A Dual-Fuzzy Decoupling Adaptive Weight Fusion Control Method for Smart Breathing Helmets
By constructing a dual-channel independent control architecture through a dual-fuzzy decoupling adaptive weight fusion control method, the problems of decreased air supply accuracy and insufficient control stability of the smart breathing helmet under all motion conditions are solved, and smooth control and efficient operation and maintenance under all working conditions are achieved.
Patent Information
- Application Number
- CN202611510615.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-28
- Publication Date
- 2026-11-17
AI Technical Summary
Existing smart breathing helmets suffer from reduced air supply accuracy and insufficient control stability under full motion conditions when facing nonlinear airflow disturbances. Traditional PID control requires frequent manual calibration, resulting in high maintenance costs, and single-channel fuzzy PID cannot effectively decouple errors in a layered manner.
A dual-fuzzy decoupling adaptive weighted fusion control method is adopted. A dual-channel independent control architecture is constructed through error orthogonal decoupling. It is combined with a PID parameter iterative correction algorithm with stability constraints and a second-order damping compensation model. Combined with data-driven fuzzy rules, smooth control under all operating conditions is achieved.
It achieves stability and accuracy of airflow output under different working conditions, reduces equipment debugging and maintenance costs, improves system control accuracy and operational stability, and provides a better wearing experience.
Smart Images

Figure CN123113387A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of automatic control technology for powered air-purifying wearable respiratory protective equipment, and particularly relates to a dual-fuzzy decoupling adaptive weighted fusion control method for smart breathing helmets. Background Technology
[0002] Industrial dust masks, outdoor cycling masks, and other equipment require closed-loop, coordinated, and precise control of both airway pressure and breathing flow in smart breathing helmets to adapt to various motion conditions involving nonlinear airflow disturbances, such as when the human body is stationary, walking at low speed, running at high speed, and undergoing posture changes. The intelligent breathing helmet supplies air by matching the human breathing rhythm through a built-in fan and regulating valve. However, high-frequency airflow disturbances are easily generated by human movement, swaying, posture deflection, and running impacts. Long-term operation of the device can also cause the reference pressure to drift, resulting in a decrease in air supply accuracy. Traditional fixed-parameter PID is only suitable for static conditions, requiring frequent manual calibration and resulting in high maintenance costs. Existing single-channel fuzzy PID cannot decouple errors (breathing demand, disturbance, drift) in layers, lacks a dynamic damping compensation mechanism to suppress transient impacts, and the rules rely on experience and solidification, resulting in insufficient system adaptability and control stability under complex conditions. Summary of the Invention
[0003] The purpose of this invention is to provide a dual-fuzzy decoupled adaptive weight fusion control method for smart breathing helmets. By constructing a dual-channel independent control architecture through error orthogonal decoupling, the steady-state accuracy and dynamic disturbance rejection capability are optimized respectively. A PID parameter iterative correction algorithm with stability constraints and a second-order damping compensation model are designed, which are combined with data-driven fuzzy rules and an adaptive weight fusion mechanism to achieve smooth control under all working conditions, reduce equipment debugging and maintenance costs, and improve system control accuracy and operational stability.
[0004] To achieve the above objectives, the technical solution adopted by this invention is: a dual-fuzzy decoupling adaptive weight fusion control method for smart breathing helmets, comprising the following steps: S1. Data Sensing Layer Acquisition: The data sensing layer collects equipment operation data and human motion condition data in real time, covering respiratory steady-state parameters and motion disturbance parameters. The collected raw signals undergo preprocessing operations such as filtering and noise reduction, anomaly removal, and normalization to be unified into a standardized data format. S2. Core algorithm control layer processing: The core algorithm control layer is equipped with a hierarchical control algorithm and an adaptive optimization program. It has built-in functions such as error decoupling, parameter iterative correction, dynamic damping compensation, rule base update, and weight fusion. It performs hierarchical calculations on the data input in step one, processes the steady-state deviation and dynamic motion disturbance of the system respectively, adaptively optimizes the control parameters, and fuses the output of the optimal digital control quantity. S3. The control output layer executes the control command output by the algorithm control layer in step two, completes signal conversion and power drive, adjusts the airflow output state of the device in real time, adapts to different human movement conditions and breathing rhythms, stabilizes the airway pressure and airflow in the helmet, and forms a complete closed-loop control.
[0005] Preferably, the core algorithm control layer includes dual fuzzy decoupling error modeling, steady-state channel high-order exponential iterative adaptive correction control strategy, dynamic channel second-order damping adaptive compensation control strategy, and operating condition disturbance quantification and adaptive weight fusion.
[0006] Preferably, the double-fuzzy decoupling error modeling is as follows: set up Let k be the target baseline parameters for respiration. Define the system's real-time integrated error as the actual output parameters of the system: It includes two coupled components: slow-varying steady-state deviation of respiration and fast-varying disturbance deviation of limb movement. This invention constructs an incremental discrete PID control framework, mathematically described as follows: In the formula: These are the basic reference parameters for the PID controller; The current PID control increment; the total system control output: The overall error is decomposed into two types of error components with completely independent physical properties: Steady-state error, caused by equipment reference deviation, static offset of respiratory rhythm, and parameter mismatch, determines the long-term steady-state accuracy of the system. Dynamic disturbance error, caused by limb swinging, posture switching, and running impact, can easily lead to system overshoot, oscillation, and adjustment lag.
[0007] Extend the scalar error into a three-dimensional error state vector. Characterize the instantaneous error, the rate of change of error, and the cumulative error respectively; construct two sets of third-order orthogonal transformation matrices. Satisfying orthogonal constraints This achieves decoupled separation of the error subspace, mathematically described as follows: Matrix construction rules: This is the low-pass filter transformation matrix, with weights biased towards the cumulative error term, to extract steady-state offset features; The high-pass filter transform matrix is used to extract perturbation abrupt change features by first-order and second-order difference of the weight bias error; where, The specific values are obtained through least squares regression fitting driven by operating condition data.
[0008] The preferred steady-state channel higher-order exponential iterative adaptive correction control strategy is as follows: 1) Parameter-based exponential convergence correction model Based on Lyapunov stability theory, an exponentially saturated smoothing map is constructed to constrain the unbounded fuzzy original increment into a continuous, bounded, and effective correction increment: : Parameter increment after smoothing correction; : Fuzzy inference original increment; : Corresponding dimension error state quantity; The global convergence smoothing coefficient needs to be determined jointly by considering the mathematical properties of the constraint function, system simulation traversal, and physical prototype calibration; it is a range value. At that time, the convergence speed is too slow, and the correction is delayed in the large error condition. At that time, the correction range is drastic, and small oscillations are easily generated under small errors; 2) PID three-parameter differential high-order correction formula To address the different physical meanings of errors corresponding to proportional, integral, and differential parameters, three sets of differentiated exponential convergence correction models are constructed to adapt to the constraint correction requirements of instantaneous error, cumulative error, and error rate of change, respectively. The mathematical descriptions are as follows: In the formula: These represent the increments of the original parameters for the proportional, integral, and differential values of the fuzzy inference output; (Left side) The effective correction increment is obtained after exponential constraint smoothing, and it converges and is bounded throughout the entire process with no abrupt changes in the time series. Iterative parameter updates to the complete model: 3) System stability constraints and convergence characteristic analysis ① Construction of positive definite energy function: Constantly satisfied ,only hour It conforms to the discrete Lyapunov positive definite function specification; ② Monotonic convergence characteristic: After exponential constraint correction, the PID increment is bounded and continuous, with no parameter drift or jumps; under any non-zero steady-state error, the error continuously decays after iteration, satisfying: ③ Negative definiteness criterion for difference: Lyapunov first-order difference ; When steady-state error exists The system energy continues to decrease; after the error returns to zero... The system maintains steady-state equilibrium without oscillation or deviation. ④ Stability enhancement: The exponential constraint operator provides global smoothing and limiting, which mathematically constrains the parameter iteration amplitude and rate of change, ensuring asymptotic stability under all operating conditions.
