient; is the actual operating capacity of the disinfection system at time t; is the maximum operating capacity of the disinfection system; is the user-set disinfection demand parameter at time t; is the reference value of user demand; k is the temperature sensitivity coefficient; is the actual temperature of the current environment; is the optimal temperature for disinfection effect; A dynamic adjustment mechanism based on real-time environmental data and user behavior is introduced: ; in equation (2), i = 1, 2, 3; is a weight base value; is an adjustment coefficient; is a daily period time constant; the function makes the weight coefficient dynamically adjust according to the microbial concentration deviation, power budget difference and time period.

9. The control method according to claim 7, characterized by In step three, the multi-modal disinfection co-execution is implemented through the following cross-modal co-optimization function: ; In formula (3), is the total disinfection effect at spatial position and time t; are the weight coefficients of UV, plasma, and photocatalysis modalities, respectively; N is the number of UV disinfection devices; is the radiation intensity of the i-th UV device; is the spatial position vector; is the spatial position vector of the i-th UV device; is the spatial diffusion coefficient of UV radiation; is the correction coefficient of local microbial concentration to UV effect; erf is the error function; is the local microbial concentration at position ; is the reference value of microbial concentration; represents the Laplacian of plasma potential ; t is the time variable; is the characteristic time constant of plasma action; is the absorption coefficient of photocatalytic material; is the ambient light intensity at time t; is the photocatalytic light period.

10. The control method according to claim 7, characterized by In step six, the model update is achieved by the following privacy-protected federated evolutionary learning function: ; In formula (4), is the updated local model parameter; is the local model parameter before updating; is the learning rate; represents the gradient of the local loss function with respect to the parameter ; represents the gradient mask matrix; represents the Hadamard product; is the global consistency weight coefficient; is the global shared model parameter in federated learning; is the KL divergence of the local model and the global model; is the benchmark threshold of the KL divergence; is the personalized weight coefficient; is the personalized objective function; represents the sample size of the local data volume; is the data volume benchmark threshold; k is the data volume sensitivity coefficient; represents the encrypted weight matrix; the personalized objective function is introduced, and is defined as a regularization term based on user preferences and local environmental data: ; in formula (5), M is the model parameter dimension, is the parameter weight, is the actual value of the jth model parameter, is the parameter value learned from historical data according to user preferences; is the balance coefficient; is the optimal disinfection effect; is the actual disinfection effect; is the disinfection effect scale factor; is the target satisfaction degree of user demand; is the actual user demand satisfaction degree; is the user demand scale factor.

Images