Plateau oxygen supply controller and method based on event trigger learning mechanism

Abstract

The invention discloses a plateau oxygen supply controller and method based on an event-triggered learning mechanism, and belongs to the technical field of plateau oxygen supply. The plateau oxygen supply controller comprises a blood oxygen saturation acquisition module and an event-triggered learning controller module of oxygen supply concentration; an oxyhemoglobin saturation SpO2 sequence is obtained through an oxyhemoglobin saturation acquisition module, and an arterial blood oxygen partial pressure PaO2 sequence is obtained through calculation according to an oxygen separation curve equation; the oxygen supply concentration percentage FiO2 is obtained; and the learning controller module is triggered by the event of the oxygen supply concentration to optimize the oxygen supply concentration percentage FiO2 through the minimum quadratic performance index. According to the plateau oxygen supply controller and method based on the event-triggered learning mechanism, the oxygen supply controller is designed for the resource-limited plateau hypoxia exercise process, so that the oxyhemoglobin saturation is effectively adjusted, and strong adverse reactions occurring in the hypoxia exercise process are relieved.

Description

Technical Field

[0001] The present invention relates to the technical field of plateau oxygen supply, and in particular to a plateau oxygen supply controller and method based on an event-triggered learning mechanism. Background Art

[0002] my country's plateau region is vast and strategically located, which is of great strategic significance to national security and economic construction. With the development of economy and society, the demand for economic construction, leisure travel and other activities in the plateau region is growing, which has led to an increasing number of people entering the plateau region to engage in related activities; in particular, emergency tasks such as border security maintenance and disaster relief support require a large number of personnel to enter the plateau in a very short time to carry out high-intensity work.

[0003] The plateau environment has the characteristics of low air pressure, low oxygen concentration, strong radiation, etc. Among them, atmospheric pressure and oxygen concentration directly affect human physiological reactions. For example, high-altitude hypoxia will cause local pathological changes in human breathing or cause organ damage. When the physiological reaction exceeds the tolerance limit, it will seriously affect people's life and health. Experimental studies have shown that acute hypoxia, mild hypoxia, and moderate hypoxia will have a negative impact on people's neurocognitive memory function, such as short-term memory confusion, distraction, disordered thinking, decreased judgment, etc., and the symptoms of hypoxia will significantly increase with the increase of hypoxia. Studies have shown that when the brain tissue is acute to moderately hypoxic, the movement attention space breadth and spatial transfer ability of the human brain tissue will be severely reduced. However, the effects of hypoxia are reversible before reaching the limit, and timely restoration of oxygen supply can restore cognitive function to normal levels. The process of hypoxia stress in the human body varies greatly from individual to individual. For individuals who have strong adverse reactions in the early stage of hypoxia acclimatization, it is necessary to alleviate them in time through artificial oxygen supply system to avoid irreversible damage.

[0004] Currently, the manufacturing technology of oxygen supply equipment is relatively mature, but the oxygen supply mechanism is mostly manually adjusted or pre-set, and it is impossible to supply oxygen according to the needs of the human body. Therefore, the research on adaptive adjustment of oxygen supply based on dynamic blood oxygen monitoring information has received widespread attention.

