A model prediction based ventilator oxygen concentration control method

Abstract

This invention discloses a model-predictive method for controlling oxygen concentration in a ventilator, relating to the field of medical gas regulation technology. The method includes: fusing multi-source state data within the ventilator; calculating in real-time the system control margin, the current oxygen concentration transfer model mismatch, and the actuator load balancing factor; adaptively adjusting the aggressiveness of the parameter identification algorithm based on the model mismatch, injecting a small-amplitude test signal to correct the model online and update the model mismatch; constructing an optimization objective function based on the corrected oxygen concentration transfer model, constrained by the system control margin and the actuator load balancing factor, and performing rolling optimization to obtain the optimal control command sequence; smoothing the commands according to the system control margin and then issuing the final control command, while simultaneously providing dynamic early warning based on the updated model mismatch and load balancing factor. This invention effectively improves the accuracy and dynamic response speed of ventilator oxygen concentration control.

Description

Technical Field

[0001] This invention relates to the field of medical gas regulation technology, and more specifically to a model-predictive method for controlling oxygen concentration in ventilators. Background Technology

[0002] With the increasing demands for respiratory support equipment in fields such as emergency medical care and intensive care, ventilators, as critical life support devices, directly impact treatment outcomes due to their oxygen concentration control performance. Current technologies mostly employ traditional feedback control or fixed-parameter model control methods, using real-time oxygen concentration monitoring signals for closed-loop adjustment, which can meet the basic respiratory support needs of routine clinical scenarios.

[0003] However, traditional control methods rely on deviation correction, and the dynamic response has an inherent lag, making it difficult to quickly adapt to the dynamic changes in the patient's respiratory rhythm and lung function. Fixed parameter models cannot cope with complex operating conditions such as ventilator tubing leaks and gas mixing characteristic drift, resulting in insufficient accuracy of oxygen concentration control and failing to meet the clinical needs of critically ill patients for rapid and precise adjustment of oxygen concentration. Summary of the Invention

[0004] This application provides a model-based method for controlling oxygen concentration in ventilators, aiming to solve the technical problem of insufficient accuracy in oxygen concentration control in existing technologies.

[0005] In view of the above problems, this application provides a model-predictive method for controlling ventilator oxygen concentration, including:

[0006] Based on multi-source state data within the ventilator, the system control margin is calculated in real time through nonlinear fusion, and the model mismatch of the current oxygen concentration delivery model and the actuator load balancing factor are calculated simultaneously.

[0007] Based on the aggressiveness of the adaptive adjustment parameter identification algorithm for model mismatch, a small-amplitude test signal is injected into the gas mixing system to correct the oxygen concentration transfer model online and calculate and update the model mismatch.

[0008] Based on the modified oxygen concentration transfer model, and constrained by the system control margin and the actuator load balancing factor, an optimization objective function is constructed, and rolling optimization is performed to obtain the optimal control command sequence.

[0009] Based on the system control margin, a strategy for smoothing the optimal control command sequence is determined, and the smoothed final control command is issued. At the same time, dynamic early warning is given based on the real-time status of the update model mismatch degree and the actuator load balancing factor.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] This application provides a model-based oxygen concentration control method for ventilators. By fusing multi-source state data of the ventilator, it obtains the system control margin, model mismatch, and actuator load balancing factor in real time, providing comprehensive and accurate state support for control decisions. Based on the model mismatch, it adaptively adjusts the parameter identification strategy and corrects the oxygen concentration transfer model online, effectively reducing the deviation between the model and the actual operating conditions. Based on the corrected model, it performs rolling optimization solution in combination with constraints to ensure the optimality of control commands. By smoothing control commands through system control margin and providing dynamic early warning, it ensures the stability of the control process and significantly improves the accuracy and dynamic response speed of ventilator oxygen concentration control. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 The present application provides a flowchart of a model-predictive method for controlling ventilator oxygen concentration, which is provided for the purpose of this embodiment. Detailed Implementation

[0014] This application provides a model-predictive method for controlling oxygen concentration in ventilators, which addresses the technical problem of insufficient accuracy in oxygen concentration control in existing technologies.

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0016] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0017] Examples, such as Figure 1 As shown, this application provides a model-predictive method for controlling ventilator oxygen concentration, the method comprising:

[0018] S100: Based on multi-source state data within the ventilator, it calculates the system control margin in real time through nonlinear fusion, and simultaneously calculates the model mismatch of the current oxygen concentration delivery model and the actuator load balancing factor.

