Pressure control system and method for air-oxygen mixers

Abstract

The present application belongs to the technical field of mixed gas pressure control, and particularly relates to a pressure control system and method for an air-oxygen mixer, the system comprising a data acquisition module, a calculation module, an interference identification module, an adaptive control module, a dual-parameter coordination module and an execution module; the method detects and quantifies the interference intensity corresponding to the sudden change signal, and calculates the pressure and oxygen concentration deviation; based on this, the adaptive control module adopts a feedforward-feedback compound structure, the front end uses a model reference adaptive algorithm to generate a preliminary control instruction, the rear end uses a particle swarm optimization-radial basis neural network algorithm to perform high-precision optimization on the control parameters, and combines a dual-parameter coupling model to perform feedforward compensation to obtain a secondary correction control instruction. By intelligently identifying the interference and dynamically optimizing the control parameters, the coupling interference problem between the pressure and the oxygen concentration is effectively solved, and the control accuracy, response speed and robustness of the system under conditions such as sudden flow changes are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of mixed gas pressure control technology, specifically relating to a pressure control system and method for an air-oxygen mixer. Background Technology

[0002] Air-oxygen mixers, as core components of respiratory support equipment, are used to provide controlled oxygen concentration mixed gases for patients such as newborns and infants who breathe spontaneously. They are a key component of continuous positive airway pressure (CPAP) systems, ventilators, and other similar devices. Their core function is to precisely adjust the inhaled oxygen concentration (FiO2) according to a preset air-oxygen volume mixing ratio and maintain a stable output pressure. This provides respiratory support for patients with acute respiratory distress syndrome (RDS) or recovering children, improving oxygenation and preventing alveolar collapse. The mixing principle of the air-oxygen mixer is as follows: the supply end first distributes the mixed gas according to the set oxygen concentration by adjusting the flow ratio of the oxygen valve and the air valve. Then, a pressure regulating valve ensures a stable output of the mixed gas at the target pressure. Stable pressure is a prerequisite for accurate oxygen delivery; pressure fluctuations will indirectly cause the actual oxygen concentration to deviate from the set value. Subsequent feedback from the output adjusts the front-end control process. Traditional air-oxygen mixer pressure control schemes generally rely on PID algorithms and improved PID algorithms such as fuzzy PID. Although these algorithms have basic regulation capabilities under steady-state conditions, the operation of the air-oxygen mixer is easily affected by the interference of the gas input at the gas supply end. This includes: sudden increases in flow rate, pressure changes, or substandard purity of the air and oxygen valves at the gas supply end; blockage of the gas supply pipeline or abnormal temperature and humidity; and internal pressure imbalance caused by blockage of the equipment's exhaust port or exhaust valve failure. This can lead to sudden changes in the originally regulated input flow rate and appropriate internal air or oxygen ratio of the oxygen or air valves at the gas supply end. Moreover, when interference occurs, the system needs to simultaneously correct the flow ratio on both sides of the gas supply end. Since the change in the input gas flow rate at the gas supply end affects the pressure regulation instantaneously, it places stringent requirements on the rapid response and dynamic adaptation capabilities of the control algorithm. Traditional PID algorithms rely on fixed parameters or a limited parameter optimization space for their adjustment logic. When faced with the combined effects of multiple sources of interference or sudden large fluctuations, they are prone to problems such as adjustment lag and untimely parameter adaptation. They cannot quickly adjust the control strategy to counteract the interference, which in turn causes the air-oxygen mixer pressure to deviate from the set threshold, the mixing ratio to become inaccurate, and the control effect to be poor. Summary of the Invention

[0003] To address the aforementioned problems in the prior art, this invention provides a pressure control system and method for an air-oxygen mixer, specifically:

[0004] A pressure control system for an air-oxygen mixer includes:

[0005] The data acquisition module is used to acquire the sudden change signals of the input flow of the air valve and the sudden change signals of the oxygen valve, acquire the real-time oxygen concentration and real-time output pressure value of the mixed gas output from the mixer output port, and obtain the output pressure setpoint and the target oxygen concentration setpoint.

[0006] The calculation module calculates the pressure deviation and oxygen concentration deviation based on the real-time oxygen concentration, the real-time output pressure value, the target oxygen concentration setpoint, and the output pressure setpoint, respectively.

[0007] The interference identification module is used to extract the time-domain abrupt change characteristics of the sudden change signal of the air valve input flow and the sudden change signal of the oxygen valve flow, and to synchronously calculate the interference intensity through the instantaneous deviation amplitude quantization method.

[0008] The adaptive control module includes a front-end unit, a back-end unit, and a correction unit. After interference identification, the front-end unit uses the MRAC algorithm to generate preliminary control commands based on the interference intensity, pressure deviation value, and oxygen concentration deviation value.

[0009] The back-end unit synchronously receives the interference intensity, pressure deviation value and oxygen concentration deviation value, and uses the particle swarm optimization-radial basis neural network algorithm to complete the high-precision optimization of the flow ratio coefficient and pressure regulation gain, and generates control commands based on the preliminary control command, flow ratio coefficient and pressure regulation gain.

[0010] The correction unit, based on the optimized flow ratio coefficient and pressure regulation gain coefficient, corrects the action amplitude of the initial control command from the front end and then outputs a correction control command.

[0011] The execution module is used to respond to the correction control command to control the flow ratio distribution of the air valve or oxygen valve and the valve opening of the pressure regulating valve.

[0012] Preferably, the high-precision optimization of the flow ratio coefficient and pressure regulation gain using the particle swarm optimization-radial basis neural network algorithm includes the following sub-steps:

[0013] A fitness function is constructed using the disturbance intensity, pressure deviation, and oxygen concentration deviation as inputs.

[0014] An adaptive particle swarm optimization algorithm is used to globally search for the flow ratio coefficient and pressure regulation gain within a set search range.