[0005] Gajdos et al. (Gajdos, Marek, et al., Effects of a new device for automated closed loop control of inspired oxygen concentration on fluctuations of arterial and different regional organ tissue oxygen saturations in preterm infants, Archives of Disease in Childhood-Fetal and Neonatal Edition 104.4 (2019): F360-F365.) evaluated the effects of closed-loop control of inspired oxygen fraction on arterial changes and regional tissue oxygen saturation changes through a randomized crossover experiment. Schwaiberger et al. (Schwaiberger, David, et al., Closed-loop mechanical ventilation for lung injury: a novel physiological-feedback mode following the principles of the open lung concept, Journal of Clinical Monitoring and Computing 32 (2018): 493-502.) tested a new fully closed-loop mechanical ventilation algorithm in the treatment of acute respiratory distress syndrome to increase the recruitment and ventilation of the dorsal lung region and improve the arterial oxygenation index. Morozoff et al. (Morozoff, Edmund, John A. Smyth, and Mehrdad Saif, Applying computer models to realize closed-loop neonatal oxygen therapy, Anesthesia & Analgesia 124.1 (2017): 95-103.) proposed a method for mathematical modeling of mechanical equipment such as air-oxygen mixers and pulse oximeters in the context of automated oxygen therapy, and proposed a blood oxygen adaptive control model.Brogi et al. (Brogi, Etrusca, et al, Clinical performance and safety of closed-loop systems: a systematic review and meta-analysis of randomized controlled trials, Anesthesia & Analgesia 124.2 (2017): 446-455.) conducted a randomized controlled trial on the accuracy and safety of closed-loop control versus manual control. It was observed that closed-loop systems tend to better control a given target within a selected range.

[0006] Designing wearable devices for plateau environments is challenging. First, due to factors such as size and weight, the computing and storage capabilities of wearable devices are relatively limited, which brings additional challenges to the online evaluation of plateau physiological indicators based on historical data of physiological information such as heart rate and blood oxygen saturation; second, existing commercial hypoxia training devices can only operate according to pre-designed open-loop training plans. Designing an embedded controller that matches this type of training device and has communication and control capabilities is the basic prerequisite for ensuring the implementation of the closed-loop control strategy of hypoxia pre-training clothing. Finally, the plateau environment conditions are complex, and it is necessary to ensure that the performance of the wearable device controller is not overly affected when computing and communication resources are limited.

[0007] Based on the above analysis, this patent designs a data-driven oxygen supply system based on an event-triggered learning mechanism for the plateau hypoxia acclimatization process. Event-triggered learning refers to learning and updating the controller only when necessary by designing trigger conditions, thereby saving computing and communication resources. The event-triggered learning method can reduce the power consumption of the control system during operation and ensure the performance of the system while saving communication and computing resources as much as possible. Data-driven control refers to the control method of designing the controller by directly using the input and output data of the system without using explicit or implicit information of the mathematical model of the control system. The change in blood oxygen saturation SpO2 concentration is mainly affected by the change in external oxygen concentration. After oxygen enters the human body, it will undergo a series of complex physiological processes, and there are large differences in the functions of pulmonary ventilation, gas transport, tissue material exchange, etc. among different individuals. Therefore, it is difficult to accurately mathematically model each functional mechanism. In the absence of an accurate system model, it is very important to use online or offline data to directly design the controller, predict and evaluate the system status. Therefore, this patent establishes a data-driven model for human blood oxygen saturation without the need to measure parameters such as the time constant of the human respiratory system of different users or the sensitivity coefficient of arterial oxygen partial pressure to oxygen supply in advance, thereby ensuring that the oxygen supply system can adapt to more common high-altitude hypoxia usage scenarios and individual differences. Summary of the invention

[0008] The purpose of the present invention is to provide a plateau oxygen supply controller and method based on an event-triggered learning mechanism, and to design an oxygen supply controller for the resource-limited plateau hypoxia acclimatization process, so as to effectively regulate blood oxygen saturation and alleviate the strong adverse reactions occurring during the hypoxia acclimatization process.

[0009] To achieve the above purpose, the present invention provides a plateau oxygen supply control method based on an event-triggered learning mechanism, which is as follows:

[0010] S1. Obtain the blood oxygen saturation SpO2 sequence through the blood oxygen saturation acquisition module, and calculate the arterial blood oxygen partial pressure PaO2 sequence according to the oxygen dissociation curve equation;

[0011] S2, obtaining the oxygen concentration percentage FiO2;

[0012] S3. Utilize the oxygen supply concentration event to trigger the learning controller module to optimize the oxygen supply concentration percentage FiO2 through the minimum quadratic performance index.