[0009] Preferably, the steady-state channel standardization iteration process includes the following steps: Step 1: With Periodically collect airway pressure and airflow rate Combined with target parameters Solve for instantaneous error, statistically analyze cumulative error and error rate of change, and construct multidimensional steady-state error characteristics; Step 2: The steady-state fuzzy rule base completes the Mamdani fuzzy inference, and the centroid method is used to defuzzify and output the unconstrained PID original correction increment; Step 3: Substitute The exponential nonlinear constraint correction is completed in proportion, integral, and differential dimensions; Step 4: Iteratively update the real-time optimal parameters based on the factory baseline parameters. ; Step 5: Use the Lyapunov criterion to verify the error convergence characteristics and continuously constrain the parameter correction magnitude; Step 6: Substitute the updated PID parameters into the incremental PID controller to solve for the steady-state control components. It is then transmitted to the weight fusion unit.
[0010] The preferred dynamic channel second-order damping adaptive compensation control strategy specifically includes: 1) Complete Modeling and Parameter Determination of Second-Order Underdamped Continuous Dynamics Continuous domain dynamic compensation benchmark model: Parameter definition and preferred values: The system's natural angular frequency, based on human motion and hardware constraints, ranges from [value range missing]. ; Damping ratio, the preferred critical damping ratio in engineering applications. Suppress overshoot; Incremental smoothing correction coefficient, optimized within a certain range based on human movement and hardware constraints. Suppressing instantaneous jumps in control input; Second-order damping enhancement coefficient, adapted to human movement and hardware constraints, within a certain range. To counteract curvature-type abrupt disturbances; The left side of the formula is the second-order damped inherent amplitude-stabilizing term; the right side is the disturbance advance compensation driving term, which uses the second-order differential term to predict the curvature of the disturbance change, and the first-order differential term to smooth the output jitter. 2) Preprocessing of dynamic disturbance features Four types of disturbance features were selected: respiratory rate deviation, human body triaxial composite acceleration, attitude rotation angular velocity, and airflow velocity deviation. The raw data were filtered by moving mean, 3σ outlier removal, and linearly normalized to the [0,1] standard domain to eliminate dimensional differences.
[0011] 3) Adaptive fuzzy rule-based end-to-end data-driven optimization 4) Discretization of the second-order damping model.
[0012] Preferably, adaptive fuzzy rule-based end-to-end data-driven optimization specifically includes: ①Incremental generation of dataset A full-gradient working condition sample set was constructed, covering five typical working conditions in actual use of smart breathing helmets: stationary, low-speed walking, high-speed running, posture deflection, and strenuous exercise. Multi-dimensional sensor data samples were collected evenly for each working condition. The sample dimensions include airway pressure error, airflow error, human body composite acceleration, posture angular velocity, and airflow fluctuation deviation. First, all collected raw operating condition samples were normalized to map the multi-dimensional heterogeneous data to a standard domain. Eliminating dimensional differences yields a standardized sample vector: In the formula: For the first A standardized working condition sample vector is generated, with each dimension corresponding to the characteristics of pressure error, flow error, motion acceleration, and attitude angular velocity, respectively. Calculate the Euclidean spatial distance between the new sample and the existing cluster centers. The distance calculation formula is as follows: In the formula: For newly added standardized samples collected in real time, For each cluster center sample already stored in the rule base; set a fixed clustering threshold. This serves as a criterion for generating new rules. Incremental rule generation judgment logic: When the minimum Euclidean distance between a new sample and all existing cluster centers is greater than or equal to a set threshold, it is judged as a completely new working condition feature sample, and a new fuzzy rule is automatically generated; otherwise, it is judged as a sample of the same type as an existing rule, and no new rule is added, only the corresponding cluster center parameters are updated. Complete judgment formula: In the formula: This represents the addition of a fuzzy rule. This means no new rules will be added; The newly added rule synchronously completes the initialization of membership parameters, updates the Gaussian membership center value and bandwidth with the new sample as the center, and automatically incorporates it into the fuzzy rule base, realizing continuous incremental iteration of working condition data and adaptive expansion of the rule base to cover boundary extreme working conditions; ② Rule similarity pruning Combined similarity of the two rules: Membership center, :bandwidth, Output mapping value; weighting coefficient It can be freely adapted and adjusted, with a coefficient sum of 1; pruning threshold , Eliminate redundant rules with low activation and high error, among which Based on experience values, this rule base can be adjusted; it also synchronously deletes long-term inactive dormant rules, resulting in a lightweight rule library. ③ Adaptive selection of multi-objective cost function A multi-objective cost function is constructed that takes into account dynamic overshoot, convergence speed, and output stability. Global optimization iteration is performed on the membership centers, membership bandwidth, and rule output weights of the fuzzy rule base to achieve adaptive optimal updating of fuzzy control parameters. The specific implementation steps are as follows: Step 1: Determine the control performance evaluation indicators. Based on the control requirements of the intelligent breathing helmet, three core quantitative performance indicators were selected as the optimization basis: the system's maximum overshoot. System adjustment of convergence time Control output fluctuation variance Among them, the maximum overshoot is used to evaluate the ability to suppress airflow fluctuations under motion conditions; the convergence time is used to evaluate the dynamic response speed of the system; and the output variance is used to evaluate the stability of long-term regulation. Step 2: Construct a weighted comprehensive cost function. Considering the usage characteristics of respiratory protective equipment, assign fixed optimization weights to the three performance indicators, prioritizing low overshoot and high stability, to construct a quantifiable multi-objective cost function. In the formula, the sum of the three weighting coefficients is 1, which can be freely adapted and adjusted. They correspond to the overshoot suppression priority, response speed priority, and output stability priority, respectively. The parameters are determined by multi-condition traversal simulation and optimization to adapt to various human movement and breathing regulation scenarios. Step 3: Parameter iterative optimization rule, with minimizing the numerical value of the cost function as the optimization objective, i.e., satisfying... The system performs reverse iterative correction on the three core parameters of the fuzzy rule base, including the activation weight, Gaussian membership center value, and Gaussian membership bandwidth of each fuzzy rule; after each round of working condition sampling, the system automatically calculates the overall control cost function value corresponding to the current rule base. Step 4: Parameter correction and rule update. If the current cost function value is greater than the cost function value of the previous iteration, it is determined that the current fuzzy rule parameters are not optimal. The system automatically fine-tunes the membership center, bandwidth, and rule weights to reduce system overshoot, shorten convergence time, and weaken output fluctuations. If the cost function value continues to converge to a stable minimum value, it is determined that the current fuzzy rule parameters are optimal, and the set of rule parameters is fixed to complete this round of optimization update. Step 5: Global rule base refresh. After completing the single-round parameter optimization, the system automatically replaces the original fuzzy rule parameters, realizing adaptive iterative upgrades of the rule base. This ensures that the damping compensation coefficient output by fuzzy inference always matches the real-time operating condition disturbance characteristics, guaranteeing control accuracy and dynamic stability under different motion intensities and breathing states. ④ Adaptive Gaussian membership degree and weighted centroid defuzzification Gaussian membership degree: Multidimensional coupled Mamdani inference, weighted mental method to solve for damping compensation coefficient ,constraint : .
