An individualized high altitude hypoxia pre-acclimatization closed-loop intervention algorithm

Abstract

The application discloses an individualized highland hypoxia pre-service closed-loop intervention algorithm, relates to the technical field of acute highland disease intervention, and comprises the following steps: S1: continuous data of oxygen supply concentration and blood oxygen saturation during IHT is obtained, and the data set is divided into a training set, a verification set and a test set; S2: based on the training set, a group prediction model of blood oxygen saturation is constructed through a Krylov operator, model updating and prediction are carried out through online iteration, a blood oxygen saturation prediction value is output, and based on the verification set, the group prediction model is used for individualization of the model; S3: blood oxygen saturation is predicted based on the individualized model, a lower-layer optimization cost function taking the low oxygen concentration as a variable is constructed, and upper-layer optimization design of safety constraint self-adaptation is carried out; S4: the training result is evaluated, and the evaluation result is fed back to the step S3; dynamic evaluation indexes are calculated, the low oxygen stimulation intensity of the IHT is adjusted through control optimization strategy, and the rapidity and effectiveness of highland adaptation capability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of acute high altitude disease intervention, and in particular to an individualized high altitude hypoxia pre-exposure closed-loop intervention algorithm. BACKGROUND

[0002] Acute high altitude disease is one of the high altitude diseases, and the risk of the disease can be reduced by intermittent hypoxic training (IHT), a pre-exposure method. IHT is a training method that can effectively improve the body's adaptation to high altitude environment by stimulating the body's tissues to undergo a series of complex adaptive changes through periodic hypoxia. During the training, key continuous physiological indicators can be detected and recorded in real time by wearable devices, and used to evaluate the training level and hypoxic adaptation ability. Among them, blood oxygen saturation is considered the most direct and effective physiological indicator, followed by heart rate. A large number of medical studies use these two indicators and their variant indicators as the evaluation criteria for high altitude environment adaptation ability.

[0003] However, the effectiveness of IHT depends on the training protocol developed. Within the individual's tolerance range, a greater intensity and longer duration of hypoxic stimulation can more effectively improve the adaptation ability. If the intensity set exceeds the tolerance range, it will have the opposite effect, and even cause irreversible damage. Existing high altitude disease research based on IHT is limited to the evaluation of hypoxic tolerance and the exploration of its adaptive improvement ability. It has not considered how to design an optimal decision algorithm to dynamically adjust the IHT training strategy to achieve rapid adaptation to hypoxic environment and improve the ability to go to high altitude.

[0004] Specifically, the hypoxic strategy used during an IHT is fixed, so the training is essentially an open-loop structure, and there is no feedback loop to effectively adjust the strategy based on the evaluation results. Considering the specificity between individuals and within individuals, the same fixed oxygen supply protocol may not be universally applicable to all individuals.

[0005] Therefore, an individualized high altitude hypoxia pre-exposure closed-loop intervention algorithm is provided to solve the above problems. SUMMARY

[0006] The purpose of the present application is to provide an individualized high altitude hypoxia pre-exposure closed-loop intervention algorithm to accelerate the training period of individuals in the plain area, improve the hypoxic tolerance and reduce the risk of acute high altitude disease.

[0007] To achieve the above purpose, the present application provides an individualized high altitude hypoxia pre-exposure closed-loop intervention algorithm, comprising the following steps:

[0008] S1: periodically oxygenate the subject during IHT, collect physiological signals in real time, obtain continuous data of oxygenation concentration and blood oxygen saturation during IHT, and divide the continuous data of blood oxygen saturation into a training set, a validation set and a test set according to different subjects;

[0009] S2: based on the training set, learning the variation rule of blood oxygen saturation under intermittent hypoxia through the Koopman operator, constructing a population prediction model of blood oxygen saturation, updating and predicting the model through online iteration, outputting the blood oxygen saturation prediction value, and individualizing the model based on the validation set using the population prediction model;

[0010] S3: using the individualized model to predict blood oxygen saturation, constructing a lower layer optimization cost function with low oxygen concentration as a variable through a low oxygen concentration double-layer optimization module, and designing a safety-constrained adaptive upper layer optimization according to the historical learning performance;

[0011] S4: evaluating the training results through a training effect evaluation module, and feeding back the evaluation results to the low oxygen concentration double-layer optimization module.

