CN116028845A - Training method of alveolar gas recognition model and alveolar gas acquisition device - Google Patents
Training method of alveolar gas recognition model and alveolar gas acquisition device Download PDFInfo
- Publication number
- CN116028845A CN116028845A CN202211618120.9A CN202211618120A CN116028845A CN 116028845 A CN116028845 A CN 116028845A CN 202211618120 A CN202211618120 A CN 202211618120A CN 116028845 A CN116028845 A CN 116028845A
- Authority
- CN
- China
- Prior art keywords
- alveolar gas
- alveolar
- gas
- matrix
- sparse
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention provides a training method of an alveolar gas recognition model and an alveolar gas acquisition device, which are characterized in that a redundant matrix is pre-constructed, carbon dioxide sensor and oxygen sensor data are randomly sub-sampled to obtain sparse signals, an observation matrix is constructed to carry out projection transformation on the sparse signals, data signals are reconstructed through orthogonal matching tracking, trend information of sensor data signal changes is obtained, and nodes where alveolar gas appears are judged; in the invention, the components of the exhaled air are analyzed by the programmable logic controller with the built-in alveolar gas recognition model, so that the nodes where the alveolar gas appears are judged, and the use is convenient; the alveolar gas recognition model recovers the original signal from the observed value, adopts a greedy tracking algorithm, namely utilizes the correlation between atoms in the observed matrix and signal residual errors, solves the optimal sparse value of the original data signal by adding iteration conditions and using a least square method to judge the node where the alveolar gas appears, and improves the accuracy of alveolar gas collection.
Description
Technical Field
The invention relates to the technical field of alveolar gas collection, in particular to a training method of an alveolar gas identification model and an alveolar gas collection device.
Background
Alveolar gas is defined as the gas in the human breath after air exchange is completed, as distinguished from the gas in the human breath that does not participate in air exchange. In the first half of the human emetic process, air is usually inhaled and present in the respiratory tract, while alveolar gas tends to occur in the second half of the deep respiratory emetic process. According to the data, the exhaled breath of a human body contains various volatile organic compounds, and the metabonomics related research has proved that the occurrence of various disease lesions can cause the composition of the exhaled breath of the human body to change, for example, diabetes is often accompanied by the increase of the concentration of acetone in the exhaled breath of the human body. Therefore, the analysis of the exhaled breath of the human body becomes a new hot spot in the fields of medical monitoring and disease diagnosis and early warning in the last twenty years.
The collecting of exhaled breath before diagnosing diseases by using the volatile component analysis method is required, and the collecting flow Cheng Daduo commonly used in domestic and foreign researches is aimed at collecting exhaled breath in human breath, in fact, the analysis of exhaled breath of human is best to analyze alveolar gas, because the alveolar gas is exhaled gas after air exchange is completed at alveoli in exhaled breath of human, and the occurrence of alveolar gas needs to be monitored. The exhaled breath of the subject is typically collected by the subject's exhalation to an airbag, then the concentration of the volatile organic compounds in the exhaled breath is increased by an enrichment method, and finally the exhaled breath of the subject is analyzed in combination with gas chromatography-mass spectrometry or electronic nose. Although generally applicable to the field of human exhaled breath analysis, the method has some defects and shortcomings, such as difficulty in collecting gas for people with pulmonary insufficiency, easiness in interference of target volatile organic compounds in exhaled breath by background air, extremely low concentration and excessively complex process for treating exhaled gas.
Chinese patent application publication No. CN112957077a discloses a mask type breath collection device and method thereof, comprising: a mask provided with a breathing hole and an air inlet hole; the first three-way valve is connected with the input end of the first three-way valve, and the exhaled gas is separated through the first three-way valve; the output end of the first three-way valve is connected with the input end of the second three-way valve, and alveolar gas is conveyed through the second three-way valve; an output end of the second three-way valve is connected with an input end of the first gas collecting mechanism; a second air pump; and the other output end of the third three-way valve is connected with the input end of the third three-way valve, and one output end of the third three-way valve is connected with the extraction opening of the second air pump. The node that this collection system when using can't accurate judgement alveolus gas appear, and the alveolus gas collection of single exhale is incomplete, influences the accuracy of detection data.