[0013] Preferably, the discretization of the second-order damping model is as follows: The standard discretization derivation of the second-order damped dynamic model in the continuous domain is completed, with a sampling period of . The complete difference definition and iterative formula are as follows: First-order difference rate: Second-order curvature difference: Complete iterative model for second-order damped dynamic compensation in discrete domain: = In the formula: The final output control component of the dynamic channel is specifically designed to counteract system oscillations and adjustment lags caused by high-frequency motion impacts and attitude switching through first-order differential jitter suppression and second-order curvature prediction of disturbance abrupt changes.
[0014] Preferably, the fusion of operating condition disturbance quantization and adaptive weights specifically includes: 1) Quantitative indicators of multi-dimensional operating condition disturbance intensity By integrating three core parameters—triaxial composite acceleration, attitude rotation angular velocity, and airflow velocity deviation—a multi-dimensional perturbation quantification model is constructed to accurately quantify the degree of perturbation under different motion conditions. The specific calculation formula is as follows: In the formula: The composite acceleration of the three axes of human motion characterizes the impact disturbance of limb movement; The angular velocity of the helmet's attitude rotation represents the attitude deflection vibration disturbance; The deviation in respiratory airflow velocity characterizes the intrinsic respiratory airflow disturbance, and the weighting coefficient is set to [value missing]. The coefficients sum to 1. After multi-condition simulation calibration, it is primarily adapted to complex motion scenarios such as running and posture switching. The calculated values are... This is a dimensionless disturbance intensity coefficient; the larger the value, the stronger the overall disturbance experienced by the system. 2) Continuous nonlinear saturation weight iteration An exponentially saturated nonlinear weighted iterative model is constructed, and the fused weights are continuously and smoothly updated according to the real-time disturbance intensity to achieve seamless switching of operating conditions. The specific iterative formula is as follows: In the formula: This is the weight adjustment coefficient, used to control the sensitivity of the weight response. The smaller the value, the smoother the weight change; the larger the value, the faster the weight response. Under low-disturbance steady-state conditions, Approaching 0, fusion weight When the value approaches 0, the system primarily operates with steady-state precision control; under high-intensity motion disturbance conditions, Significantly increased, When the value approaches 1, the system prioritizes using dynamic damping compensation to suppress disturbances; 3) Dual-channel adaptive weighted global fusion control model Based on adaptively updated nonlinear weights, the steady-state control components and dynamic compensation components are weighted and fused to address the limitation of a single control channel in adapting to all operating conditions. The optimal control quantity that adapts to real-time operating conditions is then solved, and the fusion model is as follows: In the formula: This is the final closed-loop control quantity for a single sampling period; The weighting of steady-state control components is used to ensure the accuracy of equipment reference control and eliminate static deviations. The dynamic compensation component is weighted to offset real-time motion disturbances and suppress system overshoot oscillations.
[0015] Preferably, the weighted fusion standardization step involves adaptively weighting and fusing the steady-state control components and the dynamic compensation control components based on the real-time intensity of human motion disturbances. The specific implementation steps are as follows: Step 1: Data preprocessing. The system collects three types of disturbance data in real time: human motion acceleration, posture angular velocity, and respiratory airflow fluctuation deviation. After filtering and outlier removal, the data is normalized to eliminate dimensional interference and obtain standard quantitative disturbance characteristic data. Step 2: Disturbance intensity quantification. Substitute the preprocessed disturbance features into the quantification formula, combine them with preset weight coefficients, integrate the multi-dimensional disturbance effects, and calculate the overall disturbance intensity coefficient of the current working condition in real time to accurately distinguish between steady state, weak disturbance, strong disturbance and other working conditions. Step 3: Adaptive weight update. Based on the real-time disturbance intensity, continuously smooth fusion weights are generated iteratively. Under low disturbance conditions, the weights are biased towards the steady-state control channel to ensure control accuracy. Under strong disturbance conditions, the weights are biased towards the dynamic compensation channel to suppress system oscillations. There are no weight jumps during the switching of operating conditions. Step 4: Control component retrieval and matching. The steady-state control component output from the steady-state channel and the dynamic compensation component output from the dynamic channel are read in real time. Both control quantities have undergone stability correction, providing a reliable input basis for weight fusion. Step 5: Weighted fusion of control quantities. Through an adaptive weighted fusion model, the two control components are weighted and superimposed to dynamically balance steady-state accuracy and dynamic disturbance rejection performance, and to solve for the optimal closed-loop control quantity that adapts to the current real-time operating conditions. Step Six: Output Smoothing and Limiting Processing. The amplitude of the fused final control quantity is limited to match the output limit of the hardware device. At the same time, timing smoothing processing is performed to ensure continuous and stable control output. Finally, an effective control signal is output to drive the actuator and complete single-cycle closed-loop control.
[0016] The technical effects of this invention are as follows: This invention decouples dynamic and static errors through orthogonal matrices, achieving independent dual-channel control. This effectively reduces steady-state pressure deviation and airflow fluctuations during operation, balancing equipment operating accuracy and disturbance rejection. Differential exponential constraints are used to correct PID parameters, avoiding parameter drift and fine-tuning oscillations, ensuring long-term stable system operation. Simultaneously, data-driven optimization of fuzzy rules eliminates the need for manual adjustments, significantly reducing the workload of mass production and maintenance. A second-order damping compensation structure suppresses high-frequency disturbances, shortening control time, reducing overshoot, and improving wearing comfort. Adaptive weighting smoothly adapts to different operating conditions, eliminating airflow impact during condition switching, and the algorithm is suitable for various applications. Ordinary embedded microcontrollers offer greater applicability and a higher level of product intelligence. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of a dual-fuzzy decoupling adaptive weight fusion control method for smart breathing helmets. Detailed Implementation
[0018] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings. The description in this part is only exemplary and explanatory, and should not be used to limit the scope of protection of the present invention in any way.
[0019] like Figure 1 As shown, this application presents a dual-fuzzy decoupling adaptive weight fusion control method for smart breathing helmets.
[0020] This invention adopts a closed-loop control architecture that combines hardware and software, and is divided into three layers: data sensing, algorithm regulation, and control output. Each layer works together to achieve adaptive and precise regulation of respiratory airflow. The overall architecture functions are as follows: (1) Data perception layer The data perception layer is mainly responsible for the real-time acquisition of equipment operation data and human motion condition data, covering respiratory steady-state parameters and motion disturbance parameters. The acquired raw signals undergo preprocessing operations such as filtering and noise reduction, anomaly removal, and normalization to be unified into a standardized data format before being transmitted to the algorithm control layer to provide reliable data support for subsequent control operations.
[0021] (2) Core Algorithm Control Layer The core algorithm control layer is the control core of the system. It is equipped with the self-developed hierarchical control algorithm and adaptive optimization program of this invention. It has built-in core functions such as error decoupling, parameter iterative correction, dynamic damping compensation, rule base update, and weight fusion. It can perform hierarchical calculations on input data, process the steady-state deviation and dynamic motion disturbance of the system respectively, adaptively optimize control parameters and fuse and output the optimal digital control quantity.
[0022] (3) Control output layer The control output layer is responsible for receiving control commands from the algorithm regulation layer, completing signal conversion and power drive, adjusting the airflow output state of the device in real time, adapting to different human movement conditions and breathing rhythms, stabilizing the airway pressure and airflow inside the helmet, and forming a complete closed-loop control.