[0013] Preferably, the process of obtaining the blood oxygen saturation SpO2 sequence in S1 is as follows:

[0014] The LED light source emits red light and near-infrared light, which penetrates the multi-component mixed system, is scattered and absorbed by human tissue in the light path, and is detected by the photodetector at the end of the light path. The signal in the photodetector is processed by the signal processing circuit and converted into a digital signal that can be processed by the host computer through AD sampling.

[0015] According to the timing of LED light emission, the blood oxygen information of each wavelength in the channel of each blood oxygen saturation acquisition module is separated, and the blood oxygen value is calculated using spatially resolved spectroscopy technology and continuous wave spectroscopy technology. The blood oxygen saturation SpO2 sequence is indirectly measured based on the calculated blood oxygen value.

[0016] Preferably, the oxygen dissociation curve equation in S1 is as follows:

[0017]

[0018] Among them, PaO2(t) represents the arterial oxygen partial pressure within time t, and SpO2(t) represents the blood oxygen saturation within time t. It can be seen from the above formula that there is a positive correlation between the arterial oxygen partial pressure PaO2 and the blood oxygen saturation SpO2. The SpO2 sequence can be substituted into the above formula to obtain the PaO2 sequence as the state quantity of the system.

[0019] Preferably, the process for obtaining the S2 oxygen supply concentration percentage FiO2 is as follows:

[0020] Arterial oxygen partial pressure PaO2 is determined by the oxygen supply concentration percentage FiO2. The relationship between arterial oxygen partial pressure PaO2 and oxygen supply concentration percentage FiO2 can be expressed by the following first-order linear differential equation:

[0021]

[0022] Among them, FiO2(t) represents the percentage of oxygen supply concentration within time t, τ p Represents the time constant of the human respiratory system, G p It represents the sensitivity coefficient of arterial oxygen partial pressure to oxygen supply.

[0023] Preferably, the event-triggered learning controller module of the oxygen supply concentration in S3 uses the Actor-Critic method of reinforcement learning, using the function approximator on the Critic side to approximate the Q function, while the approximator on the Actor side is implemented by a zero-order holder to approximate the optimal controller.

[0024] Preferably, a cost function is provided in the event-triggered learning controller module of the oxygen supply concentration, and the cost function calculates the control signal of the oxygen supply concentration according to the minimum distance between the measured blood oxygen saturation SpO2 sequence and the blood oxygen saturation SpO2 target interval, and the cost function J is:

[0025]

[0026] Among them, x(t) is the state variable of the system, which represents the arterial oxygen partial pressure PaO2(t) of the system in this method, x(0) is the initial value of the system state at t=0, u(t) is the control input of the system, which represents the oxygen supply concentration percentage FiO2(t) at time t in this method, and the superscript (·) T represents the transpose of a matrix or vector, τ represents integration over time, Q is the weight matrix of blood oxygen saturation, and R is the weight matrix of oxygen supply concentration control variables. The optimization goal of the oxygen supply system design is to minimize the quadratic performance index.

[0027] Preferably, in the oxygen supply concentration event-triggered learning controller module, the trigger threshold for updating the oxygen supply concentration is as follows:

[0028]

[0029] Among them, Q is the weight matrix of blood oxygen saturation, R is the weight matrix of oxygen supply concentration control variable, e u (t) is the error between the current oxygen supply concentration setting value and the oxygen supply concentration value calculated based on real-time data, λ (·) represents the smallest eigenvalue of the matrix, represents the maximum eigenvalue of the matrix, L1 is an artificial constant used to adjust the trigger efficiency of the algorithm, It indicates the oxygen supply concentration value under the event trigger mechanism. It is only updated at the trigger moment and maintains the concentration value at the last trigger moment at the non-trigger moment.