Disclosure of Invention
Aiming at the technical defects, the invention aims to provide a training method of an alveolar gas recognition model and an alveolar gas acquisition device, which omit complicated processes of adsorption and desorption, save cost, avoid changing components in exhaled air in the adsorption-desorption process and improve the accuracy of detection data.
In order to achieve the above purpose, the invention adopts the following technical scheme:
a training method of an alveolar gas recognition model, comprising the steps of:
s1, presetting an alveolar gas identification model based on a compressed sensing model;
s2, acquiring a complete respiration process of a first subject as a verification sample through an alveolar gas acquisition device;
s3, intercepting part of the respiratory process of the first subject as a first training sample, inputting the first training sample into an alveolar gas identification model, predicting the obtained data as a first prediction mode, and identifying nodes where alveolar gas appears in the first prediction mode;
s4, comparing the verification sample with the first prediction mode to evaluate an alveolar gas recognition model;
s5, optimizing and correcting an alveolar gas identification model by taking accurate prediction of alveolar gas occurrence nodes as a target;
s6, training the alveolar gas recognition model according to the steps S2 to S5 by using a plurality of subjects.
Preferably, the method for processing the respiration process of the subject by the alveolar gas identification model comprises the following steps: a redundant matrix with the number of lines smaller than the number of columns is pre-constructed, the carbon dioxide sensor and oxygen sensor data of the alveolar gas collection device are randomly sub-sampled to obtain sparse signals, then an observation matrix is constructed to carry out projection transformation on the sparse signals, reconstruction of data signals is achieved through orthogonal matching tracking, trend information of data signal changes of the carbon dioxide sensor and the oxygen sensor is obtained, and nodes where alveolar gas appears are judged.
Preferably, the method of alveolar gas recognition model for treating a respiratory process of a subject comprises: setting a one-dimensional signal X= (X) according to a compressed sensing theoretical model 1 ,X 2 ,X 3 ,…,X N )，X∈R N X is a discrete data signal of dimension N X1, sparse matrix ψ= (ψ) 1 ,Ψ 2 ,Ψ 3 ,…,Ψ N ),dΨ∈R N*N Obtaining a sparse signal through sparse transformation, wherein the sparsity of X is K (K<<N), according to the formula: x=ψx performing sparse signal projective transformation;
wherein, psi is a sparse matrix, X is an original signal; x epsilon R N*1 Is a sparse matrix transformed signal;
the observation matrix is assumed to be Φ= (Φ) 1 ,Φ 2 ,Φ 3 ,…,Φ N )，Φ∈R M*N Is a unit vector, and M<<And N, carrying out projective transformation on the sparse signals through an observation matrix, wherein the formula is as follows: y=Φ×x to obtain an observed value y∈r M ；
Wherein phi is an observation matrix, and y is an observation value;
by the above formula, the y=Φx=Φ×ψ=η×x, where η≡Φ×ψ is a sensing matrix;
the method comprises the following steps of selecting an orthogonal matching pursuit algorithm to reconstruct a target gas concentration signal:
input: m×n-dimensional perceptual matrix η=Φψ, m×1 measured value y, signal sparsity K;
and (3) outputting: sparse representation coefficient of signal is Yib;
wherein use is made of
Representing the empty set, Λ t Set of indices representing t iterations, J t Index number representing t iterations;
initializing: residual r 0 =y, index set
The number of iterations t=1;
the method comprises the following steps:
step 1), finding out residual error r t-1 And a perception matrix eta j The maximum value in the inner product is found by index J t So that J t ＝argmax j ＝1，2，…，n|<r t-1 ，η j >I, wherein eta j Is the j-th column of the sensing matrix;
step 2), updating the index set Λt=Λ t-1 ∪{J t Recording the reconstructed atom set D in the found perceptual matrix t ＝[D t-1 ，η Jt ]；
Step 3), obtaining an approximate solution of the signal by a least square method: b t ＝argmin‖y-D t b t ‖ 2 ；
Step 4), updating residual r t ＝y-Dt＾bt；
Step 5), let t=t+1, if t < K or ||r t ‖ 2 And (2) if the number is not less than epsilon, executing the step (1), and sequentially iterating;
and circularly executing the steps 1) to 5), and solving the optimal sparse value of the original signal.