[0015] The initial control command direction of the front-end unit is used as a hard constraint to ensure that the optimized parameters are consistent with the control intention. Figure 1 To;

[0016] The optimization results are locally corrected using an RBF neural network to obtain the flow ratio coefficient and pressure regulation gain.

[0017] Preferably, the particle swarm optimization algorithm uses an adaptive inertia weight w to update the particle velocity. The formula for calculating the adaptive inertia weight is as follows:

[0018] ;

[0019] in, The maximum inertia weight value is set. The minimum inertia weight value is set, and t is the current iteration speed. It is the maximum number of iterations. It is the basic attenuation term of interference. This is the interference compensation term, where I is the interference intensity.

[0020] Preferably, the fitness function is calculated using the following formula:

[0021] ;

[0022] Where f is the fitness function and I is the interference intensity. It is the pressure deviation value. This is the oxygen concentration deviation value. It is the weighting coefficient.

[0023] Preferably, when performing particle swarm optimization, the back-end unit treats the direction of the initial control command generated by the front-end unit as a hard constraint;

[0024] When calculating particle fitness, it is determined whether the parameter combination represented by the particle conflicts with the direction of the initial control command. If there is a conflict, a penalty factor is applied to the particle's fitness value to reduce the probability that the particle is selected as the optimal solution, ensuring that the optimization parameters are consistent with the front-end control intention.

[0025] Preferably, it also includes a dual-parameter coordination module, which receives data collected by the data acquisition module, calculates the oxygen concentration coupling amplitude and pressure coupling amplitude according to the preset dual-parameter coupling model, and then calculates the gas supply end feedforward compensation and pressure feedforward compensation respectively. The gas supply end feedforward compensation is used to offset the coupling interference of pressure regulation on oxygen concentration, and the gas supply end feedforward compensation is used to offset the coupling interference of oxygen concentration regulation on pressure.

[0026] The gas supply feedforward compensation and pressure feedforward compensation are transmitted to the adaptive control module. The adaptive control module performs reverse cancellation and superposition of the gas supply feedforward compensation and pressure feedforward compensation with the action amplitude corresponding to the correction control command, and generates a secondary correction control command after reverse cancellation and superposition.

[0027] The execution module responds to and executes the secondary correction control command.

[0028] Preferably, after calculating the oxygen concentration coupling amplitude and the pressure coupling amplitude, the gas supply end feedforward compensation and pressure feedforward compensation are calculated respectively, including:

[0029] Gas supply end feedforward compensation The calculation formula is: ;

[0030] Pressure regulating end feedforward compensation The calculation formula is: ;

[0031] in, is the system feedforward gain, with a value of 0.8-1.2, n is the oxygen concentration coupling amplitude, and m is the pressure coupling amplitude.

[0032] Preferably, the two-parameter coupling model is a two-parameter coupling model based on least squares support vector machine, used to represent the nonlinear mapping relationship between oxygen concentration and pressure;

[0033] The nonlinear mapping relationship is expressed as follows: ,in, To share the weight matrix, is the RBF kernel mapping, and b is the bias vector.

[0034] A pressure control method for an air-oxygen mixer includes the following steps:

[0035] S1: Initialize various sensors, set system target control parameters and system safety protection pressure values;

[0036] S2: Continuously collect real-time operating data of the system and synchronously monitor for any sudden changes in input flow.

[0037] S3: Based on the collected real-time operating data, calculate the pressure deviation value and oxygen concentration deviation value simultaneously, extract the time-domain abrupt change characteristics of the input flow rate sudden change signal, and calculate and quantify the current interference intensity through the instantaneous deviation amplitude quantization method;

[0038] S4: Based on the identified interference intensity, pressure deviation, and oxygen concentration deviation, a preliminary control command is generated. The PSO-RBF is used to optimize the flow proportional coefficient and pressure regulation gain. The optimized flow proportional coefficient and pressure regulation gain are used to correct the preliminary control command to obtain the corrected control command. Then, based on the preset coupling model, the feedforward compensation at the gas supply end and the feedforward compensation at the pressure regulation end are calculated. The feedforward compensation at the gas supply end and the feedforward compensation at the pressure regulation end are used to correct the corrected control command to obtain the secondary corrected control command.

[0039] S5: By using secondary correction control commands, the opening of the oxygen valve and air valve is adjusted to change the gas mixing ratio and total input flow rate, and the opening of the output gas valve is adjusted to control the final output pressure.

[0040] Preferably, the protection pressure value is set as follows: the mixer output pressure is greater than or equal to 12 cmH2O and less than 18 cmH2O for more than 2 seconds, or the mixer output pressure is greater than or equal to 18 cmH2O and triggers immediately, and when the set pressure is reached, the maximum output flow of the mixer shall not exceed the target value plus 3 LPM.

[0041] The beneficial effects of this invention are as follows:

[0042] This invention uses an interference identification module to detect sudden changes in the gas supply condition in real time, a dual-parameter collaborative module for feedforward compensation to actively decouple the system, and an adaptive control module for fine-grained feedback correction. This upgrades passive response to active prediction and intelligent suppression, improving control accuracy and speed. Interference intensity is used as a core variable to dynamically adjust the fitness function weights and inertia weights in the optimization algorithm. This ensures the system prioritizes stability (pressure) under strong interference and accuracy (oxygen concentration) under weak interference. The initial front-end command direction serves as a hard constraint for back-end optimization, avoiding potential inverse logic results from the intelligent algorithm and ensuring the coordination and safety of the control process. The dual-parameter collaborative module uses a dual-parameter coupling model to quantitatively calculate the interference of pressure regulation on oxygen concentration (oxygen concentration coupling amplitude) and the interference of oxygen concentration regulation on pressure (pressure coupling amplitude), generating precise feedforward compensation signals to fundamentally suppress internal disturbances within the system. Through intelligent optimization algorithms and feedforward compensation, the output pressure and oxygen concentration can be stabilized near the set values ​​simultaneously, effectively avoiding the coupling phenomenon where adjusting one parameter leads to a change in another parameter. This meets the stringent requirements of medical applications for gas parameters. The interference identification module enables it to quickly detect disturbances, and the combined effect of feedforward compensation and adaptive algorithms allows it to quickly counteract the impact of disturbances, greatly shortening the time for the system to recover stability. By using front-end commands as optimization constraints, drastic changes in control commands are avoided, making the control process smoother. This achieves fast, accurate, stable, and safe intelligent control of the air-oxygen mixer's pressure and oxygen concentration. Attached Figure Description