[0023] Core Adaptive Fusion Control Algorithm This invention combines operating condition data with a multi-objective optimization approach to establish two independent fuzzy rule bases: steady-state and dynamic. The system sampling period is set to 0.02s to adapt to human breathing rhythm and sensor response characteristics. The steady-state fuzzy rule base is optimized based on stable breathing conditions, using parameters such as steady-state error and cumulative error to correct PID parameters, eliminating static errors caused by equipment reference offsets and ensuring steady-state operating accuracy. The dynamic fuzzy rule base is iteratively generated based on various motion conditions, and can be automatically updated and redundant rules simplified. It solves for damping compensation coefficients based on motion state parameters, effectively suppressing system oscillations and control lags caused by motion and attitude switching. Both rule bases can be called and iterated in real time on embedded devices.
[0024] Dual fuzzy decoupling error modeling System synthesis error and discrete PID control model set up The target respiratory parameters (target airway pressure, target airflow rate) are given at time k. Define the system's real-time integrated error as the actual output parameters of the system: It includes two coupled components: slow-varying steady-state deviation of respiration and fast-varying disturbance deviation of limb movement; this invention constructs an incremental discrete PID control framework, mathematically described as follows: In the formula: These are the basic reference parameters for the PID controller; The current PID control increment; the total system control output: Scalar decomposition of dynamic and static errors The overall error is decomposed into two types of error components with completely independent physical properties: Steady-state error, caused by equipment reference deviation, static offset of respiratory rhythm, and parameter mismatch, determines the long-term steady-state accuracy of the system. Dynamic disturbance error, caused by limb swinging, posture switching, and running impact, can easily lead to system overshoot, oscillation, and adjustment lag.
[0025] Two-dimensional error state space orthogonal decoupling model Extend the scalar error into a three-dimensional error state vector. Characterize the instantaneous error, the rate of change of error, and the cumulative error respectively; construct two sets of third-order orthogonal transformation matrices. Satisfying orthogonal constraints This achieves decoupled separation of the error subspace, mathematically described as follows: Matrix construction rules: This is the low-pass filter transformation matrix, with weights biased towards the cumulative error term, to extract steady-state offset features; The high-pass filter transform matrix is used to extract perturbation abrupt change features by first-order and second-order difference of the weight bias error; where, The specific values are obtained through least squares regression fitting driven by operating condition data.
[0026] This section is the core decoupling step before the overall control algorithm, completing the hierarchical decomposition of dynamic and static coupling errors, and finally outputting two sets of independent error state vectors without coupling interference. The decoupled... The steady-state error state vector serves as the input to the subsequent steady-state channel high-order exponential iterative adaptive correction control strategy, used for steady-state correction of PID parameters and elimination of long-term steady-state error; after decoupling... The dynamic disturbance error state vector serves as the input to the subsequent dynamic channel second-order damping adaptive compensation control strategy. It is used for high-frequency motion disturbance identification and dynamic overshoot suppression, providing the necessary preliminary data foundation for dual-channel hierarchical independent control and subsequent adaptive correction and compensation calculations.
[0027] Steady-state channel higher-order exponential iterative adaptive correction control strategy Partial-parameter exponential convergence correction model Based on Lyapunov stability theory, an exponentially saturated smoothing map is constructed to constrain the unbounded fuzzy original increment into a continuous, bounded, and effective correction increment: : Parameter increment after smoothing correction; : Fuzzy inference original increment; : Corresponding dimension error state quantity; The global convergence smoothing coefficient needs to be determined jointly by considering the mathematical properties of the constraint function, system simulation traversal, and physical prototype calibration; it is a range value. At that time, the convergence speed is too slow, and the correction is delayed in the large error condition. At that time, the correction range is drastic, and small oscillations are easily generated under small errors.
[0028] Model characteristics: The constraint coefficient for large error conditions approaches 1, quickly eliminating steady-state error; the constraint coefficient for small error conditions adaptively decays, suppressing fine-tuning oscillations; the exponential function is continuously differentiable over the entire domain, with no abrupt breakpoints in parameter increments.
[0029] PID three-parameter differential high-order correction formula Since the proportional, integral, and differential parameters correspond to different physical meanings of errors, this invention constructs three sets of differentiated exponential convergence correction models to adapt to the constraint correction requirements of instantaneous error, cumulative error, and error rate of change, respectively. The mathematical description is as follows: In the formula: These represent the increments of the original parameters for the proportional, integral, and differential values of the fuzzy inference output; (Left side) It is an effective correction increment after exponential constraint smoothing, which converges and is bounded throughout the process with no abrupt changes in time.
[0030] The real-time control parameters are updated iteratively by combining traditional baseline PID parameters. The baseline parameters are based on a general engineering range. , , The embodiments of the present invention are calibrated. Iterative parameter updates to the complete model: System stability constraints and convergence characteristics analysis ① Construction of positive definite energy function: Constantly satisfied ,only hour It conforms to the discrete Lyapunov positive definite function specification.
[0031] ② Monotonic convergence characteristic: After exponential constraint correction, the PID increment is bounded and continuous, with no parameter drift or jumps; under any non-zero steady-state error, the error continuously decays after iteration, satisfying: ③ Negative definiteness criterion for difference: Lyapunov first-order difference ; When steady-state error exists The system energy continues to decrease; after the error returns to zero... The system maintains steady-state equilibrium without oscillations or deviations.
[0032] ④ Stability enhancement: The exponential constraint operator provides global smoothing and limiting, which mathematically constrains the parameter iteration amplitude and rate of change, ensuring asymptotic stability under all operating conditions.
[0033] Steady-state channel standardization iterative process Step 1: With Periodically collect airway pressure and airflow rate Combined with target parameters Solve for instantaneous error, statistically analyze cumulative error and error rate of change, and construct multidimensional steady-state error characteristics; Step 2: The steady-state fuzzy rule base completes the Mamdani fuzzy inference, and the centroid method is used to defuzzify and output the unconstrained PID original correction increment; Step 3: Substitute The exponential nonlinear constraint correction is completed in proportion, integral, and differential dimensions; Step 4: Iteratively update the real-time optimal parameters based on the factory baseline parameters. ; Step 5: Use the Lyapunov criterion to verify the error convergence characteristics and continuously constrain the parameter correction magnitude; Step 6: Substitute the updated PID parameters into the incremental PID controller to solve for the steady-state control components. It is then transmitted to the weight fusion unit.
[0034] This section outlines the execution phase for optimizing the system's steady-state accuracy. It utilizes the steady-state error state vector obtained from the pre-decoupling process to perform adaptive parameter iterative correction. The final output is a stable, drift-free steady-state control component. This output serves as one of the core inputs for the subsequent adaptive weighted fusion process. It is specifically used to compensate for equipment static deviations, eliminate long-term steady-state static errors, ensure the control accuracy under stable operating conditions, and provide a steady-state reference control basis for subsequent weighted fusion of dynamic and static control components and the solution of the optimal control quantity in the entire domain.
[0035] Dynamic channel second-order damped adaptive compensation control strategy Complete Modeling and Parameter Determination of Second-Order Underdamped Continuous Dynamics Continuous domain dynamic compensation benchmark model: Parameter definition and preferred values: The system's natural angular frequency, based on human motion and hardware constraints, ranges from [value range missing]. ; Damping ratio, the preferred critical damping ratio in engineering applications. Suppress overshoot; Incremental smoothing correction coefficient, optimized within a certain range based on human movement and hardware constraints. Suppressing instantaneous jumps in control input; Second-order damping enhancement coefficient, adapted to human movement and hardware constraints, within a certain range. This counteracts curvature-type abrupt disturbances.
[0036] The left side of the formula is the second-order damped inherent amplitude term; the right side is the disturbance advance compensation driving term, which uses the second-order differential term to predict the curvature of the disturbance change, and the first-order differential term to smooth the output jitter.
[0037] Dynamic perturbation feature preprocessing Four types of disturbance characteristics were selected: respiratory rate deviation, human body triaxial composite acceleration (units). ), attitude rotation angular velocity (unit) ), airflow velocity deviation (unit) The original data is filtered by moving mean, 3σ outlier removal, and linearly normalized to the standard domain of [0,1] to eliminate dimensional differences.