[0030] The present invention also provides a plateau oxygen supply controller based on an event-triggered learning mechanism, comprising:

[0031] Blood oxygen saturation acquisition module: indirectly measures the blood oxygen saturation SpO2 sequence based on the absorption characteristics of hemoglobin in the blood to light of different wavelengths, and calculates the arterial blood oxygen partial pressure PaO2 sequence based on the oxygen dissociation curve equation;

[0032] Event-triggered learning controller module of oxygen supply concentration: The human body's arterial oxygen partial pressure PaO2 sequence and oxygen supply concentration percentage FiO2 are used as the state quantity and input quantity of the plateau oxygen supply controller system respectively, and the minimum quadratic performance index is designed to optimize the oxygen supply concentration percentage FiO2 value in the plateau oxygen supply controller system.

[0033] Therefore, the present invention adopts the above-mentioned plateau oxygen supply controller and method based on event-triggered learning mechanism to design an oxygen supply controller for the resource-limited plateau hypoxia acclimatization process, thereby effectively regulating blood oxygen saturation and alleviating the strong adverse reactions occurring during the hypoxia acclimatization process.

[0034] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is an oxygen dissociation curve relationship diagram between arterial blood oxygen partial pressure and blood oxygen saturation of a plateau oxygen supply controller and method embodiment based on an event-triggered learning mechanism of the present invention;

[0036] Figure 2 It is a structural schematic diagram of a closed-loop event-triggered oxygen supply control module of a plateau oxygen supply controller and method embodiment based on an event-triggered learning mechanism of the present invention;

[0037] Figure 3 It is a schematic diagram of the triggering update of the control quantity FiO2 of an embodiment of a plateau oxygen supply controller and method based on an event-triggered learning mechanism of the present invention. DETAILED DESCRIPTION

[0038] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0039] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0040] Embodiment 1

[0041] Artificial oxygen supply equipment can form a gas pressure environment by supplying oxygen in plateau environments, create a microenvironment for the human body, protect the human body from environmental factors such as low pressure, and adjust it through human intervention to protect life safety. This type of equipment is mainly composed of an artificial oxygen mask, an oxygen regulator and an oxygen source, and is a comprehensive respiratory health maintenance system.

[0042] First, blood oxygen saturation acquisition module

[0043] Wearable devices measure blood oxygen saturation non-invasively through photoelectric measurement technology. This technology is based on the principle of pulse oximetry monitoring, which indirectly measures SpO2 based on the absorption characteristics of hemoglobin in the blood to light of different wavelengths.

[0044] The LED light source emits red light and near-infrared light, which penetrates the multi-component mixed system such as skin, subcutaneous tissue, bone, brain tissue, etc. After being scattered and absorbed by human tissue in the light path, it is detected by the photodetector at the end of the light path. The signal in the photodetector is processed by the signal processing circuit and converted into a digital signal that can be processed by the host computer through AD sampling. According to the timing of LED light emission, the blood oxygen information of each wavelength of each blood oxygen collection channel is separated, and the calculation of blood oxygen value is completed using spatial resolution spectroscopy technology and continuous wave spectroscopy technology. According to the above process, the SpO2 measurement sequence can be obtained, which can be used for the design of oxygen supply concentration controller.

[0045] The oxygen dissociation curve is a curve that describes the binding and dissociation relationship between hemoglobin and oxygen molecules. It shows the affinity of hemoglobin for binding to oxygen. The shape of the oxygen dissociation curve is affected by many factors, including temperature, pH value, carbon dioxide partial pressure and 2,3-diphosphodiesterone (DPG). A typical oxygen dissociation curve is as follows: Figure 1 shown.

[0046] According to the oxygen dissociation curve equation, the arterial oxygen partial pressure PaO2 and blood oxygen saturation SpO2 at any time have the following relationship:

[0047]

[0048] Among them, PaO2(t) represents the arterial oxygen partial pressure within time t, and SpO2(t) represents the blood oxygen saturation within time t. It can be seen from the above formula that there is a positive correlation between PaO2 and SpO2. The SpO2 sequence can be substituted into the above formula to obtain the PaO2 sequence as the state quantity of the system. The arterial oxygen partial pressure PaO2 is determined by the oxygen supply, and the relationship between it and the oxygen supply can be expressed by the following first-order linear differential equation:

[0049]

[0050] Among them, FiO2(t) represents the percentage of oxygen supply concentration within time t, τ p Represents the time constant of the human respiratory system, G p It represents the sensitivity coefficient of arterial oxygen partial pressure to oxygen supply. As long as τ is determined p and G p These two parameters can establish the relationship between oxygen supply and blood oxygen saturation. Taking into account individual differences, the above two parameters need to be measured according to the different conditions of the subjects using the oxygen supply system.