[0019] In this embodiment, based on multi-source state data within the ventilator, the system control margin is calculated in real time through nonlinear fusion, and the model mismatch of the current oxygen concentration transfer model and the actuator load balancing factor are calculated simultaneously. Ventilator oxygen concentration control needs to simultaneously address the influence of multiple dimensions such as gas mixing state, actuator coordination performance, sensor data reliability, and dynamic coupling of the flow field. Real-time changes in these influencing factors directly affect control accuracy and response speed. Therefore, multi-source data fusion and specialized index calculation are necessary to comprehensively capture the system state and provide a reliable basis for precise control.

[0020] Step S100 in the method provided in this application embodiment includes:

[0021] Among these, based on multi-source state data within the ventilator, the system control margin is calculated in real time through nonlinear fusion, including:

[0022] Oxygen concentration values ​​are collected from multiple sampling points downstream of the gas mixing chamber, and the spatial variance of the multiple oxygen concentration values ​​is calculated as the mixing uniformity index.

[0023] Collect the sequence of differences between the actual opening degree and the commanded opening degree of multiple gas proportional valves operating in parallel, calculate the degree of consistency, and use it as the actuator coordination deviation;

[0024] Compare the readings of the primary oxygen sensor and the backup oxygen sensor to calculate the sensor self-consistency score;

[0025] Analyze the pipeline pressure fluctuation spectrum under the current ventilation flow rate, and extract the energy value of a specific frequency band as a flow-pressure coupled oscillation index;

[0026] After normalizing the mixing uniformity index, actuator coordination deviation, sensor self-consistency score, and flow-pressure coupling oscillation index, the results are input into the nonlinear fusion calculation unit, which outputs the system control margin.

[0027] First, oxygen concentration values ​​are collected from multiple sampling points downstream of the gas mixing chamber, and the spatial variance of these oxygen concentration values ​​is calculated as the mixing uniformity index. The mixing uniformity index reflects the degree of uniformity of oxygen and air mixing in the gas mixing chamber; the smaller the value, the more uniform the mixing. Implementation method: Three sampling points are set downstream of the gas mixing chamber, such as at the outlet, 5cm from the outlet, and 10cm from the outlet. Oxygen concentration values ​​at each point are collected in real time, and their spatial variance is calculated as the mixing uniformity index. For example, taking a scenario where a ventilator provides ventilation support to critically ill patients with an oxygen concentration of 40%, the oxygen concentrations at the three sampling points are 39.8%, 40.2%, and 40.0%, respectively. The calculated spatial variance is [(39.8-40)²+(40.2-40)²+(40.0-40)²] / 3≈0.027, where 0.027 is the mixing uniformity index.

[0028] Secondly, the difference sequence between the actual opening and the commanded opening of multiple parallel gas proportional valves is collected, and the consistency is calculated as the actuator coordination deviation. The gas proportional valve is the core component controlling the ratio of oxygen to air input. Parallel gas proportional valves are a hardware design strategy. Ventilators require precise control of the oxygen and air mixing ratio; to achieve a wider flow range, higher reliability, or finer control, two or more proportional valves are installed in parallel. The commanded opening is the command or target issued by the control system; it is an expected value, usually a percentage, calculated by the control algorithm, indicating how wide the proportional valve should theoretically open to output the target flow rate. For example, to achieve a specific oxygen concentration, the controller might issue a command to the oxygen proportional valve to open it to 60%. The actual opening is the actual physical position or state of the valve, reflecting its true response. The actuator coordination deviation reflects the consistency between the actual actions and commands of multiple parallel gas proportional valves; the smaller the value, the better the coordination.

[0029] For example, the difference sequence between the actual opening and the commanded opening of the parallel proportional valves is collected. The consistency is calculated as 1 - average absolute difference / commanded opening, with 1 - consistency as the coordination deviation. For instance, if the commanded opening of two parallel proportional valves is 50%, the difference sequence between the actual opening and the commanded opening after 10 samplings is [1, -0.5, 0.8, -1.2, 0.3, -0.7, 1.1, -0.9, 0.6, -0.4]. The average absolute difference is 0.75, and the consistency is 1 - (0.75 / 50) = 0.985. Therefore, the actuator coordination deviation is 1 - 0.985 = 0.015.

[0030] In addition, the sensor self-consistency score is calculated by comparing the readings of the main oxygen sensor and the backup oxygen sensor. The sensor self-consistency score reflects the consistency of the readings of the main oxygen sensor and the backup oxygen sensor; the higher the value, the better the consistency. It is necessary to calculate the sensor self-consistency score by comparing the readings of the main oxygen sensor and the backup oxygen sensor to quantify the consistency of the two data, thereby judging whether the main sensor data is reliable and providing a valid data basis for subsequent control calculations, avoiding control errors caused by sensor anomalies. The absolute difference between the readings of the main oxygen sensor and the backup oxygen sensor is calculated and compared with the preset maximum allowable difference. Sensor self-consistency score = 1 - (actual difference / maximum allowable difference). For example, if the preset maximum allowable difference is 1%, the main sensor reading is 40.1%, the backup sensor reading is 39.9%, and the actual difference is 0.2%, then the self-consistency score = 1 - (0.2 / 1) = 0.8.