[0043] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0044] Figure 1 This is a structural block diagram of the control system of the present invention;

[0045] Figure 2 This is a flowchart illustrating the steps of the control method of the present invention;

[0046] Figure 3 This is a control flowchart of the control system of the present invention. Detailed Implementation

[0047] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0048] Please see Figure 1 This embodiment provides a pressure control system for an air-oxygen mixer, including:

[0049] The data acquisition module is used to acquire the sudden change signals of the input flow of the air valve and the input flow of the oxygen valve, acquire the real-time oxygen concentration and real-time output pressure value of the mixed gas output from the mixer output port, and obtain the output pressure setpoint and target oxygen concentration.

[0050] When the flow sensor collects the real-time flow rate of the gas input to the mixer from the air or oxygen valve at the gas supply end, and the rate of change of flow rate calculated based on the real-time flow rate exceeds 5% of the set value, it is determined that a sudden change in flow rate has occurred. The set value here is the input flow rate value on both sides of the air or oxygen valve under stable operating conditions during the previous stable period.

[0051] A pressure sensor is used to detect the real-time pressure value of the mixed gas output from the output terminal. A pressure sensor is a device or apparatus that can sense pressure signals and convert them into usable output electrical signals according to a certain rule. The pressure sensor reads the current pressure value in real time.

[0052] The cause of the sudden flow change signal is abnormal transmission interference such as blockage in the gas supply pipeline, as well as the imbalance of the gas supply and output flow caused by blockage of the exhaust port or failure of the exhaust valve. When it is determined that there is a sudden flow change signal in the input flow of the air valve and the oxygen valve, the fault signal is uploaded to the central control terminal.

[0053] When the gas supply is in a steady state, high-frequency sampling is not required, which can reduce system power consumption. However, when interference occurs, the sampling frequency needs to be increased to ensure that the details of the sudden signal are captured. The data acquisition module also includes the use of a high-frequency dynamic variable sampling strategy, that is, when it is determined that pressure sudden signal and flow sudden signal occur, the original sampling time is shortened to half of the original sampling time, balancing the sampling real-time performance and system resource consumption, avoiding missing key interference signals, and providing a high-quality data foundation for subsequent processing.

[0054] Since the parameters of the original Kalman filter are fixed, it is difficult to adapt to the fluctuations in the operating conditions of the air-oxygen mixer. Therefore, the data acquisition module also includes a noise reduction process for the acquired data using an improved fast Kalman filter algorithm. By adjusting the value ranges of the process noise covariance matrix and the observation noise covariance matrix, the signal-to-noise ratio of the filtered data is improved, the noise reduction error is reduced, and the acquisition noise is effectively eliminated.

[0055] The calculation module calculates the pressure deviation and oxygen concentration deviation based on the filtered real-time oxygen concentration value, the real-time output pressure value, the target oxygen concentration setpoint, and the output pressure setpoint, respectively.

[0056] The calculation module subtracts the real-time output pressure value from the output pressure setpoint to calculate the current pressure deviation value. If the current pressure deviation value is equal to 0, the control quantity remains unchanged, and the pressure deviation value for the next moment is calculated. If the current pressure deviation value is not equal to 0, the control quantity to maintain the system pressure is adjusted through subsequent module calculations, and the pressure deviation value for the next moment is calculated.

[0057] The same applies to the oxygen concentration value. The real-time oxygen concentration value is subtracted from the output pressure setpoint to calculate the current oxygen concentration deviation value. If the current pressure deviation value is equal to 0, the control quantity remains unchanged, and the oxygen concentration deviation value for the next moment is calculated. If the current pressure error is not equal to 0, the control quantity to maintain the system oxygen concentration is adjusted through subsequent module calculations, and the oxygen concentration deviation value for the next moment is calculated.

[0058] The interference identification module is used to extract the time-domain abrupt change characteristics of the sudden change signals of the air valve input flow and the oxygen valve input flow. The interference intensity is calculated by the instantaneous deviation amplitude quantization method. The time-domain abrupt change characteristics include the peak slope and the deviation growth rate.

[0059] Receive the flow rate change rate calculated from the data acquisition module to calculate the time-domain abrupt change characteristics, including:

[0060] Among them, peak slope

[0061] ,in, It refers to the peak values ​​of the sudden change signals, specifically the peak values ​​of the sudden change signals in the air valve input flow and oxygen valve flow, including the peak flow rate during the sudden change in the air valve input flow and the peak flow rate during the sudden change in the oxygen valve flow. It is the steady-state value of the signal, that is, the input flow rate of the oxygen valve and the input flow rate of the air valve or oxygen valve at the gas supply end under stable operating conditions in the previous moment. It is the time for the flow signal to go from steady state to peak value, and the peak slope is used to quantify the steepness of the change in the interference itself.

[0062] Deviation growth rate , This refers to real-time input parameters, i.e., the actual flow rate at the current moment, including real-time air valve flow rate and real-time oxygen valve flow rate. The actual flow rate at steady state includes the air valve flow rate and the oxygen valve flow rate at steady state. The deviation growth rate is used to quantify the rate at which disturbances deviate from the system's target value.

[0063] The interference intensity is calculated by weighting the deviation growth rate, peak slope, and real-time change rate together with the instantaneous deviation amplitude quantification method. The interference intensity ranges from 0 to 1, with larger values ​​indicating stronger interference.