[0012] Preferably, in step S2, based on the training set, the variation rule of blood oxygen saturation under intermittent hypoxia is learned through the Koopman operator, and a population prediction model of blood oxygen saturation is constructed, which specifically includes the following steps:

[0013] Step 1: obtain a nonlinear form dynamic characteristic equation of blood oxygen saturation with oxygen concentration, which is represented as:

[0014] x k+1 =f(x k ,u k )

[0015] Wherein u k represents the oxygen concentration at k sampling time, x k represents the blood oxygen saturation at k sampling time;

[0016] Step 2: obtain an augmented state equation of the nonlinear form dynamic characteristic equation, which is represented as:

[0017] X k =[x k ,u k ] ·

[0018] The augmented state equation is extended to obtain an extended equation, which is represented as:

[0019] X k+1 =[f(x k ,u k ),Su k ]·

[0020] Among them Su k =u k+1 , S represents the left shift operator;

[0021] Step 3: Obtain the Koopman operator by extending the dynamic modal decomposition method, and convert the nonlinear dynamic characteristic equation in step 1 into a linear dynamic characteristic equation by using the Koopman operator.

[0022] Preferably, step 3 specifically includes the following steps:

[0023] Step 1: Get the state augmentation matrix of blood oxygen saturation X and State augmented matrix of blood oxygen saturation X and Respectively expressed as:

[0024] X =[X1,X2,…,X M-1 ] ·

[0025]

[0026] Where M represents the total number of samples for constructing the population prediction model;

[0027] Step 2: Augment the blood oxygen saturation state matrix X and Perform state improvement respectively and obtain the state matrix Φ( X )and The state matrix Φ( X )and Respectively expressed as:

[0028] Φ( X )=[φ(X1),φ(X2),…,φ(X M-1 )] ·

[0029]

[0030] Step 3: Get the group Koopman operator Group Koopman operator Expressed as:

[0031]

[0032] Step 4: Obtain the linear dynamic characteristic equation, which is expressed as:

[0033]

[0034] Preferably, in step S2, the model updating and prediction are performed by online iteration, outputting the blood oxygen saturation prediction value, individualizing the model based on the validation set, and using the population prediction model, specifically including the following steps:

[0035] S21: performing daily model updating on the dth day online;

[0036] S22: estimating the Koopman operator value of the (d-1)th day according to the augmented state matrix obtained by iteration The iteration formula is represented as:

[0037]

[0038] wherein and X d-1 all represent the augmented state matrix obtained by the (d-1)th day data, γ d-1 represents the correction vector of the (d-1)th day, γ d-1 is represented as:

[0039] γ d-1 = Φ( X d-1 )Γ d

[0040] wherein Γ d represents the covariance matrix of the dth day, Γ d is represented as:

[0041] Γ d = Γ d-1 - Γ d-1 Φ( X d-1 )(Φ( X d-1 )Γ d-1 Φ( X d-1 )+I) -1 Φ( X d-1 ) · Γ d-1 .

[0042] Preferably, in step S22, when d is 1, X 0 , γ 0 and Γ 0 are the corresponding matrices obtained based on the population data, Γ 0 is represented as:

[0043] Γ 0 = (Φ( X)Φ( X ) · ) -1 .

[0044] Preferably, in step S3, the lower-layer optimization cost function is constructed by the low-oxygen concentration double-layer optimization module with the low-oxygen concentration as a variable, and the lower-layer optimization cost function is expressed as:

[0045]

[0046] X k =[x k ,u d ] T

[0047]

[0048] wherein S y (u d ) represents the area under the curve of the predicted blood oxygen saturation under the action of the low-oxygen concentration u d in a low-oxygen section period, y m represents the minimum value of the predicted blood oxygen saturation corresponding to the low-oxygen section, ρ and μ are hyperparameters, y k represents the blood oxygen saturation value corresponding to the predicted low-oxygen section, represents the upper threshold of the low-oxygen concentration, and u(ΔSI d-1 ) represents the lower threshold of the low-oxygen concentration.

[0049] Preferably, in step S3, the upper-layer optimization design is adaptively designed according to the historical learning performance and safety constraints, and the upper-layer optimization design is specifically:

[0050] Design I: if ΔSI d-1 < 0, ||ΔSI d-1 || > S TH , and ||u d-1 - u d-1 || < U TH , then is expressed as:

[0051]

[0052] wherein α0, S TH and U TH are hyperparameters, u d-1 represents the lower threshold of the oxygen concentration safety range on the d-1th day, u d represents the lower threshold of the oxygen concentration safety range on the dth day.

[0053] Design II: if ΔSI d-1 > 0, ||ΔSI d-1 || > S TH , and ||u d-1 - is expressed as:

[0051]

[0052] wherein α0, S TH and U TH are hyperparameters, u d-1 represents the lower threshold of the oxygen concentration safety range on the d-1th day, u d represents the lower threshold of the oxygen concentration safety range on the dth day.

[0053] Design III: if ΔSI d-1 > 0, ||ΔSI d-1 || > S TH , and ||u d-1 - is expressed as:

[0051]

[0052] wherein α0, S TH and U TH are hyperparameters, u d-1 represents the lower threshold of the oxygen concentration safety range on the d-1th day, u d represents the lower threshold of the oxygen concentration safety range on the dth day.d-1 > 0, || ASI d-1 || > S TH , and then is expressed as:

[0054]

[0055] wherein α1 is a hyperparameter, represents the lower threshold of the oxygen concentration safety range on the d-1th day, represents the lower threshold of the oxygen concentration safety range on the dth day.

[0056] Preferably, the hyperparameter is determined by the validation set data, and the determination method adopts a grid search method.

[0057] Preferably, step S4 specifically comprises the following steps:

[0058] S41: calculating a training evaluation index SI, which is expressed as:

[0059]

[0060] wherein represents the average area of the blood oxygen saturation in all hypoxic segments of the IHT per day, represents the average hypoxic concentration;

[0061] S42: comparing the training evaluation index of the first day with the training evaluation index of the last day to evaluate the overall training result;

[0062] S43: evaluating the training effect by the change ASI d-1 of the training evaluation index of the previous day, and feeding back the change ASI d-1 of the training evaluation index of the previous day to the low oxygen concentration double-layer optimization module.

[0063] Preferably, in step S43, the change ASI d-1 of the training evaluation index of the previous day is expressed as:

[0064] ASI d-1 = SI d-1 - SI d-2 .

[0065] Therefore, the present application adopts the above-mentioned individualized highland hypoxia pre-training closed-loop intervention algorithm, and has the following beneficial effects:

[0066] (1) The individualized prediction model learning module of the present application utilizes multiple hypoxic periods in the IHT per day to perform Kupman operator modeling, which can fully mine data information and find out key features for decision optimization;

[0067] (2) The constraint part of the low oxygen concentration double-layer optimization module of the present application considers the physiological limitations of the human body and the adaptive differences in training intensity of different individuals, controls the low oxygen concentration within a reasonable bearing range, accelerates the high altitude pre-service training, and avoids irreversible damage to individuals due to hypoxia;

[0068] (3) The present application is a one-time change from open loop structure to closed loop structure of IHT, which solves the problem of no guidance and direction in current high altitude pre-service training through IHT. Considering the specificity between individuals, different subjects will have different reactions when IHT uses the same low oxygen stimulation intensity, and some subjects may have the situation that the training effect is worse. After closed loop intervention, the algorithm allows real-time feedback according to the different training effects of each individual, intelligently optimizes the next oxygen supply strategy, effectively realizes personalized optimal training, and improves the training effect.

[0069] The method scheme of the present application will be further described in detail below by means of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0070] Figure 1 The flowchart of the individualized high altitude hypoxia pre-service closed loop intervention algorithm of the present application;

[0071] Figure 2 The correlation results of the test set based on the embodiment of the present application;

[0072] Figure 3 The trend correctness evaluation results based on the test set of the embodiment of the present application. DETAILED DESCRIPTION

[0073] The method scheme of the present application will be further described in detail below by means of the accompanying drawings and examples.

[0074] Unless otherwise defined, the method terms or scientific terms used in the present application shall have the usual meaning understood by those skilled in the art to which the present application belongs.

[0075] The terms "comprising" or "including" or similar phrases as used herein is intended to encompass the elements listed thereafter, without precluding the presence or addition of one or more other elements. The terms "inner", "outer", "upper", "lower", and the like, indicate relative positions or orientation of an apparatus or element shown in the drawings, and are used only to facilitate the description of the application and the understanding of the drawings, and do not indicate or imply that the apparatus or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore should not be interpreted as limiting the application. In the present application, unless otherwise specified and limited, the term "attached" and the like should be interpreted broadly, for example, it can be fixedly connected, or detachably connected, or integrated; it can be directly connected, or indirectly connected through an intermediate medium, or it can be an internal connection of two elements or an interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0076] Embodiment

[0077] In this embodiment, IHT data of 13 subjects are selected, of which 6 are used as a training set to determine a population model, 3 are used to individualize the model based on the population model, and are used as a validation set to determine the hyperparameters of the optimization problem, and the other 4 subjects are used as a test set to verify the effectiveness of the method.

[0078] As shown in Figure 1 The present application provides an individualized high-altitude hypoxia pre-exposure closed-loop intervention algorithm, comprising the following steps:

[0079] S1: During the high-altitude pre-exposure, the training subject performs one hour of IHT per day for a total of 7 days, and periodic oxygen supply is performed during the IHT, with a square wave concentration, 5 minutes of hypoxia and 3 minutes of hyperoxia. The hypoxic concentration u is an optimized variable. The training subject wears a blood oxygen meter during the IHT, and the oxygen supply concentration and blood oxygen saturation during the IHT are detected by a population and individual prediction model learning module, online data are acquired and recorded in real time, and the data are divided into a training set, a validation set and a test set;

[0080] S2: Based on the training set, the variation of blood oxygen saturation under intermittent hypoxia is learned by a Koopman operator, and a population prediction model of blood oxygen saturation is constructed using a data-driven method of the Koopman operator; the model is updated and predicted by online iteration, and the blood oxygen saturation prediction value is output. Based on the validation set, the population prediction model is individualized;

[0081] In step S2, based on the training set, the variation law of blood oxygen saturation under intermittent hypoxia is learned by the Koopman operator, a population prediction model of blood oxygen saturation is constructed by using the data-driven method of the Koopman operator, and the specific steps include the following steps:

[0082] Step 1: Obtain the nonlinear form dynamic characteristic equation of blood oxygen saturation with oxygen concentration, and the nonlinear form dynamic characteristic equation is represented as:

[0083] x k+1 =f(x k ,u k )

[0084] Wherein u k represents the oxygen concentration at k sampling time, x k represents the blood oxygen saturation at k sampling time;

[0085] Step 2: Obtain the augmented state equation of the nonlinear form dynamic characteristic equation, and the augmented state equation is represented as:

[0086] X k =[x k ,u k ] ·

[0087] The augmented state equation is extended to obtain the extended equation, and the extended equation is represented as:

[0088] X k+1 =[f(x k ,u k ),Su k ] ·

[0089] Wherein Su k =u k+1 , S represents a left shift operator;

[0090] Step 3: Obtain the Koopman operator by the extended dynamic modal decomposition method, and convert the nonlinear form dynamic characteristic equation in step 1 into a linear form dynamic characteristic equation by the Koopman operator;

[0091] Step 3 includes the following steps:

[0092] Step one: obtain the state augmented matrix of blood oxygen saturation X And The state augmented matrix of blood oxygen saturation X And Is represented as:

[0093] X =[X1,X2,…,X M-1 ] ·

[0094]

[0095] where M represents the total number of samples for constructing the population prediction model;

[0096] Step two: augmented matrix of the state of blood oxygen saturation X and respectively, to obtain the state matrix Φ( X ) and obtain the state matrix Φ( X ) and Φ( X ) and respectively, as follows:

[0097] Φ( X ) = [φ(X1), φ(X2), …, φ(X M-1 )] ·

[0098]

[0099] Step three: obtain the population library kapman operator Population library kapman operator can depict the average dynamic characteristics, and the population library kapman operator is expressed as:

[0100]

[0101] Step four: obtain the linear form dynamic characteristic equation, and the linear form dynamic characteristic equation is expressed as:

[0102]

[0103] In step S2, considering the difference in the change of individual blood oxygen saturation, the model is individualized through different individual data on the basis of the population model, and the daily model is updated online for the dth day. According to the kapman operator estimation value of the (d-1)th day the kapman operator estimation value of the dth day is obtained by iteration The iteration formula is expressed as:

[0104]

[0105] where and X d-1 all represent the augmented state matrix obtained from the d-1th day data, γ d-1 represents the correction vector of the d-1th day, γ d-1 is expressed as:

[0106] γd-1 = Φ X d-1 )Γ d

[0107] where Γ d denotes the covariance matrix on day d, Γ d is given by:

[0108] Γ d = Γ d-1 - Γ d-1 Φ X d-1 (Φ X d-1 )Γ d-1 Φ X d-1 + I) -1 Φ X d-1 ) · Γ d-1 ;

[0109] When d is 1, X 0 , γ 0 and Γ 0 are the corresponding matrices obtained based on the population data, Γ 0 is given by:

[0110] Γ 0 = (Φ X )Φ X ) · ) -1 .

[0111] S3: Based on the blood oxygen saturation prediction value, a safety constraint adaptive double-layer optimization framework is constructed for the low oxygen concentration variable. Through the established individualized model, we can predict the blood oxygen saturation level under different low oxygen concentration conditions, and based on the results, design the daily low oxygen concentration optimization problem. Through the low oxygen concentration double-layer optimization module, the lower layer optimization cost function is constructed for the low oxygen concentration variable, and the upper layer optimization is designed adaptively according to the safety constraints and historical learning performance;

[0112] In step S3, the lower layer optimization cost function is constructed for the low oxygen concentration variable through the low oxygen concentration double-layer optimization module, and the lower layer optimization cost function is represented as:

[0113]

[0114] X k = [x k , u d ] T

[0115]

[0116] where S y (u d ) represents the area under the curve of predicted blood oxygen saturation under the action of hypoxic concentration u d in a hypoxic period, y m represents the minimum value of the predicted blood oxygen saturation corresponding to the hypoxic period, and p and m are hyperparameters, y k represents the predicted blood oxygen saturation value corresponding to the hypoxic period, represents the upper threshold of the hypoxic concentration, and u (AS d-1 ) represents the lower threshold of the hypoxic concentration, both of which are related to the change of the training evaluation index of the previous day and are determined by the upper layer optimization design.

[0117] The ratio is set in the form of a ratio to ensure that the body's blood oxygen saturation level is as high as possible while ensuring sufficient hypoxic stimulation to trigger adaptive changes.

[0118] In step S3, a safety-constrained adaptive upper layer optimization design is performed according to the historical learning performance, and the upper layer optimization design is specifically:

[0119] Design I: If AS d-1 < 0, || AS d-1 || > S TH , and || u d-1 - u d-1 || < U TH , then is represented as:

[0120]

[0121] where a0, S TH , and U TH are hyperparameters, u d-1 represents the lower threshold of the safe range of oxygen concentration on the d-1 day, u d represents the lower threshold of the safe range of oxygen concentration on the d day.

[0122] Design II: If AS d-1 > 0, || AS d-1 || > S TH , and then is represented as:

[0123]

[0124] wherein a1 is a hyperparameter, denotes the lower threshold of the safe range of oxygen concentration on the d-1th day, denotes the lower threshold of the safe range of oxygen concentration on the dth day.

[0125] The hyperparameter is determined by the validation set data, and the determination method adopts the grid search method.

[0126] The basic optimization principle followed by the upper layer optimization design of the embodiment conforms to the hypoxic concentration regulation criterion in the clinic, that is, when the adaptability declines, the hypoxic concentration should be increased to prevent further damage to the body, so the upper bound of the constraint is increased, and vice versa, when the adaptability improves, it means that the body can adapt well under the current hypoxic concentration, so we will lower the lower bound of the constraint to explore the feasibility of lower hypoxic concentration and improve the rapidity of adaptation.

[0127] As shown in Figure 2 and Figure 3 , the decision direction of the method of the embodiment conforms to the basic law of strategy adjustment, that is, when the performance decreases, the optimization result tends to increase the hypoxic concentration, which verifies the feasibility of the embodiment.

[0128] S4: Evaluate the training result by the training effect evaluation module to evaluate whether the training effect reaches the required level, and feed back the evaluation result to the low oxygen concentration double-layer optimization module;

[0129] Step S4 specifically includes the following steps:

[0130] S41: After each training, the training evaluation index SI can be calculated by using the data recorded by IHT, which is used to measure the ability of the individual to maintain a high blood oxygen saturation at the lowest possible oxygen concentration. The training evaluation index SI is represented as:

[0131]

[0132] wherein denotes the average area of blood oxygen saturation of all hypoxic segments of IHT in a day, denotes the average hypoxic concentration;

[0133] S42: Compare the training evaluation index of the first day with the training evaluation index of the last day to evaluate the overall training result;

[0134] S43: Evaluate the training effect by the change of the training evaluation index of the last day ΔSI d-1 and feed back the change of the training evaluation index of the last day ΔSI d-1 to the low oxygen concentration double-layer optimization module for guiding the optimization of the safety constraint;

[0135] In step S43, the change of the training evaluation index of the last day ΔSId-1 is represented as:

[0136] ΔSI d-1 = SI d-1 -SI d-2 .

[0137] Therefore, the application adopts the above-mentioned individualized high-altitude hypoxia pre-training closed-loop intervention algorithm, can calculate dynamic evaluation indexes by using IHT data, and can regulate the hypoxic stimulation intensity of IHT by controlling the optimization strategy to improve the rapidity and effectiveness of high-altitude adaptation, speed up the training cycle of individuals in the plain area, improve the hypoxic tolerance and reduce the risk of acute high-altitude disease.

[0138] Finally, it should be noted that: the above examples are only used to illustrate the method scheme of the present application, but not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: the method scheme of the present application can still be modified or replaced by the equivalent, and these modifications or equivalent replacements cannot make the modified method scheme deviate from the spirit and scope of the method scheme of the present application.

Claims

1. An individualized closed-loop intervention algorithm for high-altitude hypoxia preparatory service, characterized by: The following steps are involved: S1: During IHT, subjects were given periodic oxygen supply, and physiological signals were collected in real time to obtain continuous data on oxygen concentration and blood oxygen saturation during IHT. The continuous data on blood oxygen saturation were divided into training, validation, and test sets according to different subjects. S2: Based on the training set, the Koopman operator is used to learn the changing pattern of blood oxygen saturation under intermittent hypoxia, and a group prediction model for blood oxygen saturation is constructed. The model is updated and predicted through online iteration, and the predicted blood oxygen saturation value is output. Based on the validation set, the group prediction model is used for model individualization; In step S2, based on the training set, the Koopman operator is used to learn the changing pattern of blood oxygen saturation under intermittent hypoxia, and a group prediction model of blood oxygen saturation is constructed, which specifically includes the following steps: Step 1: Obtain a nonlinear dynamic characteristic equation of how blood oxygen saturation changes with oxygen concentration. The nonlinear dynamic characteristic equation is expressed as: ； in express The oxygen concentration at the sampling time, express Blood oxygen saturation at the sampling moment; Step 2: Obtain the augmented state equation of the nonlinear dynamic characteristic equation. The augmented state equation is expressed as: ； Expanding the augmented state equation, we get the extended equation, which is expressed as: ； in , S represents the left shift operator; Step 3: Obtain the Koopman operator by extending the dynamic modal decomposition method, and convert the nonlinear dynamic characteristic equation in step 1 into a linear dynamic characteristic equation by using the Koopman operator; S3: Uses an individualized model to predict blood oxygen saturation. A dual-layer optimization module for supply and hypoxia concentration is used to construct a lower-layer optimization cost function with hypoxia concentration as a variable. Based on historical learning performance, an upper-layer optimization design is performed with safety constraints and self-adaptation. Step 3 specifically includes the following steps: Step 1: Get the state augmentation matrix of blood oxygen saturation and , the state augmented matrix of blood oxygen saturation and Respectively expressed as: ； ； in M Indicates the total number of samples for constructing the group prediction model; Step 2: Augment the blood oxygen saturation state matrix and Perform state improvement separately to obtain the state matrix and , the state matrix and Respectively expressed as: ； ； Step 3: Get the group Koopman operator , group Koopman operator Expressed as: ； Step 4: Obtain the linear dynamic characteristic equation, which is expressed as: ； S4: The training results are evaluated through the training effect evaluation module, and the evaluation results are fed back to the low oxygen concentration double-layer optimization module.

2. The individualized closed-loop intervention algorithm for high-altitude hypoxia preparatory service according to claim 1 is characterized in that: In step S2, the model is updated and predicted through online iteration, and the blood oxygen saturation prediction value is output. Based on the validation set, the model is individualized using the group prediction model, which specifically includes the following steps: S21: Online Daily model updates are conducted on the day; S22: According to Koopman operator estimate for the day Iterate to get Koopman operator estimate for the day , the iterative formula is expressed as: ； in and All indicate that The augmented state matrix obtained from the data of the day, Indicates the The correction vector of the day, Expressed as: ； in Indicates the The covariance matrix of the day, Expressed as: 。 3. The individualized closed-loop intervention algorithm for high altitude hypoxia preparatory service according to claim 2 is characterized in that: In step S22, when When 1, 、 、 and is the corresponding matrix obtained based on population data, Expressed as: 。 4. The individualized closed-loop intervention algorithm for high-altitude hypoxia preparatory service according to claim 1 is characterized in that: In step S3, a lower-layer optimization cost function with the supply and low oxygen concentration as a variable is constructed through the low oxygen concentration double-layer optimization module. The lower-layer optimization cost function is expressed as: ； in Indicates that during a low oxygen period, the The area under the curve of the predicted blood oxygen saturation under the action of Indicates the minimum value of blood oxygen saturation corresponding to the predicted hypoxic segment, and is a hyperparameter, Indicates the blood oxygen saturation value corresponding to the predicted hypoxic segment, Indicates the upper threshold of low oxygen concentration. Indicates the lower threshold of low oxygen concentration.

5. The individualized closed-loop intervention algorithm for high-altitude hypoxia preparatory service according to claim 4 is characterized in that: In step S3, the upper-level optimization design of security constraint adaptation is performed based on the historical learning performance. The upper-level optimization design is specifically as follows: Design 1: If , ,and ,but Expressed as: ； in , and are all hyperparameters, Indicates the The lower threshold of the safe range of oxygen supply concentration per day, Indicates the The lower threshold of the safe range of oxygen supply concentration per day; Design 2: If , ,and ,but Expressed as: ； in is a hyperparameter, Indicates the The lower threshold of the safe range of oxygen supply concentration per day, Indicates the The lower threshold of the safe range of oxygen supply concentration for the day.

6. The individualized closed-loop intervention algorithm for high-altitude hypoxia preparatory service according to claim 5 is characterized in that: The hyperparameters are determined by the validation set data using the grid search method.

7. The individualized closed-loop intervention algorithm for high altitude hypoxia preparatory service according to claim 1 is characterized in that: Step S4 specifically includes the following steps: S41: Calculate training evaluation indicators , training evaluation indicators Expressed as: ； in It represents the average area of ​​blood oxygen saturation in all hypoxic segments of IHT in one day. Indicates the average low oxygen concentration; S42: Compare the training evaluation indicators of the first day and the training evaluation indicators of the last day to evaluate the overall training results; S43: Changes in evaluation indicators based on the previous day’s training Evaluate the training effect and report the changes in the previous day's training evaluation indicators Feedback is provided to the low oxygen concentration double-layer optimization module.

8. The individualized closed-loop intervention algorithm for high-altitude hypoxia preparatory service according to claim 7 is characterized in that: In step S43, the change of the training evaluation index of the previous day Expressed as: 。

Images