[0038] Adaptive fuzzy rule-based end-to-end data-driven optimization ①Incremental generation of dataset A full-gradient working condition sample set is constructed, covering five typical working conditions in actual use of smart breathing helmets: stationary, low-speed walking, high-speed running, posture deflection, and strenuous exercise. Multi-dimensional sensor data samples are collected evenly for each working condition. The sample dimensions include airway pressure error, airflow error, human body composite acceleration, posture angular velocity, and airflow fluctuation deviation. This invention is based on a distance threshold-based incremental generation mechanism to achieve unsupervised automatic addition of fuzzy rules without the need for manual preset of the number of rules.
[0039] First, all collected raw operating condition samples were normalized to map the multi-dimensional heterogeneous data to a standard domain. Eliminating dimensional differences yields a standardized sample vector: In the formula: For the first A standardized working condition sample vector is generated, with each dimension corresponding to pressure error, flow error, motion acceleration, and attitude angular velocity characteristics, respectively.
[0040] Calculate the Euclidean spatial distance between the new sample and the existing cluster centers. The distance calculation formula is as follows: In the formula: For newly added standardized samples collected in real time, For each cluster center sample already stored in the rule base; set a fixed clustering threshold. This serves as a criterion for generating new rules.
[0041] Incremental rule generation judgment logic: When the minimum Euclidean distance between a new sample and all existing cluster centers is greater than or equal to a set threshold, it is judged as a completely new working condition feature sample, and a new fuzzy rule is automatically generated; otherwise, it is judged as a sample of the same type as an existing rule, and no new rule is added, only the corresponding cluster center parameters are updated. Complete judgment formula: In the formula: This represents the addition of a fuzzy rule. This means no new rules will be added.
[0042] The newly added rule completes the initialization of membership parameters simultaneously, updates the Gaussian membership center value and bandwidth with the new sample as the center, and automatically incorporates it into the fuzzy rule base, realizing continuous incremental iteration of working condition data and adaptive expansion of the rule base to cover boundary extreme working conditions.
[0043] ② Rule similarity pruning Combined similarity of the two rules: Membership center, :bandwidth, Output mapping value; weighting coefficient It can be freely adapted and adjusted, with a coefficient sum of 1; pruning threshold , Eliminate redundant rules with low activation and high error, among which Based on experience, this value is adjustable; long-term inactive dormant rules are deleted synchronously, resulting in a lightweight rule library.
[0044] ③ Adaptive selection of multi-objective cost function This invention constructs a multi-objective cost function that considers dynamic overshoot, convergence speed, and output stability. It then performs global optimal iteration on the membership centers, membership bandwidth, and rule output weights of the fuzzy rule base to achieve adaptive optimal updating of fuzzy control parameters. The specific implementation steps are as follows: Step 1: Determine the control performance evaluation indicators. Based on the control requirements of the intelligent breathing helmet, three core quantitative performance indicators were selected as the optimization basis: the system's maximum overshoot. System adjustment of convergence time Control output fluctuation variance Among them, the maximum overshoot is used to evaluate the ability to suppress airflow fluctuations under motion conditions; the convergence time is used to evaluate the dynamic response speed of the system; and the output variance is used to evaluate the stability of long-term control.
[0045] Step 2: Construct a weighted comprehensive cost function. Considering the usage characteristics of respiratory protective equipment, fixed optimization weights are assigned to the three performance indicators, prioritizing low overshoot and high stability, to construct a quantifiable multi-objective cost function: In the formula, the sum of the three weighting coefficients is 1, which can be freely adjusted to correspond to overshoot suppression priority, response speed priority, and output stability priority, respectively. The parameters are determined by multi-condition traversal simulation and optimization to adapt to various human movement and breathing regulation scenarios.
[0046] Step 3: Parameter Iteration Optimization Rules. The optimization objective is to minimize the numerical value of the cost function, i.e., satisfy... The system performs reverse iterative correction on three core parameters of the fuzzy rule base, including the activation weight, Gaussian membership center value, and Gaussian membership bandwidth of each fuzzy rule. After each round of operating condition sampling, the system automatically calculates the overall control cost function value corresponding to the current rule base.
[0047] Step 4: Parameter Correction and Rule Update. If the current cost function value is greater than the cost function value of the previous iteration, the current fuzzy rule parameters are determined to be non-optimal. The system automatically fine-tunes the membership centers, bandwidth, and rule weights to reduce system overshoot, shorten convergence time, and weaken output fluctuations. If the cost function value continues to converge to a stable minimum value, the current fuzzy rule parameters are determined to be optimal, and the set of rule parameters is solidified, completing this round of optimization and update.
[0048] Step 5: Global rule base refresh. After completing the single-round parameter optimization, the system automatically replaces the original fuzzy rule parameters, realizing adaptive iterative upgrades of the rule base. This ensures that the damping compensation coefficient output by fuzzy inference always matches the real-time operating condition disturbance characteristics, guaranteeing control accuracy and dynamic stability under different motion intensities and breathing states.
[0049] This mechanism enables fully data-driven automatic selection of fuzzy rules, eliminating the need for manual parameter adjustment, avoiding subjective errors caused by human experience, and significantly improving the adaptability and robustness of the control system under complex boundary conditions.
[0050] ④ Adaptive Gaussian membership degree and weighted centroid defuzzification Gaussian membership degree: Multidimensional coupled Mamdani inference, weighted mental method to solve for damping compensation coefficient ,constraint : Discretization of the second-order damping model To adapt to embedded discrete sampling computation scenarios, a standard discretization derivation of the second-order damped dynamic model in the continuous domain is performed, with a sampling period of [missing information]. The complete difference definition and iterative formula are as follows: First-order difference rate: Second-order curvature difference: Complete iterative model for second-order damped dynamic compensation in discrete domain: = In the formula: The final output control component of the dynamic channel is specifically designed to counteract system oscillations and adjustment lags caused by high-frequency motion impacts and attitude switching through first-order differential jitter suppression and second-order curvature prediction of disturbance abrupt changes.
[0051] The final output control component of the dynamic channel is specifically designed to offset the oscillation lag caused by high-frequency motion impacts and attitude switching.
[0052] This section is the core execution step of the system's dynamic disturbance rejection optimization. It relies on the dynamic disturbance error state vector obtained from the pre-decoupling, combined with a data-driven adaptive fuzzy rule mechanism and a second-order underdamped dynamic discrete compensation algorithm. The final output is a dynamic compensation control component with high response and low oscillation. This output serves as another core input for the subsequent adaptive weight fusion stage. It is specifically used to suppress high-frequency airflow disturbances and system overshoot lag caused by human movement and posture deflection, make up for the lack of dynamic response in the steady-state channel, and provide a dynamic anti-disturbance control basis for the fusion of dynamic and static dual-path control components and the realization of adaptive and precise regulation under all operating conditions.
[0053] Fusion Design of Operating Condition Disturbance Quantization and Adaptive Weights Multi-dimensional operating condition disturbance intensity quantification index When a person wears a breathing helmet for exercise, limb movement, posture rotation, and changes in breathing rhythm will generate coupled airflow disturbances. A single detection parameter cannot accurately determine the intensity of the disturbance under different working conditions. This invention integrates three core parameters—triaxial composite acceleration, posture rotation angular velocity, and airflow velocity deviation—to construct a multi-dimensional disturbance quantification model, accurately quantifying the degree of disturbance under different motion conditions. The specific calculation formula is as follows: In the formula: The composite acceleration of the three axes of human motion characterizes the impact disturbance of limb movement; The angular velocity of the helmet's attitude rotation represents the attitude deflection vibration disturbance; The deviation in respiratory airflow velocity characterizes the intrinsic respiratory airflow disturbance. The weighting coefficients are set to... The coefficients sum to 1, and after multi-condition simulation calibration, it is primarily adapted to complex motion scenarios such as running and posture switching. The calculated values are... This is a dimensionless disturbance intensity coefficient; the larger the value, the stronger the overall disturbance experienced by the system.