[0064] The adaptive control module includes a front-end unit, a back-end unit, and a correction unit. After disturbance identification, the front-end unit uses the Model Reference Adaptive Control (MRAC) algorithm to generate preliminary control commands based on the disturbance intensity, pressure deviation, and oxygen concentration deviation.

[0065] First, the disturbance intensity, pressure deviation, and oxygen concentration deviation are received, and the model is initialized to generate model parameters. Using the disturbance intensity, pressure deviation, and oxygen concentration deviation as model inputs, the model output is calculated.

[0066] The model here is the reference model, a core component of the Model Reference Adaptive Control (MRAC) algorithm. It defines the desired dynamic response characteristics of the system and is typically represented using linear differential equations. The prediction error between the actual output and the reference model's output is calculated. Then, based on historical data and disturbance intensity, the sensitivity matrix is ​​updated in real-time using recursive least squares. The sensitivity matrix describes the impact of parameter changes on the output. The model parameters are those of the linear differential equations. The goal is to generate ideal pressure or oxygen concentration response curves. Based on the prediction error and the sensitivity matrix, the adaptive parameters are updated using gradient descent, and a dynamic gain adjustment mechanism ensures rapid parameter convergence. Control quantities are then allocated according to the proportional relationship between pressure and oxygen concentration deviations, mapped to action commands for oxygen valves, air valves, and pressure regulating valves, with amplitude limits set to avoid over-adjustment. Finally, initial control commands are output in the form of PWM signals, specifying the adjustment direction and base amplitude. By predicting future output deviations, parameters are optimized, avoiding the hysteresis problem of traditional MRAC.

[0067] The back-end unit synchronously receives interference intensity, pressure deviation, and oxygen concentration deviation values. It then uses the Particle Swarm Optimization-Radial Basis Function Neural Network (APSO-RBFNN) algorithm to optimize the control parameters. This optimization process includes: using the flow rate proportionality coefficient and pressure regulation gain coefficient as optimization targets, and optimizing the particle search efficiency through adaptive inertial weights to complete the control optimization; simultaneously, using the coarse adjustment direction of the initial front-end control command as an optimization constraint to ensure that the optimized control parameters are consistent with the direction of the initial control command. The control parameters include the flow rate proportionality coefficient and pressure regulation gain coefficient. Based on the initial control command and the control parameters, a control command is generated. The specific process is as follows:

[0068] Step 1: Initialize the optimization parameters, set the search range of the flow ratio coefficient and pressure regulation gain coefficient, and configure the initial parameters of the particle swarm optimization algorithm, including the number of particles, the maximum number of iterations, and the learning factor.

[0069] Step 2: Receive pressure deviation value and oxygen concentration deviation value, and at the same time obtain the interference intensity output by the interference identification module and the direction of the preliminary control command generated by the front-end unit;

[0070] Step 3: Construct the fitness function of the algorithm. The fitness function dynamically adjusts the weights of pressure deviation and oxygen concentration deviation according to the intensity of disturbance. Pressure stability is given priority in the case of strong disturbance, and oxygen concentration accuracy is given priority in the case of weak disturbance. The larger the fitness function value, the better the parameter combination.

[0071] Since the air-oxygen mixer needs to control both oxygen concentration and pressure simultaneously, the fitness function must comprehensively consider the deviations of both parameters to avoid optimizing one parameter leading to the deterioration of the other. Furthermore, the system prioritizes oxygen concentration and pressure control differently under varying disturbance intensities. For example, under strong disturbances, pressure stability must be prioritized to prevent the mixed gas from failing to reach the patient's lungs in a timely manner due to excessively low pressure, and to avoid excessive pressure fluctuations that could damage the patient's alveoli. Therefore, the fitness function needs to dynamically adjust its weights. The formula for calculating the fitness function f is:

[0072] ;

[0073] Where I is the interference intensity, It is the pressure deviation value. This is the oxygen concentration deviation value. This refers to the weighting coefficient. The value of the weighting coefficient is determined by the numerical value corresponding to the interference intensity and the interference intensity range. For example, the interference range includes the first, second, and third intervals. The later the interval is ranked, the larger the numerical value corresponding to the interference intensity, and the higher the interference intensity. Different weighting coefficient values ​​are set for different interference intervals, namely 0.3, 0.5, and 0.7. The value of f ranges from 0 to 1. The larger the value, the better the parameter combination. Based on the interference intensity, the weights of pressure deviation and oxygen concentration deviation are dynamically adjusted to ensure that pressure stability is prioritized during strong interference and oxygen concentration accuracy is prioritized during weak interference. This allows for the selection of the most suitable combination of control parameters (flow rate proportional coefficient, pressure regulation gain coefficient) for the current operating conditions, avoiding the deterioration of one parameter due to the optimization of a single parameter.

[0074] Step 4: Since strong interference at the gas supply end requires a larger weight to expand the search range to find the optimal solution, while weak interference requires a smaller weight to improve convergence accuracy, in the adaptive particle swarm optimization stage, each parameter combination is encoded as a particle. The optimization process begins by initializing the particle swarm position and velocity. Based on the historical optimal position and global optimal position of the particles, combined with adaptively adjusted inertial weights, the particle positions are updated. The inertial weights are automatically adjusted with the number of iterations and the intensity of interference. Under strong interference, this expands the search range, avoids getting trapped in local optima, and improves the global exploration capability of the balancing algorithm. The formula for calculating the inertial weight value in the current iteration is:

[0075] ;

[0076] in, The maximum inertia weight value is set. The minimum inertia weight value is set, and t is the current iteration speed. I is the maximum number of iterations, and I is the interference intensity.