Preferably, the method of optimizing and correcting the alveolar gas recognition model includes: acquiring the time and concentration (Ti, ci, ti, ci) and corresponding trigger nodes (Si, si) of the carbon dioxide sensor and the oxygen sensor of the alveolar gas collection device at the occurrence of alveolar gas in the respiratory cycle of a plurality of subjects; training an alveolar gas recognition model by taking the (Ti, ci, ti, ci) as an independent variable and a trigger node (Si, si) corresponding to the exhalation state of a plurality of subjects as the dependent variable; acquiring concentration signals (Ti ', ci ', ti ', ci ') in unknown fragment exhalation states, and processing the signals (Ti ', ci ', ti ', ci ') by using a trained prediction model to obtain current trigger nodes (Si ' ); when the trigger node is reached, the alveolar gas collection device starts active gas collection.
An alveolar gas collection device comprises a half mask, a nose wing clamp, a carbon dioxide sensor, an oxygen concentration sensor, a differential pressure meter, a micropump, a conductive gas path, a superfine glass fiber filter membrane and a programmable logic controller with an alveolar gas identification model arranged in the middle; the nose wing clip is connected with the top of the half mask; the conducting gas circuit is fixed outside the half mask and communicated with the half mask; the programmable logic controller is fixed on the outer side surface of the half mask, and the carbon dioxide sensor and the oxygen concentration sensor are arranged on the inner side surface of the half mask; the superfine glass fiber filter membrane, the differential pressure gauge and the micropump are sequentially arranged in the conducting gas path from top to bottom; the conduction gas circuit is connected with a gas collecting device; the flow guiding direction of the micropump always flows from the inside of the half mask to the outside through a guiding air passage; the carbon dioxide sensor, the oxygen concentration sensor, the differential pressure meter, the micropump and the gas collecting device are respectively and electrically connected with the programmable logic controller.
Preferably, the carbon dioxide sensor, the oxygen concentration sensor, the differential pressure gauge and the micropump are uniformly coated with polytetrafluoroethylene microporous films on the surfaces.
Preferably, the carbon dioxide sensor and the oxygen concentration sensor are detachably arranged in the center of the inner side surface of the half mask.
Compared with the prior art, the invention has the following beneficial effects:
1. according to the invention, the carbon dioxide sensor, the oxygen concentration sensor and the differential pressure meter which are arranged on the half mask are used for monitoring the gas exhaled by the patient, and the programmable logic controller with the alveolar gas recognition model is used for analyzing the components of the exhaled gas, so that the nodes where the alveolar gas appears are accurately judged, and the alveolar gas in the respiratory cycle of the human body is collected through the gas collecting device, and the device is convenient to use and easy to operate.
2. In the invention, an alveolar gas recognition model recovers an original signal from an observed value, a greedy tracking algorithm is adopted, namely, the correlation between atoms in an observation matrix and signal residual errors is utilized, iteration conditions are added, and a least square method is used for solving the optimal sparse value of an original data signal to judge the node where alveolar gas appears, so that the accuracy of alveolar gas collection is improved.
Drawings
FIG. 1 is a schematic diagram of the structure of the present invention;
FIG. 2 is a schematic diagram of the connection of a programmable logic controller to a half mask in accordance with the present invention;
FIG. 3 is a schematic diagram showing the connection of the ultra-fine glass fiber filter membrane and the conducting gas path in the invention;
FIG. 4 is a graph showing the concentration signal of the carbon dioxide sensor output by the alveolar gas recognition model in the present invention;
FIG. 5 is a graph showing the concentration signal of the oxygen sensor output by the alveolar gas recognition model according to the present invention.