[0054] Continuous nonlinear saturation weight iteration This invention constructs an exponentially saturated nonlinear weighted iterative model, which can continuously and smoothly update the fused weights according to the real-time disturbance intensity, achieving seamless switching of operating conditions. The specific iterative formula is as follows: In the formula: This is the weight adjustment coefficient, used to control the sensitivity of the weight response. The smaller the value, the smoother the weight change; the larger the value, the faster the weight response. Under low-disturbance steady-state conditions, Approaching 0, fusion weight When the value approaches 0, the system primarily operates with steady-state precision control; under high-intensity motion disturbance conditions, Significantly increased, When the value approaches 1, the system prioritizes dynamic damping compensation to suppress disturbances. This weight is continuously differentiable throughout the entire process, completely avoiding the control abrupt change problem caused by piecewise weights.
[0055] Dual-channel adaptive weighted global fusion control model To balance system steady-state accuracy and dynamic disturbance rejection performance, this invention uses adaptively updated nonlinear weights to weight and fuse steady-state control components and dynamic compensation components. This addresses the limitation of a single control channel in adapting to all operating conditions, and solves for the optimal control quantity adapted to real-time operating conditions. The fusion model is as follows: In the formula: This is the final closed-loop control quantity for a single sampling period; The weighting of steady-state control components is used to ensure the accuracy of equipment reference control and eliminate static deviations. The dynamic compensation component is weighted to offset real-time motion disturbances and suppress system overshoot oscillations. This model can automatically adjust the two control components according to the intensity of the disturbance under operating conditions. Under stable operating conditions, it ensures high-precision voltage and current stabilization, and under complex disturbance conditions, it enhances the ability to resist disturbances and suppress vibrations. The two control components complement and cooperate to achieve stable, accurate, and lag-free adaptive regulation under all operating conditions.
[0056] Weight fusion standardization steps This section is the core aggregation unit for the system's full-condition control. Based on the real-time intensity of human motion disturbances, it adaptively weights and fuses the steady-state control components and the dynamic compensation control components. This solves the problems of traditional fixed weights failing to adapt to condition switching and poor integration of dynamic and static control performance. The specific implementation steps are as follows: Step 1: Data Preprocessing. The system collects three types of disturbance data in real time: human motion acceleration, posture angular velocity, and respiratory airflow fluctuation deviation. After filtering and outlier removal, the data is normalized to eliminate dimensional interference and obtain standard quantified disturbance characteristic data.
[0057] Step 2: Disturbance Intensity Quantification. Substitute the preprocessed disturbance characteristics into the quantification formula, combine them with preset weighting coefficients, integrate the influence of multi-dimensional disturbances, and calculate the overall disturbance intensity coefficient of the current operating condition in real time to accurately distinguish between steady-state, weak disturbance, and strong disturbance operating conditions.
[0058] Step 3: Adaptive Weight Update. A continuous and smooth fusion weight is generated iteratively based on the real-time disturbance intensity. Under low-disturbance conditions, the weights are biased towards the steady-state control channel to ensure control accuracy; under strong-disturbance conditions, the weights are biased towards the dynamic compensation channel to suppress system oscillations. There are no weight jumps during the condition switching process.
[0059] Step 4: Control Component Retrieval and Matching. The steady-state control components from the steady-state channel output and the dynamic compensation components from the dynamic channel output are read in real time. Both control quantities have undergone stability correction, providing a reliable input basis for weighted fusion.
[0060] Step 5: Weighted Fusion of Control Components. Using an adaptive weighted fusion model, the two control components are weighted and superimposed to dynamically balance steady-state accuracy and dynamic disturbance rejection performance, thus solving for the optimal closed-loop control quantity that best suits the current real-time operating conditions.
[0061] Step Six: Output Smoothing and Limiting. The amplitude of the fused final control quantity is limited to match the output limits of the hardware. Simultaneously, timing smoothing is performed to ensure continuous and stable control output. The final output is an effective control signal that drives the actuator, completing single-cycle closed-loop control.
[0062] This section is the final output summary of the entire control algorithm. Following the steady-state control component and dynamic compensation control component mentioned earlier, it integrates real-time operating condition disturbance quantization and adaptive nonlinear weighting to finally output the globally optimal control quantity U(k) that balances steady-state accuracy and dynamic disturbance rejection performance. This section solves the problem that traditional fixed weights cannot adapt to multiple operating condition switching and that dynamic and static control performances are mutually constrained. It can automatically allocate the two control components according to different human motion states to achieve high-precision voltage and current stabilization under stable operating conditions and adaptive regulation to suppress vibration and disturbances under strong disturbance conditions, providing the final output guarantee for stable, accurate, and lag-free closed-loop control of the system under all operating conditions.
[0063] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0064] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be noted that due to the limitations of textual expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of the present invention.
Claims
1. A dual-fuzzy decoupling adaptive weighted fusion control method for smart breathing helmets, characterized in that, Includes the following steps: S1. Data Sensing Layer Acquisition: The data sensing layer collects equipment operation data and human motion condition data in real time, covering respiratory steady-state parameters and motion disturbance parameters. The collected raw signals undergo preprocessing operations such as filtering and noise reduction, anomaly removal, and normalization to be unified into a standardized data format. S2. Core algorithm control layer processing: The core algorithm control layer is equipped with a hierarchical control algorithm and an adaptive optimization program. It has built-in functions such as error decoupling, parameter iterative correction, dynamic damping compensation, rule base update, and weight fusion. It performs hierarchical calculations on the data input in step one, processes the steady-state deviation and dynamic motion disturbance of the system respectively, adaptively optimizes the control parameters, and fuses the output of the optimal digital control quantity. S3. The control output layer executes the control command output by the algorithm control layer in step two, completes signal conversion and power drive, adjusts the airflow output state of the device in real time, adapts to different human movement conditions and breathing rhythms, stabilizes the airway pressure and airflow in the helmet, and forms a complete closed-loop control.
2. The dual-fuzzy decoupling adaptive weight fusion control method for intelligent breathing helmets according to claim 1, characterized in that, The core algorithm control layer includes dual-fuzzy decoupling error modeling, steady-state channel high-order exponential iterative adaptive correction control strategy, dynamic channel second-order damping adaptive compensation control strategy, and operating condition disturbance quantification and adaptive weight fusion.
3. The dual-fuzzy decoupling adaptive weight fusion control method for intelligent breathing helmets according to claim 2, characterized in that, The specific steps for modeling the double fuzzy decoupling error are as follows: set up Let k be the target baseline parameters for respiration. Define the system's real-time integrated error as the actual output parameters of the system: It includes two coupled components: slow-varying steady-state deviation of respiration and fast-varying disturbance deviation of limb movement. This invention constructs an incremental discrete PID control framework, mathematically described as follows: In the formula: These are the basic reference parameters for the PID controller; The current PID control increment; the total system control output: The overall error is decomposed into two types of error components with completely independent physical properties: Steady-state error, caused by equipment reference deviation, static offset of respiratory rhythm, and parameter mismatch, determines the long-term steady-state accuracy of the system. Dynamic disturbance error, caused by limb swinging, posture switching, and running impact, can easily lead to system overshoot, oscillation, and adjustment lag. Extend the scalar error into a three-dimensional error state vector. Characterize the instantaneous error, the rate of change of error, and the cumulative error respectively; construct two sets of third-order orthogonal transformation matrices. Satisfying orthogonal constraints This achieves decoupled separation of the error subspace, mathematically described as follows: Matrix construction rules: This is a low-pass filter transformation matrix, with weights biased towards the cumulative error term, to extract steady-state offset features; The high-pass filter transform matrix is used to extract perturbation abrupt change features by first-order and second-order difference of the weight bias error; where, The specific values are obtained through least squares regression fitting driven by operating condition data.