[0077] This is the interference-based decay term, which decreases linearly with the number of iterations, simulating the decay of the inertial weight in traditional PSO. This is an interference compensation term used to dynamically adjust the weights based on the interference intensity. The stronger the interference, the smaller the compensation term. By dynamically adjusting the inertia weight value, when the interference increases, the weight value is increased to improve the particle's ability to escape local optima. Under strong interferences such as sudden increases or decreases in oxygen flow, this improves the success rate of parameter optimization. When the interference is weak, the weight value is decreased to allow the particle to search more precisely near the optimal solution, improving robustness and control accuracy. This makes subsequent adjustments to control parameters smoother, reduces system output fluctuations, and avoids frequent actuator movements.

[0078] Step 5: Use the initial control command direction of the front-end unit as a hard constraint for particle swarm optimization to ensure that the parameters are within a reasonable range and consistent with the initial control command direction. If the particle parameters conflict with the initial command direction, such as the command requiring a reduction in the oxygen ratio while the particle parameters are suggested to be increased, then the fitness value is penalized by multiplying by 0.5 to avoid system oscillation caused by parameter mutations. Based on the parameter range constraint, the front-end control intent is integrated to reduce the possibility of conflicts between the optimization parameters and the initial command and to avoid system adjustment chaos.

[0079] Step 6: When the particle swarm optimization reaches the maximum number of iterations or the optimal solution does not change significantly over multiple generations, output the optimized parameters;

[0080] Step 7: Input the parameters into the radial basis neural network for further correction. This neural network takes pressure deviation, oxygen concentration deviation and disturbance intensity as inputs, and establishes a parameter correction model through training with historical data. It performs local fine-tuning of the particle swarm optimization results to improve parameter accuracy.

[0081] Step 8: Verify whether the optimized parameters meet the accuracy requirements and system stability conditions. If the conditions are met, the final optimized parameters are passed to the correction unit to accurately correct the amplitude of the initial control command and generate the final control command.

[0082] By combining the global optimization capability of adaptive particle swarm optimization with the local correction capability of radial basis neural network, the system can improve parameter accuracy while ensuring optimization speed, and ensure that the system can maintain stable operation in complex interference environment.

[0083] The correction unit, based on the optimized flow ratio coefficient and pressure regulation gain coefficient, corrects the amplitude of the initial control command from the front end and outputs a corrected control command.

[0084] The core control objects at the input end of the air-oxygen mixer are the air valve and the oxygen valve. The flow ratio between the two directly determines the oxygen concentration of the mixed gas. The total input flow indirectly affects the output pressure (increasing the total flow tends to lead to higher pressure, and vice versa). The mixed gas pressure needs to be readjusted through the pressure regulating valve. When oxygen concentration adjustment is performed (such as adjusting the air or oxygen ratio), the output pressure is affected through the coupling effect. For example, increasing the oxygen valve opening causes the oxygen concentration to increase, the flow rate to increase, and the output pressure to passively increase. The pressure regulating end feedforward compensation adjusts the total flow (or ratio) at the input end to offset this interference in advance in order to maintain pressure stability.

[0085] Specifically, the dual-parameter collaborative module is used to receive data collected by the data acquisition module, calculate the oxygen concentration coupling amplitude and pressure coupling amplitude according to the preset dual-parameter coupling model, and then calculate the gas supply end feedforward compensation and the pressure regulation end feedforward compensation respectively. The gas supply end feedforward compensation is used to offset the coupling interference of pressure regulation on oxygen concentration, and the pressure regulation end feedforward compensation is used to offset the coupling interference of oxygen concentration regulation on pressure.

[0086] The gas supply-side feedforward compensation and pressure feedforward compensation are transmitted to the adaptive control module to generate secondary correction control commands. The feedforward compensation essentially uses the amplitude of coupled disturbances predicted by a two-parameter coupled model to pre-correct the opening degrees of the proportional valves of the air and oxygen valves, offsetting the interference of pressure regulation on oxygen concentration and oxygen concentration regulation on pressure, ensuring that the input flow ratio matches the total flow. The specific implementation process includes the following steps:

[0087] A two-parameter coupled model based on least squares support vector machine (LSSVM) was constructed through offline training to represent the nonlinear mapping relationship between oxygen concentration and pressure.

[0088] Training sample data was collected and preprocessed, including steady-state operating conditions of the mixer (routine oxygen therapy for adults, low-flow oxygen therapy for newborns), weak interference conditions (gas source pressure fluctuations of ±10%, slight changes in patient respiratory rate), and moderate to strong interference conditions (partial airway obstruction, a sudden 50% increase in oxygen valve flow, and a sudden change in air valve pressure). Each sample contained 5-dimensional input and 2-dimensional output. Input data was synchronously collected via sensors to ensure data time alignment; output data was obtained through a regulation-measurement closed-loop experiment, i.e., manually adjusting the proportional valve amplitude of the air valve or oxygen valve and the adjustment amplitude of the pressure regulating valve, while simultaneously recording the actual coupled change of the other parameter to ensure label accuracy. The input parameters were standardized to eliminate the influence of dimensions.

[0089] The basic structure of LSSVM is constructed using a multi-output fusion structure with a shared kernel matrix. This directly models the coupling mapping relationship between the input and dual outputs, avoiding the parameter redundancy and prediction conflicts of traditional independent single-output models. The nonlinear mapping relationship formula of the model is as follows:

[0090] ;in, To share the weight matrix, The RBF kernel function is used because the coupling relationship in an air-oxygen mixer is typically nonlinear. For example, increasing the oxygen flow rate not only increases the oxygen concentration but also indirectly affects the pressure due to the change in total flow rate, and the degree of influence varies with the intensity of the disturbance. This relationship cannot be accurately described by a linear model in low-dimensional space, so the RBF kernel function is used. The 5-dimensional input is mapped to a high-dimensional feature space. In the high-dimensional space, the originally complex nonlinear coupling relationship is transformed into a linear relationship. b is the bias vector. The complexity of the coupling relationship is solved by nonlinear mapping, collaborative prediction is achieved by linear combination of dual outputs, and system error is compensated by bias correction. Subsequently, the constraint optimization is transformed into solving a system of linear equations by the Lagrange multiplier method, balancing the model complexity and fitting efficiency.