Wherein:
1. nose wing clips; 2. a half mask; 3. a carbon dioxide sensor; 4. an oxygen concentration sensor; 5. a differential pressure gauge; 6. a micropump; 7. a conductive gas path; 8. a superfine glass fiber filter membrane; 9. a polytetrafluoroethylene microporous membrane; 10. a programmable logic controller.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
As shown in fig. 1 to 3, an alveolar gas collection device comprises a half mask 2, a nose clip 1, a carbon dioxide sensor 3, an oxygen concentration sensor 4, a differential pressure meter 5, a micropump 6, a conductive gas circuit 7, an ultrafine glass fiber filter membrane 8 and a programmable logic controller 10 with an alveolar gas identification model arranged inside; the nose wing clip 1 is connected with the top of the half mask 2; the conducting gas circuit 7 is fixed outside the half mask 2 and communicated with the half mask 2; the conductive gas path 7 is easily connected to a gas collection device, such as a tenax sorbent tube, a ventilator, or a gas collection bag, for alveolar gas collection by communication interaction with the programmable logic controller 10; the programmable logic controller 10 is fixed on the outer side surface of the half mask 2, and the carbon dioxide sensor 3 and the oxygen concentration sensor 4 are arranged on the inner side surface of the half mask 2; the carbon dioxide sensor 3, oxygen concentration sensor 4 record the carbon dioxide and oxygen concentration of human breathing process when the human body wears, in the time series data or predictive data of the carbon dioxide sensor 3, oxygen concentration sensor 4, when the carbon dioxide concentration produces the increase of 50% or more of the amplitude compared with the value before one second or the oxygen concentration produces the trend of reducing of 50% or more of the amplitude compared with the value before one second, judge the alveolar gas to appear; the superfine glass fiber filter membrane 8, the differential pressure meter 5 and the micropump 6 are sequentially arranged in the conducting gas path 7 from top to bottom; the flow guiding direction of the micropump 6 always flows from the inside of the half mask 2 to the outside through the conducting air path 7; the carbon dioxide sensor 3, the oxygen concentration sensor 4, the differential pressure meter 5, the micropump 6 and the gas collecting device are respectively and electrically connected with the programmable logic controller 10, and the programmable logic controller 10 is internally provided with a program capable of carrying out fusion processing on the data forms collected by the carbon dioxide sensor 3, the oxygen concentration sensor 4 and the differential pressure meter 5, and carrying out abnormal value checking and correction processing on the data information so as to control the gas collecting device to collect alveolar gas; alveolar gas collection procedure: after the device is worn, the exhaled air is exhaled from the mouth and enters the inner space of the half mask 2, the exhaled air is contacted with the carbon dioxide sensor 3 and the oxygen concentration sensor 4 to obtain the real-time concentration of carbon dioxide and oxygen of the exhaled air, and the air enters the conducting air path 7 from the inner part of the half mask 2 and sequentially passes through the superfine glass fiber filter membrane 8, the differential pressure gauge 5 and the micropump 6; when the above collection flow is repeated for a plurality of times and the active gas collection of the alveolar gas collection device still cannot be started, the nose wing clip 1 can be used for clamping the nose wings of a human body, and the breathing process can be rapidly developed so as to reduce the interference of the nasal inhalation gas on the sampling.
Further, the surfaces of the carbon dioxide sensor 3, the oxygen concentration sensor 4, the differential pressure gauge 5 and the micropump 6 are uniformly coated with a polytetrafluoroethylene microporous membrane 9.