4. The dual-fuzzy decoupling adaptive weight fusion control method for intelligent breathing helmets according to claim 3, characterized in that, The steady-state channel higher-order exponential iterative adaptive correction control strategy is as follows: 1) Parameter-based exponential convergence correction model Based on Lyapunov stability theory, an exponentially saturated smoothing map is constructed to constrain the unbounded fuzzy original increment into a continuous, bounded, and effective correction increment: : Parameter increment after smoothing correction; : Fuzzy inference original increment; : Corresponding dimension error state quantity; The global convergence smoothing coefficient needs to be determined jointly by considering the mathematical properties of the constraint function, system simulation traversal, and physical prototype calibration; it is a range value. At that time, the convergence speed is too slow, and the correction is delayed in the large error condition. At that time, the correction range is drastic, and small oscillations are easily generated under small errors; 2) PID three-parameter differential high-order correction formula To address the different physical meanings of errors corresponding to proportional, integral, and differential parameters, three sets of differentiated exponential convergence correction models are constructed to adapt to the constraint correction requirements of instantaneous error, cumulative error, and error rate of change, respectively. The mathematical descriptions are as follows: In the formula: These represent the increments of the original parameters for the proportional, integral, and differential values of the fuzzy inference output; (Left side) The effective correction increment is obtained after exponential constraint smoothing, and it converges and is bounded throughout the entire process with no abrupt changes in the time series. Iterative parameter updates to the complete model: 3) System stability constraints and convergence characteristic analysis ① Construction of positive definite energy function: Constantly satisfied ,only hour It conforms to the discrete Lyapunov positive definite function specification; ② Monotonic convergence characteristic: After exponential constraint correction, the PID increment is bounded and continuous, with no parameter drift or jumps; under any non-zero steady-state error, the error continuously decays after iteration, satisfying: ③ Negative definiteness criterion for difference: Lyapunov first-order difference ; When steady-state error exists The system energy continues to decrease; after the error returns to zero... The system maintains steady-state equilibrium without oscillation or deviation. ④ Stability enhancement: The exponential constraint operator provides global smoothing and limiting, which mathematically constrains the parameter iteration amplitude and rate of change, ensuring asymptotic stability under all operating conditions.
5. The dual-fuzzy decoupling adaptive weight fusion control method for intelligent breathing helmets according to claim 4, characterized in that, The steady-state channel standardization iteration process includes the following steps: Step 1: With Periodically collect airway pressure and airflow rate Combined with target parameters Solve for instantaneous error, statistically analyze cumulative error and error rate of change, and construct multidimensional steady-state error characteristics; Step 2: The steady-state fuzzy rule base completes the Mamdani fuzzy inference, and the centroid method defuzzifies the output to generate the unconstrained PID original correction increment; Step 3: Substitute The exponential nonlinear constraint correction is completed in proportion, integral, and differential dimensions; Step 4: Iteratively update the real-time optimal parameters based on the factory baseline parameters. ; Step 5: Use the Lyapunov criterion to verify the error convergence characteristics and continuously constrain the parameter correction magnitude; Step 6: Substitute the updated PID parameters into the incremental PID controller to solve for the steady-state control components. It is then transmitted to the weight fusion unit.
6. The dual-fuzzy decoupling adaptive weight fusion control method for intelligent breathing helmets according to claim 4, characterized in that, The dynamic channel second-order damping adaptive compensation control strategy specifically includes: 1) Complete Modeling and Parameter Determination of Second-Order Underdamped Continuous Dynamics Continuous domain dynamic compensation benchmark model: Parameter definition and preferred values: The system's natural angular frequency, based on human motion and hardware constraints, ranges from [value range missing]. ; Damping ratio, the preferred critical damping ratio in engineering applications. Suppress overshoot; Incremental smoothing correction coefficient, optimized within a certain range based on human movement and hardware constraints. Suppressing instantaneous jumps in control input; Second-order damping enhancement coefficient, adapted to human movement and hardware constraints, within a certain range. To counteract curvature-type abrupt disturbances; The left side of the formula is the second-order damped inherent amplitude-stabilizing term; the right side is the disturbance advance compensation driving term, which uses the second-order differential term to predict the curvature of the disturbance change, and the first-order differential term to smooth the output jitter. 2) Preprocessing of dynamic disturbance features Four types of disturbance features were selected: respiratory rate deviation, human body triaxial composite acceleration, attitude rotation angular velocity, and airflow velocity deviation. The raw data were filtered by moving mean, 3σ outlier removal, and linearly normalized to the [0,1] standard domain to eliminate dimensional differences. 3) Adaptive fuzzy rule-based end-to-end data-driven optimization 4) Discretization of the second-order damping model.
7. The dual-fuzzy decoupling adaptive weight fusion control method for intelligent breathing helmets according to claim 6, characterized in that, Adaptive fuzzy rule-based end-to-end data-driven optimization specifically includes: ①Incremental generation of dataset A full-gradient working condition sample set was constructed, covering five typical working conditions in actual use of smart breathing helmets: stationary, low-speed walking, high-speed running, posture deflection, and strenuous exercise. Multi-dimensional sensor data samples were collected evenly for each working condition. The sample dimensions include airway pressure error, airflow error, human body composite acceleration, posture angular velocity, and airflow fluctuation deviation. First, all collected raw operating condition samples were normalized to map the multi-dimensional heterogeneous data to a standard domain. Eliminating dimensional differences yields a standardized sample vector: In the formula: For the first A standardized working condition sample vector is generated, with each dimension corresponding to the characteristics of pressure error, flow error, motion acceleration, and attitude angular velocity, respectively. Calculate the Euclidean spatial distance between the new sample and the existing cluster centers. The distance calculation formula is as follows: In the formula: For newly added standardized samples collected in real time, For each cluster center sample already stored in the rule base; set a fixed clustering threshold. This serves as a criterion for generating new rules. Incremental rule generation judgment logic: When the minimum Euclidean distance between a new sample and all existing cluster centers is greater than or equal to a set threshold, it is judged as a completely new working condition feature sample, and a new fuzzy rule is automatically generated; otherwise, it is judged as a sample of the same type as an existing rule, and no new rule is added, only the corresponding cluster center parameters are updated. Complete judgment formula: In the formula: This represents the addition of a fuzzy rule. This means no new rules will be added; The newly added rule synchronously completes the initialization of membership parameters, updates the Gaussian membership center value and bandwidth with the new sample as the center, and automatically incorporates it into the fuzzy rule base, realizing continuous incremental iteration of working condition data and adaptive expansion of the rule base to cover boundary extreme working conditions; ② Rule similarity pruning Combined similarity of the two rules: Membership center, :bandwidth, Output mapping value; weighting coefficient It can be freely adapted and adjusted, with a coefficient sum of 1; pruning threshold , Eliminate redundant rules with low activation and high error, among which Based on experience values, this rule set can be adjusted; it also synchronously deletes long-term inactive dormant rules, resulting in a lightweight rule library. ③ Adaptive selection of multi-objective cost function A multi-objective cost function is constructed that takes into account dynamic overshoot, convergence speed, and output stability. Global optimization iteration is performed on the membership centers, membership bandwidth, and rule output weights of the fuzzy rule base to achieve adaptive optimal updating of fuzzy control parameters. The specific implementation steps are as follows: Step 1: Determine the control performance evaluation indicators. Based on the control requirements of the intelligent breathing helmet, three core quantitative performance indicators were selected as the optimization basis: the system's