[0031] Furthermore, the pipeline pressure fluctuation spectrum under the current ventilation flow rate was analyzed, and the energy values ​​of specific frequency bands were extracted as flow-pressure coupled oscillation indices. The flow-pressure coupled oscillation index reflects the intensity of the coupled oscillation between gas flow rate and pipeline pressure during ventilation; a lower value indicates a more stable flow field. Spectral analysis of the pipeline pressure fluctuation under the current ventilation flow rate was performed, and the energy values ​​of specific frequency bands related to the respiratory rhythm were extracted as flow-pressure coupled oscillation indices. For example, the pressure fluctuation spectrum analysis showed that the energy value in the 0.5-2Hz frequency band was 0.3 Pa², and 0.3 Pa² is the flow-pressure coupled oscillation index.

[0032] Finally, the mixing uniformity index, actuator coordination deviation, sensor self-consistency score, and flow-pressure coupling oscillation index are normalized and then input into the nonlinear fusion calculation unit to output the system control margin. The system control margin comprehensively reflects the current safe controllable space of the system; the higher the value, the stronger the control flexibility. The above four indices are normalized to the range of 0-1, with the mixing uniformity index, coordination deviation, and coupling oscillation index requiring inverse normalization, i.e., the smaller the value, the larger the normalized value. These are then input into the nonlinear fusion calculation unit, such as a weighted neural network, to output the fusion result. For example, the mixing uniformity index of 0.027 is normalized to 0.1, the actuator coordination deviation of 0.015 is normalized to 0.05, the sensor self-consistency score of 0.8 is normalized to 0.8, and the flow-pressure coupling oscillation index of 0.3 Pa² is normalized to 0.3. These are input into the nonlinear fusion calculation unit, and the system control margin of 0.75 is output after fusion calculation.

[0033] This includes simultaneously calculating the model mismatch of the current oxygen concentration transfer model and the actuator load balancing factor, including:

[0034] The model mismatch degree is calculated based on the average absolute error between the short-term predicted value and the actual measured value of the sensor in the oxygen concentration transfer model within a preset time window.

[0035] The number of times and amplitude of each gas proportional valve actuation within a preset time window are statistically analyzed, the coefficient of variation is calculated, and the reciprocal is taken and normalized to obtain the actuator load balancing factor.

[0036] First, the model mismatch is calculated based on the average absolute error between the short-term predicted values ​​and the actual measured values ​​from the sensor within a preset time window using the oxygen concentration transfer model. The oxygen concentration transfer model is a mathematical expression describing the dynamic relationship between the ventilator's internal control commands and the actual output oxygen concentration. The model mismatch reflects the deviation between the predicted and actual values ​​of the oxygen concentration transfer model; a higher value indicates a less accurate model. Calculating the model mismatch allows for real-time monitoring of the model's fit with actual operating conditions, providing a basis for adjusting subsequent parameter identification algorithms. A high mismatch necessitates stronger model correction; a low mismatch maintains model stability, ensuring the model accurately reflects system characteristics and provides a reliable predictive basis for precise control. Within a preset time window, the average absolute error between the short-term predicted oxygen concentration from the oxygen concentration transfer model and the actual measured values ​​from the sensor is calculated as the model mismatch. For example, the model predicts the oxygen concentration sequence as [40.0, 40.1, 39.9, 40.2, 40.0] within 10 seconds, while the measured sequence is [40.2, 40.3, 40.0, 40.4, 40.1]. The mean absolute error is (0.2+0.2+0.1+0.2+0.1) / 5=0.16, which means the model mismatch is 0.16.

[0037] Secondly, the number of actuations and amplitudes of each gas proportional valve within a preset time window are statistically analyzed, the coefficient of variation is calculated, and the reciprocal is taken before normalization to obtain the actuator load balancing factor. The actuator load balancing factor reflects the degree of load balance among the actuations of multiple proportional valves; a higher value indicates a more balanced load, which can reduce local losses. The actuator load balancing factor is obtained by statistically analyzing the number of actuations and amplitudes of each proportional valve within a preset time window, calculating the coefficient of variation, and taking the reciprocal before normalization. The coefficient of variation is the ratio of the standard deviation to the mean, reflecting the degree of dispersion. For example, if two proportional valves actuate 8 and 12 times respectively within 10 seconds, with a mean amplitude of 5% and a standard deviation of 1% for both, and a coefficient of variation of 0.2 for both, the average coefficient of variation is 0.2. Taking the reciprocal of 0.2 as 5 and normalizing, the actuator load balancing factor is 0.8.