Further, the carbon dioxide sensor 3 and the oxygen concentration sensor 4 are detachably arranged in the center of the inner side surface of the half mask 2; the carbon dioxide sensor 3 and the oxygen concentration sensor 4 optimize the arrangement scheme based on aerodynamic analysis: taking the oral cavity of a human body as a gas source, inputting the flow of the exhaled gas of a subject and the initial positions of the carbon dioxide sensor 3 and the oxygen concentration sensor 4, and taking the time from wearing the half mask 2 to detecting alveolar gas as an objective function; the data correlation of the two sensors of the same type is included into an index; taking the minimum function value as a planning target and carrying out sensitivity analysis; an optimized sensor layout scheme is generated, and the positions of the carbon dioxide sensor 3 and the oxygen concentration sensor 4 are set accordingly.
Further, the half mask 2 is tightly attached to the lower half face of the subject through a sealing strip; the sealing strip is made of rubber material, and the half mask 2 frame and the main body are integrally injection molded by adopting resin materials.
A training method of an alveolar gas recognition model, comprising the steps of:
s1, presetting an alveolar gas identification model based on a compressed sensing model;
s2, acquiring a complete respiration process of a first subject as a verification sample through an alveolar gas acquisition device;
s3, intercepting part of the respiratory process of the first subject as a first training sample, inputting the first training sample into an alveolar gas identification model, predicting the obtained data as a first prediction mode, and identifying nodes where alveolar gas appears in the first prediction mode;
s4, comparing the verification sample with the first prediction mode to evaluate an alveolar gas recognition model;
s5, optimizing and correcting an alveolar gas identification model by taking accurate prediction of alveolar gas occurrence nodes as a target;
s6, training the alveolar gas recognition model according to the steps S2 to S5 by using a plurality of subjects.
Further, the method for treating the respiratory process of the subject by the alveolar gas identification model comprises the following steps: a redundant matrix with the number of lines smaller than the number of columns is pre-constructed, the carbon dioxide sensor and oxygen sensor data of the alveolar gas collection device are randomly sub-sampled to obtain sparse signals, then an observation matrix is constructed to carry out projection transformation on the sparse signals, reconstruction of data signals is achieved through orthogonal matching tracking, trend information of data signal changes of the carbon dioxide sensor and the oxygen sensor is obtained, and nodes where alveolar gas appears are judged.
Further, the method for treating the respiratory process of the subject by the alveolar gas identification model comprises the following steps: setting a one-dimensional signal X= (X) according to a compressed sensing theoretical model 1 ,X 2 ,X 3 ,…,X N )，X∈R N X is a discrete data signal of dimension N X1, sparse matrix ψ= (ψ) 1 ,Ψ 2 ,Ψ 3 ,…,Ψ N ),dΨ∈R N*N Obtaining a sparse signal through sparse transformation, wherein the sparsity of X is K (K<<N), according to the formula: x=ψx performing sparse signal projective transformation;
wherein, psi is a sparse matrix, X is an original signal; x epsilon R N*1 Is a sparse matrix transformed signal;
the observation matrix is assumed to be Φ= (Φ) 1 ,Φ 2 ,Φ 3 ,…,Φ N )，Φ∈R M*N Is a unit vector, and M<<And N, carrying out projective transformation on the sparse signals through an observation matrix, wherein the formula is as follows: y=Φ×x to obtain an observed value y∈r M ；
Wherein phi is an observation matrix, and y is an observation value;
by the above formula, the y=Φx=Φ×ψ=η×x, where η≡Φ×ψ is a sensing matrix;
reconstruction optimization algorithm: recovering an original signal from the observed value, adopting a greedy tracking algorithm, namely utilizing the correlation between atoms in an observation matrix and signal residual errors, adding iteration conditions, and solving the optimal sparse value of the original data signal by using a least square method to judge the node where alveolar gas appears;
the method comprises the following steps of selecting an orthogonal matching pursuit algorithm to reconstruct a target gas concentration signal:
input: m×n-dimensional perceptual matrix η=Φψ, m×1 measured value y, signal sparsity K;
and (3) outputting: sparse representation coefficient of signal is Yib;
wherein use is made of
Representing the empty set, Λ t Set of indices representing t iterations, J t Index number representing t iterations;
initializing: residual r 0 =y, index set
The number of iterations t=1;
the method comprises the following steps:
step 1), finding out residual error r t-1 And a perception matrix eta j The maximum value in the inner product is found by index J t So that J t ＝argmax j ＝1，2，…，n|<r t-1 ，η j >I, wherein eta j Is the j-th column of the sensing matrix;
step 2), updating the index set Λt=Λ t-1 ∪{J t Recording the reconstructed atom set D in the found perceptual matrix t ＝[D t-1 ，η Jt ]；
Step 3), obtaining an approximate solution of the signal by a least square method: b t ＝argmin‖y-D t b t ‖ 2 ；
Step 4), updating residual r t ＝y-Dt＾bt；
Step 5), let t=t+1, if t < K or ||r t ‖ 2 And (2) if the number is not less than epsilon, executing the step (1), and sequentially iterating;
and circularly executing the steps 1) to 5), and solving the optimal sparse value of the original signal.