maximum overshoot. System adjustment of convergence time Control output fluctuation variance Among them, the maximum overshoot is used to evaluate the ability to suppress airflow fluctuations under motion conditions; the convergence time is used to evaluate the dynamic response speed of the system; and the output variance is used to evaluate the stability of long-term regulation. Step 2: Construct a weighted comprehensive cost function. Considering the usage characteristics of respiratory protective equipment, assign fixed optimization weights to the three performance indicators, prioritizing low overshoot and high stability, to construct a quantifiable multi-objective cost function. In the formula, the sum of the three weighting coefficients is 1, which can be freely adapted and adjusted. They correspond to the overshoot suppression priority, response speed priority, and output stability priority, respectively. The parameters are determined by multi-condition traversal simulation and optimization to adapt to various human movement and breathing regulation scenarios. Step 3: Parameter iterative optimization rule, with the optimization objective being the minimization of the cost function, i.e., satisfying... The system performs reverse iterative correction on the three core parameters of the fuzzy rule base, including the activation weight, Gaussian membership center value, and Gaussian membership bandwidth of each fuzzy rule; after each round of working condition sampling, the system automatically calculates the overall control cost function value corresponding to the current rule base. Step 4: Parameter correction and rule update. If the current cost function value is greater than the cost function value of the previous iteration, it is determined that the current fuzzy rule parameters are not optimal. The system automatically fine-tunes the membership center, bandwidth, and rule weights to reduce system overshoot, shorten convergence time, and weaken output fluctuations. If the cost function value continues to converge to a stable minimum value, it is determined that the current fuzzy rule parameters are optimal, and the set of rule parameters is fixed to complete this round of optimization update. Step 5: Global rule base refresh. After completing the single-round parameter optimization, the system automatically replaces the original fuzzy rule parameters, realizing adaptive iterative upgrades of the rule base. This ensures that the damping compensation coefficient output by fuzzy inference always matches the real-time operating condition disturbance characteristics, guaranteeing control accuracy and dynamic stability under different motion intensities and breathing states. ④ Adaptive Gaussian membership degree and weighted centroid defuzzification Gaussian membership degree: Multidimensional coupled Mamdani inference, weighted mental method to solve for damping compensation coefficient ,constraint : 。 8. The dual-fuzzy decoupling adaptive weight fusion control method for intelligent breathing helmets according to claim 6, characterized in that, The discretization of the second-order damping model is specifically as follows: The standard discretization derivation of the second-order damped dynamic model in the continuous domain is completed, with a sampling period of . The complete difference definition and iterative formula are as follows: First-order difference rate: Second-order curvature difference: Complete iterative model for second-order damped dynamic compensation in discrete domain: = In the formula: The final output control component of the dynamic channel is specifically designed to counteract system oscillations and adjustment lags caused by high-frequency motion impacts and attitude switching through first-order differential jitter suppression and second-order curvature prediction of disturbance abrupt changes.
9. The dual-fuzzy decoupling adaptive weight fusion control method for intelligent breathing helmets according to claim 7, characterized in that, The quantization of operating condition disturbances and the fusion of adaptive weights specifically include: 1) Quantitative indicators of multi-dimensional operating condition disturbance intensity By integrating three core parameters—triaxial composite acceleration, attitude rotation angular velocity, and airflow velocity deviation—a multi-dimensional perturbation quantification model is constructed to accurately quantify the degree of perturbation under different motion conditions. The specific calculation formula is as follows: In the formula: The composite acceleration of the three axes of human motion characterizes the impact disturbance of limb movement; The angular velocity of the helmet's attitude rotation represents the attitude deflection vibration disturbance; The deviation in respiratory airflow velocity characterizes the intrinsic respiratory airflow disturbance, and the weighting coefficient is set to [value missing]. The coefficients sum to 1. After multi-condition simulation calibration, it is primarily adapted to complex motion scenarios such as running and posture switching. The calculated values are... This is a dimensionless disturbance intensity coefficient; the larger the value, the stronger the overall disturbance experienced by the system. 2) Continuous nonlinear saturation weight iteration An exponentially saturated nonlinear weighted iterative model is constructed, and the fused weights are continuously and smoothly updated according to the real-time disturbance intensity to achieve seamless switching of operating conditions. The specific iterative formula is as follows: In the formula: This is the weight adjustment coefficient, used to control the sensitivity of the weight response. The smaller the value, the smoother the weight change; the larger the value, the faster the weight response. Under low-disturbance steady-state conditions, Approaching 0, fusion weight When the value approaches 0, the system primarily operates with steady-state precision control; under high-intensity motion disturbance conditions, Significantly increased, When the value approaches 1, the system prioritizes using dynamic damping compensation to suppress disturbances; 3) Dual-channel adaptive weighted global fusion control model Based on adaptively updated nonlinear weights, the steady-state control components and dynamic compensation components are weighted and fused to address the limitation of a single control channel in adapting to all operating conditions. The optimal control quantity that adapts to real-time operating conditions is then solved, and the fusion model is as follows: In the formula: This is the final closed-loop control quantity for a single sampling period; The weighting of steady-state control components is used to ensure the accuracy of equipment reference control and eliminate static deviations. The dynamic compensation component is weighted to offset real-time motion disturbances and suppress system overshoot oscillations.
10. The dual-fuzzy decoupling adaptive weight fusion control method for intelligent breathing helmets according to claim 9, characterized in that, The weighted fusion standardization process involves adaptively weighting and fusing the steady-state control components and the dynamic compensation control components based on the real-time intensity of human motion disturbances. The specific implementation steps are as follows: Step 1: Data preprocessing. The system collects three types of disturbance data in real time: human motion acceleration, posture angular velocity, and respiratory airflow fluctuation deviation. After filtering and outlier removal, the data is normalized to eliminate dimensional interference and obtain standard quantitative disturbance characteristic data. Step 2: Disturbance intensity quantification. Substitute the preprocessed disturbance features into the quantification formula, combine them with preset weight coefficients, integrate the multi-dimensional disturbance effects, and calculate the overall disturbance intensity coefficient of the current working condition in real time to accurately distinguish between steady state, weak disturbance, strong disturbance and other working conditions. Step 3: Adaptive weight update. Based on the real-time disturbance intensity, continuously smooth fusion weights are generated iteratively. Under low disturbance conditions, the weights are biased towards the steady-state control channel to ensure control accuracy. Under strong disturbance conditions, the weights are biased towards the dynamic compensation channel to suppress system oscillations, and there is no weight jump during the switching of operating conditions. Step 4: Control component retrieval and matching. The steady-state control component output from the steady-state channel and the dynamic compensation component output from the dynamic channel are read in real time. Both control quantities have undergone stability correction, providing a reliable input basis for weight fusion. Step 5: Weighted fusion of control quantities. Through an adaptive weighted fusion model, the two control components are weighted and superimposed to dynamically balance steady-state accuracy and dynamic disturbance rejection performance, and to solve for the optimal closed-loop control quantity that adapts to the current real-time operating conditions. Step Six: Output Smoothing and Limiting Processing. The amplitude of the fused final control quantity is limited to match the output limit of the hardware device. At the same time, timing smoothing processing is performed to ensure continuous and stable control output. Finally, an effective control signal is output to drive the actuator and complete single-cycle closed-loop control.