[0038] In this embodiment, key information such as gas mixing uniformity, actuator coordination, sensor reliability, flow field coupling stability, and model adaptability are comprehensively integrated. The system control margin, model mismatch, and actuator load balancing factor are accurately quantified, providing a comprehensive and accurate state basis for the dynamic adjustment of subsequent control strategies. This lays the foundation for improving the accuracy and dynamic response speed of ventilator oxygen concentration control.

[0039] S200: Based on the aggressiveness of the adaptive adjustment parameter identification algorithm of the model mismatch, a small test signal is injected into the gas mixing system to correct the oxygen concentration transfer model online and calculate and update the model mismatch.

[0040] In this embodiment, based on the aggressiveness of the adaptive parameter identification algorithm according to the model mismatch, a micro-amplitude test signal is injected into the gas mixing system to correct the oxygen concentration transfer model online and calculate and update the model mismatch. The parameters of the oxygen concentration transfer model drift due to factors such as changes in pipeline sealing, fluctuations in patient breathing resistance, and aging of the gas mixing chamber, leading to an increased deviation between the model's predicted and actual values, thus affecting control accuracy. Therefore, it is necessary to dynamically adjust the aggressiveness of parameter identification according to the degree of mismatch and achieve online model correction through micro-amplitude test signals to ensure that the model always matches the actual operating conditions.

[0041] Step S200 in the method provided in this application embodiment includes:

[0042] Based on the current model mismatch, the convergence threshold and learning rate of the preset parameter identification algorithm are adaptively adjusted. The model mismatch is directly proportional to the learning rate, and the higher the model mismatch, the higher the convergence threshold.

[0043] A small-amplitude test signal is superimposed on the current control command, and the measured value of the oxygen concentration sensor of the gas mixing system is collected.

[0044] Using the adjusted parameter identification algorithm, the equivalent delay time and inertial time constant of the oxygen concentration transfer model are identified and updated online based on the micro-amplitude test signal and the measured value of the oxygen concentration sensor.

[0045] New short-term predictions are calculated based on the updated oxygen concentration transfer model, and the model mismatch is updated based on the mean absolute error between the new short-term predictions and the measured values ​​from the oxygen concentration sensor.

[0046] First, based on the current model mismatch, the preset convergence threshold and learning rate of the parameter identification algorithm are adaptively adjusted. The model mismatch is directly proportional to the learning rate, and the higher the model mismatch, the higher the convergence threshold. The convergence threshold is the critical error value at which the parameter identification algorithm stops iterating; a higher value indicates that the algorithm stops earlier. The learning rate is the step size when the algorithm updates the model parameters; a larger value indicates a larger adjustment range. For example, with a base learning rate of 0.05 and a base convergence threshold of 0.02, for every 0.1 increase in model mismatch, the learning rate increases by 0.03, and the convergence threshold increases by 0.01. If the current model mismatch is 0.16, then the adjusted learning rate = 0.05 + (0.16 / 0.1) × 0.03 = 0.098, and the convergence threshold = 0.02 + (0.16 / 0.1) × 0.01 = 0.036. If the mismatch subsequently rises to 0.3, resulting in a larger bias, the learning rate is adjusted to 0.05 + 0.3 × 0.3 = 0.14, and the convergence threshold is adjusted to 0.02 + 0.3 × 0.1 = 0.05 to correct the model more quickly.

[0047] Secondly, a micro-amplitude test signal is superimposed on the current control command, and the measured value of the oxygen concentration sensor in the gas mixing system is acquired. The micro-amplitude test signal is a small disturbance signal superimposed on the normal control command to avoid affecting patient ventilation safety. It is used to stimulate the dynamic response of the system to extract model parameters. For example, if the current oxygen proportional valve control command is 50% opening, a sinusoidal test signal with an amplitude of ±2% and a period of 5 seconds is superimposed, and the real-time measured value of the main oxygen sensor downstream of the gas mixing chamber is acquired simultaneously. The superimposed command sequence is [50,51,52,51,50,49,48,49,50,...], and the acquired measured oxygen concentration sequence is [40.0,40.1,40.3,40.2,40.0,39.9,39.7,39.8,40.0,...].