[0091] During long-term operation of the air-oxygen mixer, factors such as proportional valve wear and pipeline dust accumulation can cause a slow drift in coupling characteristics, leading to a decrease in accuracy of offline-trained fixed-parameter models. To address this, a recursive least squares (RLS) algorithm is introduced to periodically update model parameters, compensating for the coupling characteristic drift caused by proportional valve wear, pipeline dust accumulation, and sensor aging during long-term operation of the air-oxygen mixer.

[0092] During the model validation phase, the performance of the nonlinear coupled model is evaluated through cross-validation to ensure that the prediction error between the training set and the test set is less than a set threshold. During online operation, the error between the predicted value and the sensor measured value is monitored in real time. When the error exceeds the set threshold multiplied by 1.2 for three consecutive sampling periods, the latest valid samples are automatically added and background retraining is triggered, forming a closed-loop mechanism of training-validation-correction-retraining, which further enhances the robustness of the coupled model under sudden change scenarios.

[0093] Step B: The data acquisition module collects the real-time oxygen concentration and pressure values ​​of the mixed gas, the oxygen concentration deviation between the real-time oxygen concentration and the target oxygen concentration setpoint, and the pressure deviation between the real-time pressure and the output pressure setpoint. Combined with the interference intensity, these values ​​are input into the trained two-parameter coupling model. The model calculates the oxygen concentration coupling amplitude and pressure coupling amplitude based on the input. The oxygen concentration coupling amplitude n is used to cancel the coupling interference of pressure regulation on oxygen concentration regulation, and the pressure coupling amplitude m is used to cancel the coupling interference of oxygen concentration regulation on pressure regulation.

[0094] Step C: Based on the predicted amplitudes of the two coupled interferences, calculate the corresponding gas supply-side feedforward compensation. and pressure regulating end feedforward compensation , This represents the system feedforward gain, with a value ranging from 0.8 to 1.2.

[0095] Gas supply end feedforward compensation The calculation formula is: ;

[0096] Pressure regulating end feedforward compensation The calculation formula is: ;

[0097] Among them, the gas supply end feedforward compensation is used to offset the effect of pressure regulation on oxygen concentration, and the pressure regulation end feedforward compensation is used to offset the effect of oxygen concentration regulation on pressure.

[0098] The feedforward compensation at the gas supply end is mapped to the opening correction value of the oxygen proportional valve or the air proportional valve. If the feedforward compensation at the gas supply end is positive, the oxygen concentration is reduced by decreasing the opening of the oxygen valve or increasing the opening of the air valve. If the feedforward compensation at the gas supply end is negative, the oxygen concentration is increased by increasing the opening of the oxygen valve or decreasing the opening of the air valve.

[0099] The feedforward compensation of the pressure regulating end is mapped to the valve opening correction value of the pressure regulating valve.

[0100] Step D: The gas supply feedforward compensation and pressure feedforward compensation are transmitted to the adaptive control module. The adaptive control module performs reverse cancellation and superposition of the gas supply feedforward compensation and pressure feedforward compensation with the corresponding action amplitude of the correction control command, and generates a secondary correction control command after reverse cancellation and superposition. The superposition logic is reverse cancellation. This reverse cancellation method can effectively eliminate the coupling interference between parameters and ensure that the oxygen concentration and pressure can be accurately controlled within the target range.

[0101] Precise control of the gas supply input and pressure regulating valves is achieved by modifying the opening degree of the flow ratio valve of the air or oxygen valve. By anticipating coupling interference in advance, the flow ratio at the input end and the pressure regulation of the pressure regulating valve are adjusted accordingly to counteract the impact of interference on the target parameters.

[0102] The execution module is used to respond to secondary correction control commands and control the flow ratio distribution of the proportional valve of the air valve or oxygen valve and the control valve of the pressure regulating device. Specifically, it includes determining the opening step of the proportional valve according to the flow ratio coefficient and calibrating the action amplitude of the pressure regulating valve according to the pressure regulating gain coefficient. Specifically, after receiving the secondary correction control command, the execution module drives the action of the air or oxygen proportional valve and the pressure regulating valve through high-frequency pulse width modulation (PWM) technology, and calculates the duty cycle of the PWM using the secondary correction control command and the target oxygen concentration setpoint.

[0103] Traditional control algorithms typically use independent PID loops to control pressure and oxygen concentration separately, which can easily lead to seesaw-like oscillations due to internal coupling within the system, and also exhibit slow response under sudden flow disturbances.

[0104] This system uses an interference identification module to detect sudden changes in the gas supply condition in real time. A dual-parameter collaborative module performs feedforward compensation for active decoupling, and an adaptive control module performs fine-tuned feedback correction, upgrading passive response to active prediction and intelligent suppression, thus improving control accuracy and speed. The front-end unit generates preliminary control commands based on the interference intensity, pressure deviation, and oxygen concentration deviation using the MRAC algorithm. The back-end unit employs a hybrid algorithm combining a PSO-RBF neural network for high-precision optimization. MRAC ensures the speed and basic directional accuracy of control, while PSO-RBF is responsible for finding the optimal parameter combination within a complex space, achieving a balance between speed and accuracy. The system uses interference intensity as the core variable, dynamically adjusting the fitness function weights and inertia weights in the optimization algorithm. This allows the system to prioritize stability (pressure) under strong interference and prioritize accuracy (oxygen concentration) under weak interference.

[0105] By using the initial command direction from the front end as a hard constraint for back-end optimization, the reverse logic results that intelligent algorithms may produce are avoided, ensuring the coordination and safety of the control process. The dual-parameter collaborative module uses a dual-parameter coupling model to quantitatively calculate the amount of interference of pressure regulation on oxygen concentration (oxygen concentration coupling amplitude) and the amount of interference of oxygen concentration regulation on pressure (pressure coupling amplitude), and generates accurate feedforward compensation signals accordingly, thereby suppressing internal disturbances within the system from the root.

[0106] Through intelligent optimization algorithms and feedforward compensation, the system can simultaneously stabilize the output pressure and oxygen concentration near the set values, effectively avoiding the coupling phenomenon where adjusting one parameter leads to a change in another. This meets the stringent requirements of medical applications for gas parameters. The interference identification module enables it to quickly detect disturbances, and the combined effect of feedforward compensation and adaptive algorithms allows it to quickly counteract the impact of disturbances, greatly shortening the system's recovery time. By using front-end commands as optimization constraints, the system avoids drastic changes in control commands, making the control process smoother. This achieves rapid, accurate, stable, and safe intelligent control of the air-oxygen mixer's pressure and oxygen concentration.

[0107] Please see Figure 3 The control logic of the system control flowchart is executed by the built-in processing modules (such as MCU and PLC) of the control system, forming a multi-condition judgment closed loop with pressure as the priority, oxygen concentration as the constraint, and flow rate as the guarantee. Specifically, it includes:

[0108] After system power-on initialization, it enters the main control loop: First, the controller powers on, all sensors and actuators perform self-checks and resets, and enter standby mode. Then, the system determines whether the real-time output pressure exceeds the set value. If the pressure is too high, it further determines the oxygen concentration: If the oxygen concentration is also too high, it reduces the oxygen valve opening and increases the air valve opening to synergistically reduce pressure and oxygen; if the oxygen concentration is not too high, it directly reduces the output valve opening to lower the pressure. If the pressure is not too high, it increases the output valve opening to increase the pressure.

[0109] During the pressure regulation process described above, the system simultaneously determines whether the output flow rate exceeds the set value by a certain margin (e.g., set value + 3L / min). If the flow rate is insufficient, the oxygen valve and air valve at the control input end are increased in tandem to ensure the total gas supply from the gas supply end.

[0110] It should be noted that the logical output of the control flow (i.e., the control commands for each valve) can be integrated with the corrective control commands generated by the adaptive control module. For example, the flow proportional coefficient output by the adaptive control module can be used to accurately calculate the specific change in opening of the oxygen valve (decreasing) and the air valve (increasing); while its output pressure regulation gain can be used to dynamically adjust the response speed of the output valves (increasing / decreasing). This combines rule-based rapid decision-making with model-based intelligent optimization to achieve optimal system performance.

[0111] Please see Figure 2 This embodiment provides a pressure control method for an air-oxygen mixer, the method including the following steps:

[0112] S1: Initialize various sensors and control valves and set the system target control parameters. Sensors include input / output pressure sensors, ultrasonic flow sensors, and oxygen concentration sensors. Control valves include oxygen valves, air valves, and pressure regulating valves. System target control parameters include output pressure setpoint, target oxygen concentration setpoint, target output flow setpoint, and system safety protection pressure value. The system protection pressure value is set as follows: the mixer output pressure is greater than or equal to 12cmH2O and less than 18cmH2O for more than 2 seconds, or the mixer output pressure is greater than or equal to 18cmH2O and triggers immediately. When the set pressure is reached, the mixer output flow rate shall not exceed the target value plus 3LPM.

[0113] S2: Continuously collect real-time operating data of the system and synchronously monitor the sudden change signal of the input flow at the gas supply end;

[0114] S3: Based on the collected real-time operating data, calculate the pressure deviation value and oxygen concentration deviation value simultaneously, extract the time-domain abrupt change characteristics of the input flow rate sudden change signal, and calculate and quantify the current interference intensity through the instantaneous deviation amplitude quantization method;

[0115] S4: Based on the identified interference intensity, pressure deviation, and oxygen concentration deviation, a preliminary control command is generated. The PSO-RBF is used to optimize the flow proportional coefficient and pressure regulation gain. The optimized flow proportional coefficient and pressure regulation gain are used to correct the preliminary control command once to obtain the corrected control command. Then, based on the preset coupling model, the gas supply feedforward compensation and pressure feedforward compensation are calculated. The gas supply feedforward compensation and pressure feedforward compensation are used to correct the corrected control command a second time to obtain the secondary corrected control command.

[0116] S5: By using secondary correction control commands, the opening of the oxygen valve and air valve is adjusted to change the gas mixing ratio and total input flow rate, and the opening of the output gas valve is adjusted to control the final output pressure.

[0117] The method effectively solves the problem of coupled control of pressure and oxygen concentration in air-oxygen mixers through a dual mechanism of feedforward compensation and adaptive feedback optimization. Furthermore, by introducing disturbance identification and intelligent optimization algorithms, it significantly improves the system's response speed, control accuracy, and robustness under disturbance conditions such as sudden changes in flow rate.

[0118] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A pressure control system for an air-oxygen mixer, characterized in that: include: The data acquisition module is used to acquire the sudden change signals of the input flow of the air valve and the sudden change signals of the oxygen valve, acquire the real-time oxygen concentration and real-time output pressure value of the mixed gas output from the mixer output port, and obtain the output pressure setpoint and the target oxygen concentration setpoint. The calculation module calculates the pressure deviation and oxygen concentration deviation based on the real-time oxygen concentration, the real-time output pressure value, the target oxygen concentration setpoint, and the output pressure setpoint, respectively. The interference identification module is used to extract the time-domain abrupt change characteristics of the sudden change signal of the air valve input flow and the sudden change signal of the oxygen valve flow, and to synchronously calculate the interference intensity through the instantaneous deviation amplitude quantization method. The adaptive control module includes a front-end unit, a back-end unit, and a correction unit. After interference identification, the front-end unit uses the MRAC algorithm to generate preliminary control commands based on the interference intensity, pressure deviation value, and oxygen concentration deviation value. The back-end unit synchronously receives the interference intensity, pressure deviation value and oxygen concentration deviation value, and uses the particle swarm optimization-radial basis neural network algorithm to complete the high-precision optimization of the flow ratio coefficient and pressure regulation gain, and generates control commands based on the preliminary control command, flow ratio coefficient and pressure regulation gain. The correction unit, based on the optimized flow ratio coefficient and pressure regulation gain coefficient, corrects the action amplitude of the initial control command from the front end and then outputs a correction control command. An execution module is used to respond to the correction control command and control the flow ratio distribution of the air valve or oxygen valve and the valve opening of the pressure regulating valve. It also includes a dual-parameter coordination module, which receives data collected by the data acquisition module, calculates the oxygen concentration coupling amplitude and pressure coupling amplitude according to the preset dual-parameter coupling model, and then calculates the gas supply end feedforward compensation and pressure feedforward compensation respectively. The gas supply end feedforward compensation is used to offset the coupling interference of pressure regulation on oxygen concentration, and the gas supply end feedforward compensation is used to offset the coupling interference of oxygen concentration regulation on pressure. The gas supply feedforward compensation and pressure feedforward compensation are transmitted to the adaptive control module. The adaptive control module performs reverse cancellation and superposition of the gas supply feedforward compensation and pressure feedforward compensation with the action amplitude corresponding to the correction control command, respectively, to generate a secondary correction control command after reverse cancellation and superposition. The execution module responds to and executes the secondary correction control command.

2. The pressure control system for an air-oxygen mixer according to claim 1, characterized in that: The high-precision optimization of the flow ratio coefficient and pressure regulation gain using the particle swarm optimization-radial basis neural network algorithm includes the following sub-steps: A fitness function is constructed using the disturbance intensity, pressure deviation, and oxygen concentration deviation as inputs. An adaptive particle swarm optimization algorithm is used to globally optimize the flow ratio coefficient and pressure regulation gain within a set search range. The initial control command direction of the front-end unit is used as a hard constraint to ensure that the optimized parameters are consistent with the control intention. The optimization results are locally corrected using an RBF neural network to obtain the flow ratio coefficient and pressure regulation gain.

3. The pressure control system for an air-oxygen mixer according to claim 2, characterized in that: The particle swarm optimization algorithm uses an adaptive inertia weight w to update the particle velocity. The formula for calculating the adaptive inertia weight is: ； in, The maximum inertia weight value is set. The minimum inertia weight value is set, and t is the current iteration speed. It is the maximum number of iterations. It is the basic attenuation term of interference. This is the interference compensation term, where I is the interference intensity.

4. The pressure control system for an air-oxygen mixer according to claim 2, characterized in that: The fitness function is calculated using the following formula: ； Where f is the fitness function and I is the interference intensity. It is the pressure deviation value. This is the oxygen concentration deviation value. It is the weighting coefficient.

5. The pressure control system for an air-oxygen mixer according to claim 1, characterized in that: When performing particle swarm optimization, the back-end unit treats the direction of the initial control command generated by the front-end unit as a hard constraint. When calculating particle fitness, it is determined whether the parameter combination represented by the particle conflicts with the direction of the initial control command. If there is a conflict, a penalty factor is applied to the particle's fitness value to reduce the probability that the particle is selected as the optimal solution, ensuring that the optimization parameters are consistent with the front-end control intention.

6. The pressure control system for an air-oxygen mixer according to claim 1, characterized in that: After calculating the oxygen concentration coupling amplitude and pressure coupling amplitude, the gas supply end feedforward compensation and pressure feedforward compensation are calculated separately, including: Gas supply end feedforward compensation The calculation formula is: ; Pressure regulating end feedforward compensation The calculation formula is: ; in, is the system feedforward gain, with a value of 0.8-1.2, n is the oxygen concentration coupling amplitude, and m is the pressure coupling amplitude.

7. The pressure control system for an air-oxygen mixer according to claim 1, characterized in that: The two-parameter coupling model is a two-parameter coupling model based on least squares support vector machine, used to represent the nonlinear mapping relationship between oxygen concentration and pressure; The nonlinear mapping relationship is expressed as follows: ,in, To share the weight matrix, is the RBF kernel mapping, and b is the bias vector.

8. A pressure control method for an air-oxygen mixer, applied to a pressure control system for an air-oxygen mixer as described in any one of claims 1-7, characterized in that: Includes the following steps: S1: Initialize various sensors, set system target control parameters and system safety protection pressure values; S2: Continuously collect real-time operating data of the system and synchronously monitor sudden changes in the input flow rate at the gas supply end; S3: Based on the collected real-time operating data, calculate the pressure deviation value and oxygen concentration deviation value simultaneously, extract the time-domain abrupt change characteristics of the input flow rate sudden change signal, and calculate and quantify the current interference intensity through the instantaneous deviation amplitude quantization method; S4: Based on the identified interference intensity, pressure deviation, and oxygen concentration deviation, a preliminary control command is generated. The particle swarm optimization-radial basis neural network algorithm is used to optimize the flow ratio coefficient and pressure regulation gain. The optimized flow ratio coefficient and pressure regulation gain are used to correct the preliminary control command once to obtain the corrected control command. Then, based on the preset coupling model, the gas supply feedforward compensation amount and pressure feedforward compensation amount are calculated. The gas supply feedforward compensation amount and pressure feedforward compensation amount are used to correct the corrected control command a second time to obtain the secondary corrected control command. S5: By using secondary correction control commands, the opening of the oxygen valve and air valve is adjusted to change the gas mixing ratio and total input flow rate, and the opening of the output gas valve is adjusted to control the final output pressure.

9. The pressure control method for an air-oxygen mixer according to claim 8, characterized in that: The protection pressure value is set as follows: the mixer output pressure is greater than or equal to 12cmH2O and less than 18cmH2O for more than 2 seconds, or the mixer output pressure is greater than or equal to 18cmH2O and triggers immediately, and the mixer output flow rate is less than the target value plus 3LPM when the set pressure is reached.

Images