Further, the method for optimizing and correcting the alveolar gas identification model comprises the following steps: acquiring the time and concentration (Ti, ci, ti, ci) and corresponding trigger nodes (Si, si) of the carbon dioxide sensor and the oxygen sensor of the alveolar gas collection device at the occurrence of alveolar gas in the respiratory cycle of a plurality of subjects; training an alveolar gas recognition model by taking the (Ti, ci, ti, ci) as an independent variable and a trigger node (Si, si) corresponding to the exhalation state of a plurality of subjects as the dependent variable; acquiring concentration signals (Ti ', ci ', ti ', ci ') in unknown fragment exhalation states, and processing the signals (Ti ', ci ', ti ', ci ') by using a trained prediction model to obtain current trigger nodes (Si ' ); when the trigger node is reached, alveolar gas appears, and the alveolar gas collection device starts active gas collection, as shown in fig. 4 and 5.
Alveolar gas collection principle:
the active and passive gas production modes are switched through the micropump and the nose wing clamp; carrying out data fusion processing on the collected signals of the plurality of sensors through a programmable logic controller; the method comprises the steps of processing time series data acquired by a sensor by using a compressed sensing idea in machine learning, extracting and reconstructing trend information by using an orthogonal matching pursuit method, predicting the occurrence of alveolar gas in advance, and actively collecting gas in the occurrence stage of the alveolar gas by using the device; the working state of the micro pump is controlled through the pressure difference count value, so that the smoothness of an air path when a human body inhales is ensured, and the gas sampling mode error caused by the predictive failure of an algorithm is avoided.
Claims (7)
1. A training method of an alveolar gas recognition model, comprising the steps of:
s1, presetting an alveolar gas identification model based on a compressed sensing model;
s2, acquiring a complete respiration process of a first subject as a verification sample through an alveolar gas acquisition device;
s3, intercepting part of the respiratory process of the first subject as a first training sample, inputting the first training sample into an alveolar gas identification model, predicting the obtained data as a first prediction mode, and identifying nodes where alveolar gas appears in the first prediction mode;
s4, comparing the verification sample with the first prediction mode to evaluate an alveolar gas recognition model;
s5, optimizing and correcting an alveolar gas identification model by taking accurate prediction of alveolar gas occurrence nodes as a target;
s6, training the alveolar gas recognition model according to the steps S2 to S5 by using a plurality of subjects.
2. The method of training an alveolar gas recognition model as defined in claim 1, wherein the method of processing a subject's respiratory process using the alveolar gas recognition model comprises: a redundant matrix with the number of lines smaller than the number of columns is pre-constructed, the carbon dioxide sensor and oxygen sensor data of the alveolar gas collection device are randomly sub-sampled to obtain sparse signals, then an observation matrix is constructed to carry out projection transformation on the sparse signals, reconstruction of data signals is achieved through orthogonal matching tracking, trend information of data signal changes of the carbon dioxide sensor and the oxygen sensor is obtained, and nodes where alveolar gas appears are judged.