[0048] Furthermore, using the adjusted parameter identification algorithm, based on the micro-amplitude test signal and the measured value of the oxygen concentration sensor, the equivalent delay time and inertial time constant of the oxygen concentration transfer model are identified and updated online. The equivalent delay time refers to the lag time from the action of the proportional valve to the detection of the concentration change by the sensor, reflecting the transmission delay; the inertial time constant refers to the system's response speed from input change to output stability; the smaller the delay value, the faster the response. The micro-amplitude test signal and the measured oxygen concentration value are input into the adjusted parameter identification algorithm, such as the recursive least squares method. By fitting the dynamic relationship between input and output, the equivalent delay time and inertial time constant are updated. For example, the original model has an equivalent delay time of 0.8 seconds and an inertial time constant of 1.2 seconds; based on the test signal and measured value, the algorithm identifies the current actual delay time as 0.7 seconds and the inertial time constant as 1.1 seconds, therefore the oxygen concentration transfer model parameters are updated to: equivalent delay time 0.7 seconds and inertial time constant 1.1 seconds.

[0049] Finally, new short-term predicted values ​​are calculated based on the updated oxygen concentration transfer model, and the model mismatch is updated according to the mean absolute error between the new short-term predicted values ​​and the measured values ​​from the oxygen concentration sensor. Based on the updated oxygen concentration transfer model, the oxygen concentration for the next 10 seconds is predicted, and the mean absolute error between this prediction and the measured values ​​from the sensor during the same period is calculated as the new model mismatch. For example, the 10-second predicted sequence of the updated model is [40.0,40.2,40.1,39.9,40.3,40.1,39.8,40.0,40.2,40.1], while the measured sequence during the same period is [40.1,40.2,40.0,39.9,40.2,40.0,39.9,40.1,40.2,40.0]. The mean absolute error is (0.1+0+0.1+0+0.1+0.1+0.1+0.1+0+0.1) / 10=0.08. Therefore, the mismatch of the updated model is 0.08, which is significantly lower than the mismatch of 0.16 of the original model.

[0050] In this embodiment, by adaptively adjusting the aggressiveness of parameter identification based on model mismatch, it can quickly correct large model deviations to avoid control lag, and smoothly adjust small deviations to ensure stable ventilation. Superimposed micro-amplitude test signals can accurately extract the dynamic characteristics of the model without affecting the patient, enabling online updates of the oxygen concentration transfer model, significantly reducing model mismatch, and providing an accurate model basis for subsequent optimized control, thereby helping to improve the accuracy and dynamic response speed of oxygen concentration control.

[0051] S300: Based on the modified oxygen concentration transfer model, and constrained by the system control margin and actuator load balancing factor, an optimization objective function is constructed, and rolling optimization is performed to obtain the optimal control command sequence.

[0052] In this embodiment, based on the modified oxygen concentration transfer model and constrained by the system control margin and actuator load balancing factor, an optimization objective function is constructed and solved through rolling optimization to obtain the optimal control command sequence. The modified oxygen concentration transfer model provides a foundation for prediction, but the generation of control commands must simultaneously consider the system's current controllability and the actuator's load balancing state. If only oxygen concentration tracking accuracy is pursued, commands may exceed the system's safe control range or exacerbate local actuator losses, affecting control stability and equipment lifespan. Therefore, it is necessary to use the modified model as the prediction basis, combine the system control margin and actuator load balancing factor to construct an optimization objective function, and generate a control command sequence that is both accurate and safe and stable through rolling optimization.

[0053] Step S300 in the method provided in this application embodiment includes:

[0054] Using the modified oxygen concentration transfer model as the prediction engine, the predicted oxygen concentration within a preset time period is predicted in a rolling manner based on the current state of the gas mixing system.

[0055] The objective function is to minimize the tracking error between the predicted oxygen concentration and the set value, with the actuator load balancing factor as the penalty term and the system control margin and actuator load balancing factor as constraints.

[0056] The optimization objective function is solved using a rolling algorithm to output the optimal control command sequence that satisfies all constraints.

[0057] First, using the modified oxygen concentration transfer model as the prediction engine, the predicted oxygen concentration within a preset time period is predicted in a rolling manner based on the current gas mixing system state. Rolling prediction: In each control cycle, based on the current system state, the modified model predicts the oxygen concentration change within a preset time period. Using the modified oxygen concentration transfer model as the prediction engine, the current gas mixing system state is collected, such as a current oxygen concentration of 40.0%, an oxygen proportional valve opening of 50%, and an air proportional valve opening of 50%. The prediction time is set to 30 seconds, with each 5-second period as a sub-cycle, predicting the oxygen concentration for the next 6 sub-cycles. After each control cycle, the prediction is updated based on the new system state. For example, under the current state, the initial predicted oxygen concentration sequence for the next 30 seconds is [40.0, 40.1, 40.0, 39.9, 40.1, 40.0]; after 5 seconds, the measured oxygen concentration is 40.1%, and the prediction is re-updated based on the new state, with the sequence updated to [40.1, 40.0, 40.2, 40.1, 39.9, 40.0].

[0058] Secondly, with minimizing the tracking error between the predicted oxygen concentration and the setpoint as the optimization objective, the actuator load balancing factor as the penalty term, and the system control margin and actuator load balancing factor as constraints, an optimization objective function is constructed. The optimization objective function is a mathematical expression that includes the optimization objective, the penalty term, and the constraints. The optimization objective is to minimize the sum of squares of the tracking error between the predicted oxygen concentration and the setpoint; the penalty term = 1 - actuator load balancing factor, with a larger penalty for a smaller actuator load balancing factor, and a weight of 0.2; the constraints include: system control margin ≥ 0.5, actuator load balancing factor ≥ 0.6. The objective function expression is: Optimization objective = sum of squares of tracking error + 0.2 × (1 - load balancing factor); constraints: control command change ≤ system control margin × maximum allowable change, load balancing factor ≥ 0.6. For example, if the sum of squared tracking errors for a certain group of instructions is 0.05 and the load balancing factor of the actuator is 0.7, then the optimization target value = 0.05 + 0.2 × (1 - 0.7) = 0.05 + 0.06 = 0.11; if the sum of squared tracking errors for another group of instructions is 0.04, but the load balancing factor is 0.5, which is lower than the constraint of 0.6, then that group of instructions is excluded.

[0059] Finally, a rolling solution is performed based on the aforementioned objective function to output the optimal control command sequence that satisfies all constraints. The rolling solution refers to dividing the overall control time into multiple consecutive short periods, such as 5 seconds per period, during the optimization process. Each period only solves for the control command sequence within a preset short timeframe, such as 30 seconds, but only executes the command of the first period in the sequence. The next period, based on the new system state, re-solves for the command sequence for the next 30 seconds, repeating this process. For example, the current period might yield the oxygen proportional valve opening command sequence for the next 30 seconds as [50, 51, 50, 49, 50, 51], corresponding to 6 sub-cycles. After 5 seconds, based on the new measured oxygen concentration of 40.1% and the updated system state, a new command sequence [51, 50, 52, 51, 49, 50] is obtained through re-optimization.

[0060] In this embodiment, by using the corrected accurate model as the basis for prediction and combining the system control margin and actuator load balancing factor to construct the optimization objective function, it can ensure the accurate tracking of oxygen concentration to the set value and avoid commands exceeding the system's safe control range or causing actuator load imbalance. Rolling optimization, by dynamically updating the state and solving in stages, enables the control commands to adapt to system changes in real time, providing a high-quality original command sequence for smooth processing of subsequent commands, further improving the accuracy and dynamic response speed of oxygen concentration control.

[0061] S400: Based on the system control margin, determine the strategy for smoothing the optimal control command sequence and issue the smoothed final control command. At the same time, based on the real-time status of the updated model mismatch degree and the actuator load balancing factor, provide dynamic early warning.

[0062] In this embodiment, a strategy for smoothing the optimal control command sequence is determined based on the system control margin, and the smoothed final control command is issued. Simultaneously, dynamic early warnings are provided based on the real-time status of the updated model mismatch and actuator load balancing factor. The optimal control command sequence may contain abrupt changes; direct issuance could lead to drastic actuator movements, affecting gas mixing stability and patient ventilation experience. The system control margin reflects the current controllability, and its magnitude determines the degree of command smoothing. Excessive model mismatch reduces prediction accuracy, while excessively low actuator load balancing factors exacerbate equipment wear; both require real-time monitoring and early warning to ensure control safety and equipment reliability.

[0063] Step S400 in the method provided in this application embodiment includes:

[0064] The strategy for smoothing the optimal control command sequence based on the system control margin, and the issuance of the smoothed final control command, includes:

[0065] Take the first control instruction in the optimal control instruction sequence as the instruction to be output.

[0066] The output instruction is smoothed by a first-order low-pass filter to obtain the final control instruction. The time constant of the first-order low-pass filter is negatively correlated with the current system control margin value.

[0067] The smoothed final control command is sent to the gas proportional valve for execution.

[0068] First, the first control instruction in the optimal control instruction sequence is taken as the output instruction. The optimal sequence is [51,50,52,51,49,50]. The first instruction, 51, which is the oxygen proportional valve opening of 51%, is taken as the output instruction.

[0069] Next, the output command is smoothed using a first-order low-pass filter to obtain the final control command. The time constant of the first-order low-pass filter is negatively correlated with the current system control margin value. A first-order low-pass filter is a signal processing tool that filters abrupt changes in the control command, making the command change smoother. Time constant: A parameter that determines the smoothing effect of the first-order low-pass filter; the smaller the value, the weaker the smoothing effect, and the larger the value, the stronger the smoothing effect. A base time constant of 0.5 seconds is set. The time constant is negatively correlated with the system control margin; for every 0.1 increase in control margin, the time constant decreases by 0.1. The output command is input into the filter to calculate the final control command. For example, with a current control margin of 0.75, the time constant = 0.5 - (0.75 - 0.5) / 0.1 × 0.1 = 0.25 seconds. After inputting 51% of the output command into the filter, the final control command is smoothed to 50.5%, avoiding a direct jump from 50% to 51%.

[0070] Then, the smoothed final control command is sent to the gas proportional valve for execution. The smoothed final control command is transmitted to the gas proportional valve, which then smoothly adjusts its opening according to the command. For example, if a 50.5% opening command is issued, the oxygen proportional valve will smoothly adjust from the current 50% to 50.5% without any drastic movement.

[0071] Among these, dynamic early warning is provided based on the real-time status of both the updated model mismatch degree and the actuator load balancing factor, including:

[0072] The updated model mismatch degree calculated in real time is monitored. If the updated model mismatch degree value is continuously higher than the preset first warning threshold within a preset time window, a warning of decreased model accuracy is triggered. The value of the first warning threshold is negatively correlated with the current system control margin.

[0073] Obtain the real-time calculated actuator load balancing factor. If the actuator load balancing factor remains below the preset second warning threshold for a preset time window, trigger the actuator load imbalance warning.

[0074] The early warning information will be reported to the equipment status monitoring system.

[0075] First, the updated model mismatch degree calculated in real time is monitored. If the updated model mismatch degree value remains higher than a preset first warning threshold within a preset time window, a model accuracy decline warning is triggered. The value of the first warning threshold is negatively correlated with the current system control margin. The first warning threshold is a critical value used to determine whether the model mismatch degree is abnormal and is used to trigger the model accuracy decline warning. The updated model mismatch degree is monitored in real time to determine whether it remains higher than the first warning threshold within the preset time window. The larger the control margin, the lower the first warning threshold. For example, if the first warning threshold is 0.2, and the updated model mismatch degree remains at 0.22 for 15 seconds, which is higher than the first warning threshold, a model accuracy decline warning is triggered; if the current mismatch degree is 0.08, which is lower than the first warning threshold, a model accuracy decline warning is not triggered.

[0076] Secondly, the actuator load balancing factor is obtained in real time. If the actuator load balancing factor remains below a preset second warning threshold within a preset time window, an actuator load imbalance warning is triggered. The second warning threshold is a critical value used to determine whether the actuator load balancing factor is abnormal and to trigger the actuator load imbalance warning. The actuator load balancing factor is obtained in real time, and it is determined whether it remains below the second warning threshold within a preset time window. For example, if the second warning threshold is 0.6, and the load balancing factor remains at 0.55 for 15 seconds, below the second warning threshold, an actuator load imbalance warning is triggered; if the current load balancing factor is 0.8, above the second warning threshold, no warning is triggered.

[0077] Finally, the warning information is reported to the equipment status monitoring system. Upon triggering a warning, information such as the warning type, abnormal indicator value, and trigger time is reported to the equipment status monitoring system in real time. For example, if a model accuracy decline warning is triggered, the reported information would be: "Model accuracy decline warning, current mismatch 0.22, trigger time XX hours XX minutes," for medical or maintenance personnel to handle.

[0078] In this embodiment, a smoothing strategy negatively correlated with the system's control margin is employed to ensure gradual changes in control commands, preventing drastic actuator movements and guaranteeing patient ventilation stability. A dynamic early warning mechanism can promptly detect abnormalities such as excessive model deviation and actuator load imbalance, proactively mitigating control risks and equipment wear. The combination of smoothed commands and the early warning mechanism not only ensures the safety and reliability of the control process but also further improves the accuracy and dynamic response speed of oxygen concentration control.

[0079] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:

[0080] This application provides a model-based oxygen concentration control method for ventilators. It integrates multi-source state data to calculate control margin, model mismatch, and actuator load balancing factors, providing comprehensive state support for control. Based on the model mismatch, it dynamically adjusts parameter identification strategies and injects small-amplitude test signals to correct the model online, reducing bias. Using the corrected model as a foundation, it combines constraints to construct an optimization objective function for rolling optimization, generating an optimal control command sequence that balances accuracy and load balancing. Based on the control margin, it smooths commands and provides dynamic warnings, ensuring stable control. Ultimately, this significantly improves the accuracy and dynamic response speed of ventilator oxygen concentration control while reducing actuator wear and ensuring the safety and reliability of respiratory support.

[0081] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0082] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0083] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A model-based method for controlling ventilator oxygen concentration, characterized in that, The method includes: Based on multi-source state data within the ventilator, the system control margin is calculated in real time through nonlinear fusion, and the model mismatch of the current oxygen concentration delivery model and the actuator load balancing factor are calculated simultaneously. Based on the aggressiveness of the adaptive adjustment parameter identification algorithm for model mismatch, a small-amplitude test signal is injected into the gas mixing system to correct the oxygen concentration transfer model online and calculate and update the model mismatch. Based on the modified oxygen concentration transfer model, and constrained by the system control margin and the actuator load balancing factor, an optimization objective function is constructed, and rolling optimization is performed to obtain the optimal control command sequence. Based on the system control margin, a strategy for smoothing the optimal control command sequence is determined, and the smoothed final control command is issued. At the same time, dynamic early warning is given based on the real-time status of the update model mismatch degree and the actuator load balancing factor. Among these, based on multi-source state data within the ventilator, the system control margin is calculated in real time through nonlinear fusion, including: Oxygen concentration values ​​are collected from multiple sampling points downstream of the gas mixing chamber, and the spatial variance of the multiple oxygen concentration values ​​is calculated as the mixing uniformity index. Collect the sequence of differences between the actual opening degree and the commanded opening degree of multiple gas proportional valves operating in parallel, calculate the degree of consistency, and use it as the actuator coordination deviation; Compare the readings of the primary oxygen sensor and the backup oxygen sensor to calculate the sensor self-consistency score; Analyze the pipeline pressure fluctuation spectrum under the current ventilation flow rate, and extract the energy value of a specific frequency band as a flow-pressure coupled oscillation index; After normalizing the mixing uniformity index, actuator coordination deviation, sensor self-consistency score and flow-pressure coupling oscillation index, the results are input into the nonlinear fusion calculation unit, and the system control margin is output. This includes simultaneously calculating the model mismatch of the current oxygen concentration transfer model and the actuator load balancing factor, including: The model mismatch degree is calculated based on the average absolute error between the short-term predicted value and the actual measured value of the sensor in the oxygen concentration transfer model within a preset time window. The number and amplitude of each gas proportional valve's action within a preset time window are statistically analyzed, the coefficient of variation is calculated, the reciprocal is taken and normalized to obtain the actuator load balancing factor. Specifically, based on the aggressiveness of the adaptive adjustment parameter identification algorithm for model mismatch, a micro-amplitude test signal is injected into the gas mixing system to perform online correction of the oxygen concentration transfer model and calculate and update the model mismatch, including: Based on the current model mismatch, the convergence threshold and learning rate of the preset parameter identification algorithm are adaptively adjusted. The model mismatch is directly proportional to the learning rate, and the higher the model mismatch, the higher the convergence threshold. A small-amplitude test signal is superimposed on the current control command, and the measured value of the oxygen concentration sensor of the gas mixing system is collected. Using the adjusted parameter identification algorithm, the equivalent delay time and inertial time constant of the oxygen concentration transfer model are identified and updated online based on the micro-amplitude test signal and the measured value of the oxygen concentration sensor. New short-term predictions are calculated based on the updated oxygen concentration transfer model, and the model mismatch is updated based on the mean absolute error between the new short-term predictions and the measured values ​​from the oxygen concentration sensor. Based on the modified oxygen concentration transfer model, and constrained by the system control margin and actuator load balancing factor, an optimization objective function is constructed and solved through rolling optimization to obtain the optimal control command sequence, including... Using the modified oxygen concentration transfer model as the prediction engine, the predicted oxygen concentration within a preset time period is predicted in a rolling manner based on the current state of the gas mixing system. The optimization objective is to minimize the tracking error between the predicted oxygen concentration and the set value, with the actuator load balancing factor as the penalty term and the system control margin and actuator load balancing factor as constraints. The optimization objective function is solved using a rolling algorithm to output the optimal control command sequence that satisfies all constraints.

2. The method according to claim 1, characterized in that, Based on the system control margin, a strategy for smoothing the optimal control command sequence is determined, and the smoothed final control commands are issued, including: Take the first control instruction in the optimal control instruction sequence as the instruction to be output. The output instruction is smoothed by a first-order low-pass filter to obtain the final control instruction. The time constant of the first-order low-pass filter is negatively correlated with the current system control margin value. The smoothed final control command is sent to the gas proportional valve for execution.

3. The method according to claim 1, characterized in that, Simultaneously, based on the real-time status of the updated model mismatch degree and the actuator load balancing factor, dynamic early warnings are provided, including: The updated model mismatch degree calculated in real time is monitored. If the updated model mismatch degree value is continuously higher than the preset first warning threshold within a preset time window, a warning of decreased model accuracy is triggered. The value of the first warning threshold is negatively correlated with the current system control margin. Obtain the real-time calculated actuator load balancing factor. If the actuator load balancing factor remains below the preset second warning threshold for a preset time window, trigger the actuator load imbalance warning. The early warning information will be reported to the equipment status monitoring system.

Images