[0051] In order to obtain more accurate parameters, traditional methods rely on designing medium- and long-term behavioral experiments for fitting. Considering that the adverse reactions caused by plateau hypoxia stress may cause irreversible effects on individuals, this scheme regards the above parameters as unknown quantities and solves the control input, namely the oxygen supply FiO2, through a model-free event-triggered learning method.

[0052] Second, the event-triggered learning algorithm of oxygen supply concentration

[0053] This scheme is to build an event-triggered learning framework. Through the relationship between the oxygen dissociation curve and SpO2, the human arterial oxygen partial pressure PaO2 and the oxygen supply concentration percentage FiO2 are used as the state quantity and input quantity of the system respectively, and the minimum quadratic performance index is designed to optimize the FiO2 value in the oxygen supply system.

[0054] Consider the following linear system:

[0055]

[0056] in, represents the time derivative of the system state, is the system state, whose initial value is x(0)=x0, in this scheme x(t)=PaO2(t), It is the control input of the system. In this scheme, u(t)=FiO2(t). and The system matrices corresponding to the controlled object and input quantity are The optimization goal of oxygen supply system design is to minimize the secondary performance index. The secondary performance index J is:

[0057]

[0058] Where x(0) is the initial value of the system state at t = 0, u(t) is the control input of the system, and in this method, it represents the oxygen concentration percentage FiO2(t) at time t. The superscript (·) Trepresents the transpose of a matrix or vector, τ represents integration by time, Q is a positive semidefinite matrix, and R is a positive definite matrix. In combination with the practical significance of the present invention, Q is a weight matrix of blood oxygen saturation, R is a weight matrix of oxygen supply concentration control variables, and J is a cost function.

[0059] This scheme starts from the Actor-Critic perspective of reinforcement learning and designs an event-triggered learning algorithm that adjusts the parameters of the relevant value function online. The Actor-Critic structure of this scheme is set to use the function approximator on the Critic side to approximate the Q function, while the approximator on the Actor side is implemented through a zero-order holder to approximate the optimal controller. Specifically, It can be expressed as:

[0060]

[0061] where x is the system state, is the optimal control input, and U is the current control input u and x. d is the data matrix of row vectors, H xx is the matrix coefficient of the Q function corresponding to the system state, is x and u d The matrix coefficients corresponding to the Q function between yes The transpose of is the matrix coefficient of the Q function corresponding to the control input, is a positive definite matrix with the above four matrices as submatrices. vech is an operator for semi-vectorizing the symmetric matrix. Its function is to delete the lower part of the symmetric matrix and then vectorize the remaining part column by column. For an n-dimensional real symmetric matrix The result obtained after the vech operation is vech(A)=[A 1,1 …A n,1 A 2,2 …A n,2 …A n-1,n-1 A n-1, n A n,n ] T .symbol is the Kronecker product operator, which is used to find the Kronecker product between two matrices.

[0062] The ideal critic weight is W c , since the critic side function approximator is used to approximate the Q function, let Then there is The ideal Actor side weight is W a , since the Actor-side function approximator is used to approximate the controller Therefore, there is

[0063] Since the human body's blood oxygen regulation process is difficult to accurately mathematically model, all the information of the system matrix is ​​considered unknown. We can only obtain the information through the blood oxygen saturation state estimation. The above theoretical weights cannot be directly calculated. Therefore, it is necessary to estimate the learning weights. Based on the above analysis, the weight estimator on the Critic side is defined as:

[0064]

[0065] in, It is the representation of the critic side weight estimator from the event triggering perspective. It is the transpose of the Critic side weight matrix from the event triggering perspective.

[0066] The weight estimator on the Actor side satisfies:

[0067]

[0068] in, It is the representation of the Actor-side weight estimator from the perspective of event triggering. It is the transpose of the Actor-side weight matrix from the event triggering perspective. is the current control input from the perspective of event triggering. The design idea of ​​this paper is to update the Actor-side weight only at the triggering moment, and the state information used is all known. Actor-side weight estimator Defined as:

[0069]

[0070] Among them, r j represents the jth triggering moment, r j+1 Indicates the j+1th triggering moment; Represents the Actor side weight at time t.

[0071] From this, we can get the form of control strategy from the perspective of event-triggered learning, namely: in is the estimated value of the control input from the event-triggered learning perspective.

[0072] The weight error e between the Actor and Critic a 、e c They are defined as:

[0073]

[0074] in, It is the transpose of the Actor-side weight matrix from the event triggering perspective. is the inverse matrix of the matrix coefficients of the Q function obtained by substituting the estimated control input value, is the matrix coefficient of the Q function obtained by substituting the estimated values ​​of the control input and the system state, x(r j ) is the jth triggering moment r j t is the current time, T is the time constant set by the user to adjust the performance of the algorithm, is the matrix corresponding to the estimated value of the Q function.

[0075] Therefore, the goal is to find a suitable weight iteration method so that the defined weight error e a With e c As the learning process approaches zero. Combined with adaptive control theory, the square norm of these errors can be defined as:

[0076]

[0077] Among them, K a Represents the square norm of the error on the Actor side, K c Represents the square norm of the error on the critic side.

[0078] The iterative update of weights follows the following two formulas:

[0079]

[0080] in, is the derivative of the critic side weight with respect to time, is the transpose of the error vector on the critic side, is the Actor-side weight under the event triggering mechanism, is the derivative of the Actor-side weight with respect to time under the event triggering mechanism, is the Actor-side weight when there is no event trigger mechanism. is the transpose of the error vector on the Actor side; to simplify the expression, α c is a positive real constant defined by the user to adjust the learning convergence rate on the Critic side, α a It is a positive real constant defined by the user to adjust the learning convergence rate on the Actor side. The calculation rate of the learning algorithm will increase with the increase of these two constant values, but more computing resources will be consumed accordingly. When applying, the above parameters can be dynamically adjusted according to the actual needs of the plateau for control efficiency.

[0081] Since different users have different physiological indicators, it is difficult to adapt to the differences in individual conditions in the plateau hypoxic environment by designing a controller with a fixed event trigger threshold. At the same time, the data-driven control method proposed in this patent does not measure parameters such as the time constant of the human respiratory system or the sensitivity coefficient of arterial oxygen partial pressure to oxygen supply of different users in advance. Instead, it directly constructs a controller using measurement information containing physiological indicators of different individuals. Therefore, it is impossible to artificially set a fixed trigger threshold based on expert experience and the individual conditions of the user. Based on the above considerations, this scheme designs a new event trigger threshold condition, which can dynamically update the trigger threshold according to the real-time collected blood oxygen information to ensure that the controller can adapt to different usage scenarios.

[0082] Trigger error u (t) is defined as:

[0083]

[0084] The trigger threshold is as follows:

[0085]

[0086] Among them, Q is the weight matrix of blood oxygen saturation, R is the weight matrix of oxygen supply concentration control variable, e u (t) is the error between the current oxygen supply concentration setting value and the oxygen supply concentration value calculated based on real-time data, λ (·) represents the smallest eigenvalue of the matrix, represents the maximum eigenvalue of the matrix, L1 is an artificial constant used to adjust the trigger efficiency of the algorithm, It is the current control input from the event triggering perspective.

[0087] The trigger conditions proposed in this scheme can ensure the Lyapunov stability of the learning results. When the oxygen supply system is working, the above threshold conditions are used to determine whether to update the controller to update the oxygen supply concentration only when necessary. This has important application value in plateaus where oxygen resources and communication resources are limited. The execution process of the algorithm is as follows: Figure 2 shown.

[0088] In the specific implementation, the initial sequence of blood oxygen saturation SpO2 is first obtained through historical physiological data measurement information, and then used for the design of the oxygen supply concentration controller; in real-time control, the SpO2 sequence is substituted into the oxygen dissociation curve equation to obtain the PaO2 sequence as the state quantity x(t) of the system; then, u(t)=FiO2(t) is obtained through the event-triggered learning algorithm of the oxygen supply concentration, and the oxygen supply concentration of the oxygen supply device at the current moment is set; finally, the user's new physiological data is collected, and the blood oxygen saturation SpO2 sequence is updated for the next control cycle. The present invention has carried out performance verification analysis and evaluation on the above method. The update of the event-triggered oxygen supply system concentration controller is as follows: Figure 3 The model-free event-triggered learning algorithm designed in this study has good adaptability to different initial states, while ensuring control accuracy and reducing the number of network updates under the premise of ensuring the stability of the closed-loop system, thus achieving a reasonable compromise between control performance and transmission performance.

[0089] Therefore, the present invention adopts the above-mentioned plateau oxygen supply controller and method based on event-triggered learning mechanism, which has the following beneficial effects:

[0090] (1) The relationship between blood oxygen saturation and oxygen supply concentration was constructed based on the oxygen dissociation curve, and a data-driven controller was designed based on this. Without knowing the user's respiratory system time constant and other physiological indicators, the oxygen supply concentration can be set directly based on the data collected during actual use to adapt to individual differences.

[0091] (2) Reinforcement learning technology and event-triggered control ideas are applied to the controller design of the oxygen supply system. According to the state estimation results of blood oxygen saturation SpO2, the oxygen supply concentration percentage FiO2 is dynamically adjusted through the event-triggered learning mechanism. The number of controller updates is reduced while ensuring the stability of the closed-loop oxygen supply system, achieving a reasonable compromise between control performance and transmission performance, thereby reducing communication load and power consumption.

[0092] (3) Design an oxygen supply system based on real-time blood oxygen collection information, and achieve refined use of oxygen resources through closed-loop control. While ensuring that the human body's blood oxygen saturation is within the normal range, it also extends the service life of the equipment and improves its ability to protect human health.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A high-altitude oxygen supply control method based on event-triggered learning mechanism, characterized in that: The details are as follows: S1. Obtain the blood oxygen saturation SpO2 sequence through the blood oxygen saturation acquisition module, and calculate the arterial blood oxygen partial pressure PaO2 sequence according to the oxygen dissociation curve equation; S2, obtaining the oxygen concentration percentage FiO2; S3. Utilize the oxygen supply concentration event to trigger the learning controller module to optimize the oxygen supply concentration percentage FiO2 through the minimum quadratic performance index.

2. A high-altitude oxygen supply control method based on event-triggered learning mechanism according to claim 1, characterized in that: The process of obtaining the blood oxygen saturation SpO2 sequence in S1 is as follows: The LED light source emits red light and near-infrared light, which penetrates the multi-component mixed system, is scattered and absorbed by human tissue in the light path, and is detected by the photodetector at the end of the light path. The signal in the photodetector is processed by the signal processing circuit and converted into a digital signal that can be processed by the host computer through AD sampling. According to the timing of LED light emission, the blood oxygen information of each wavelength in the channel of each blood oxygen saturation acquisition module is separated, and the blood oxygen value is calculated using spatially resolved spectroscopy technology and continuous wave spectroscopy technology. The blood oxygen saturation SpO2 sequence is indirectly measured based on the calculated blood oxygen value.

3. A high-altitude oxygen supply control method based on event-triggered learning mechanism according to claim 1, characterized in that: The oxygen dissociation curve equation in S1 is as follows: Among them, PaO2(t) represents the arterial oxygen partial pressure at time t, and SpO2(t) represents the blood oxygen saturation at time t. It can be seen from the above formula that there is a positive correlation between the arterial oxygen partial pressure PaO2 and the blood oxygen saturation SpO2. The SpO2 sequence can be substituted into the above formula to obtain the PaO2 sequence as the state quantity of the system.

4. A high-altitude oxygen supply control method based on event-triggered learning mechanism according to claim 3, characterized in that: The process of obtaining the S2 oxygen supply concentration percentage FiO2 is as follows: Arterial oxygen partial pressure PaO2 is determined by the oxygen supply concentration percentage FiO2. The relationship between arterial oxygen partial pressure PaO2 and oxygen supply concentration percentage FiO2 can be expressed by the following first-order linear differential equation: Among them, FiO2(t) represents the percentage of oxygen supply concentration at time t, τ p Represents the time constant of the human respiratory system, G p It represents the sensitivity coefficient of arterial oxygen partial pressure to oxygen supply.

5. The high-altitude oxygen supply control method based on event-triggered learning mechanism according to claim 1, characterized in that: The event-triggered learning controller module of the oxygen supply concentration in S3 uses the Actor-Critic method of reinforcement learning, and uses the function approximator on the Critic side to approximate the Q function, while the approximator on the Actor side is implemented by a zero-order holder to approximate the optimal controller.

6. A high-altitude oxygen supply control method based on event-triggered learning mechanism according to claim 5, characterized in that: The event-triggered learning controller module of the oxygen supply concentration is provided with a cost function, which calculates the control signal of the oxygen supply concentration according to the minimum distance between the measured blood oxygen saturation SpO2 sequence and the blood oxygen saturation SpO2 target interval. The cost function J is: Among them, x(t) is the state variable of the system, which represents the arterial oxygen partial pressure PaO2(t) of the system in this method, x(0) is the initial value of the system state at t=0, u(t) is the control input of the system, which represents the oxygen supply concentration percentage FiO2(t) at time t in this method, and the superscript (·) T represents the transpose of a matrix or vector, τ represents integration over time, Q is the weight matrix of blood oxygen saturation, and R is the weight matrix of oxygen supply concentration control variables; the optimization goal of the oxygen supply system design is to obtain the minimum quadratic performance index.

7. The high-altitude oxygen supply control method based on event-triggered learning mechanism according to claim 5, characterized in that: In the oxygen supply concentration event-triggered learning controller module, the trigger threshold for updating the oxygen supply concentration is shown in the following formula: Among them, Q is the weight matrix of blood oxygen saturation, R is the weight matrix of oxygen supply concentration control variable, e u (t) is the error between the current oxygen supply concentration setting value and the oxygen supply concentration value calculated based on real-time data, λ(·) represents the minimum eigenvalue of the matrix, represents the maximum eigenvalue of the matrix, L1 is an artificial constant used to adjust the trigger efficiency of the algorithm, It indicates the oxygen supply concentration value under the event trigger mechanism. It is only updated at the trigger moment and maintains the concentration value at the last trigger moment at the non-trigger moment.

8. A plateau oxygen supply controller based on event-triggered learning mechanism, characterized in that: include: Blood oxygen saturation acquisition module: indirectly measures the blood oxygen saturation SpO2 sequence based on the absorption characteristics of hemoglobin in the blood to light of different wavelengths, and calculates the arterial blood oxygen partial pressure PaO2 sequence based on the oxygen dissociation curve equation; Event-triggered learning controller module of oxygen supply concentration: The human body's arterial oxygen partial pressure PaO2 sequence and oxygen supply concentration percentage FiO2 are used as the state quantity and input quantity of the plateau oxygen supply controller system respectively, and the minimum quadratic performance index is designed to optimize the oxygen supply concentration percentage FiO2 value in the plateau oxygen supply controller system.

Images