3. The method of training an alveolar gas recognition model as defined in claim 2, wherein the method of processing a subject's respiratory process using the alveolar gas recognition model comprises: setting a one-dimensional signal X= (X) according to a compressed sensing theoretical model 1 ,X 2 ,X 3 ,…,X N )，X∈R N X is a discrete data signal of dimension N X1, sparse matrix ψ= (ψ) 1 ,Ψ 2 ,Ψ 3 ,…,Ψ N ),dΨ∈R N*N Obtaining a sparse signal through sparse transformation, wherein the sparsity of X is K (K<<N), according to the formula: x=ψx performing sparse signal projective transformation;
wherein, psi is a sparse matrix, X is an original signal; x epsilon R N*1 Is a sparse matrix transformed signal;
the observation matrix is assumed to be Φ= (Φ) 1 ,Φ 2 ,Φ 3 ,…,Φ N )，Φ∈R M*N Is a unit vector, and M<<And N, carrying out projective transformation on the sparse signals through an observation matrix, wherein the formula is as follows: y=Φ×x to obtain an observed value y∈r M ；
Wherein phi is an observation matrix, and y is an observation value;
by the above formula, the y=Φx=Φ×ψ=η×x, where η≡Φ×ψ is a sensing matrix;
the method comprises the following steps of selecting an orthogonal matching pursuit algorithm to reconstruct a target gas concentration signal:
input: m×n-dimensional perceptual matrix η=Φψ, m×1 measured value y, signal sparsity K;
and (3) outputting: sparse representation coefficient of signal is Yib;
wherein use is made of
Representing the empty set, Λ t Set of indices representing t iterations, J t Index number representing t iterations;
initializing: residual r 0 =y, index set
The number of iterations t=1;
the method comprises the following steps:
step 1), finding out residual error r t-1 And a perception matrix eta j The maximum value in the inner product is found by index J t So that J t ＝argmax j ＝1，2，…，n|<r t-1 ，η j >I, wherein eta j Is the j-th column of the sensing matrix;
step 2), updating the index set Λt=Λ t-1 ∪{J t Recording the reconstructed atom set D in the found perceptual matrix t ＝[D t-1 ，η Jt ]；
Step 3), obtaining an approximate solution of the signal by a least square method: b t ＝argmin‖y-D t b t ‖ 2 ；
Step 4), updating residual r t ＝y-Dt＾bt；
Step 5), let t=t+1, if t < K or ||r t ‖ 2 And (2) if the number is not less than epsilon, executing the step (1), and sequentially iterating;
and circularly executing the steps 1) to 5), and solving the optimal sparse value of the original signal.
4. The method of training an alveolar gas recognition model as defined in claim 1, wherein the method of optimally correcting the alveolar gas recognition model comprises: acquiring the time and concentration (Ti, ci, ti, ci) and corresponding trigger nodes (Si, si) of the carbon dioxide sensor and the oxygen sensor of the alveolar gas collection device at the occurrence of alveolar gas in the respiratory cycle of a plurality of subjects; training an alveolar gas recognition model by taking the (Ti, ci, ti, ci) as an independent variable and a trigger node (Si, si) corresponding to the exhalation state of a plurality of subjects as the dependent variable; acquiring concentration signals (Ti ', ci ', ti ', ci ') in unknown fragment exhalation states, and processing the signals (Ti ', ci ', ti ', ci ') by using a trained prediction model to obtain current trigger nodes (Si ' ); when the trigger node is reached, the alveolar gas collection device starts active gas collection.
5. The alveolar gas collection device is characterized by comprising a half mask (2), a nose wing clip (1), a carbon dioxide sensor (3), an oxygen concentration sensor (4), a differential pressure meter (5), a micropump (6), a conductive gas circuit (7), an ultrafine glass fiber filter membrane (8) and a programmable logic controller (10) with an alveolar gas identification model arranged in the air chamber; the nose wing clip (1) is connected with the top of the half mask (2); the conduction gas circuit (7) is fixed outside the half mask (2) and is communicated with the half mask (2); the programmable logic controller (10) is fixed on the outer side surface of the half mask (2), and the carbon dioxide sensor (3) and the oxygen concentration sensor (4) are arranged on the inner side surface of the half mask (2); the superfine glass fiber filter membrane (8), the differential pressure gauge (5) and the micropump (6) are sequentially arranged in the conducting gas path (7) from top to bottom; the conduction gas circuit (7) is connected with a gas collecting device; the flow guiding direction of the micropump (6) always flows from the inside of the half mask (2) to the outside through the air guiding path (7); the carbon dioxide sensor (3), the oxygen concentration sensor (4), the differential pressure meter (5), the micropump (6) and the gas collecting device are respectively and electrically connected with the programmable logic controller (10).
6. An alveolar gas collection device as claimed in claim 5 wherein polytetrafluoroethylene microporous membrane (9) is uniformly coated on the surfaces of the carbon dioxide sensor (3), oxygen concentration sensor (4), differential pressure gauge (5) and micropump (6).
7. An alveolar gas collection device as claimed in claim 5 wherein the carbon dioxide sensor (3) and oxygen concentration sensor (4) are removably disposed centrally on the inner side of the half mask (2).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211618120.9A CN116028845A (en) | 2022-12-15 | 2022-12-15 | Training method of alveolar gas recognition model and alveolar gas acquisition device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211618120.9A CN116028845A (en) | 2022-12-15 | 2022-12-15 | Training method of alveolar gas recognition model and alveolar gas acquisition device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116028845A true CN116028845A (en) | 2023-04-28 |
Family
ID=86078966
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211618120.9A Pending CN116028845A (en) | 2022-12-15 | 2022-12-15 | Training method of alveolar gas recognition model and alveolar gas acquisition device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116028845A (en) |
-
2022
- 2022-12-15 CN CN202211618120.9A patent/CN116028845A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10952640B2 (en) | Method for collecting a selective portion of a subject's breath | |
| US9144396B2 (en) | Apparatus and method of analyzing constituents of gas in oral cavity and alveolar gas | |
| US20160256072A1 (en) | Portable pneumotachograph for measuring components of an expiration volume | |
| CN109310368A (en) | For capturing the system and equipment of breath sample | |
| CN110123382B (en) | Human body exhaled gas off-line acquisition device and method | |
| US9532731B2 (en) | Method and apparatus for measuring the concentration of a gas in exhaled air | |
| CN106456053A (en) | Selection, segmentation and analysis of exhaled breath for airway disorders assessment | |
| US20170191910A1 (en) | Method, computing device and system for collecting exhaled breath | |
| CN106061381B (en) | Detect the method for ARDS and the system for detecting ARDS | |
| CN102458244A (en) | Apparatus and method for monitoring degree of integration between functions of heart and lungs, and therapeutic success of resuscitative interventions | |
| US9999373B2 (en) | On-airway pulmonary function tester | |
| CN110146651A (en) | Expiratory air detection system based on gas sensor | |
| WO2009025488A2 (en) | Apparatus and method of analyzing constituents of gas in oral cavity and alveolar gas | |
| CN205228892U (en) | Last sampling device of exhaling | |
| WO2014117747A2 (en) | Systems and methods using exhaled breath for medical diagnostics and treatment | |
| US10682073B2 (en) | Measurement device and method for human respiratory system function | |
| CN205411221U (en) | Guardianship sampling device exhales | |
| CN105496412B (en) | A kind of expiration inflammation monitoring method and device | |
| CN116028845A (en) | Training method of alveolar gas recognition model and alveolar gas acquisition device | |
| CN110554028A (en) | Gas detection method and gas detection system based on same | |
| CN203465233U (en) | Handheld expiration analyzer | |
| Guo et al. | Diabetes identification and classification by means of a breath analysis system | |
| CN210742134U (en) | Gas detection system | |
| CN114177446A (en) | Intelligent oxygen supply system | |
| IT202000032249A1 (en) | PATHOLOGICAL STATE IDENTIFICATION APPARATUS AND RELATED METHOD |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |