CN114121250B - Safety detection method for breathing machine - Google Patents
Safety detection method for breathing machine Download PDFInfo
- Publication number
- CN114121250B CN114121250B CN202111292272.XA CN202111292272A CN114121250B CN 114121250 B CN114121250 B CN 114121250B CN 202111292272 A CN202111292272 A CN 202111292272A CN 114121250 B CN114121250 B CN 114121250B
- Authority
- CN
- China
- Prior art keywords
- network
- data
- breathing machine
- air leakage
- super
- 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.)
- Active
Links
- 230000029058 respiratory gaseous exchange Effects 0.000 title claims abstract description 117
- 238000001514 detection method Methods 0.000 title claims abstract description 66
- 238000003062 neural network model Methods 0.000 claims abstract description 83
- 238000000034 method Methods 0.000 claims abstract description 60
- 230000002068 genetic effect Effects 0.000 claims abstract description 28
- 238000007781 pre-processing Methods 0.000 claims abstract description 27
- 238000013528 artificial neural network Methods 0.000 claims abstract description 22
- 210000005036 nerve Anatomy 0.000 claims description 9
- 238000005215 recombination Methods 0.000 claims description 4
- 230000006798 recombination Effects 0.000 claims description 4
- 239000011148 porous material Substances 0.000 claims description 3
- 230000001502 supplementing effect Effects 0.000 claims description 3
- 238000011897 real-time detection Methods 0.000 abstract description 2
- 208000028399 Critical Illness Diseases 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
- G16H40/40—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management of medical equipment or devices, e.g. scheduling maintenance or upgrades
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Business, Economics & Management (AREA)
- Physiology (AREA)
- General Business, Economics & Management (AREA)
- Genetics & Genomics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Epidemiology (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Respiratory Apparatuses And Protective Means (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of safety detection of respirators, and discloses a method for detecting the safety of a respirator, which comprises the following steps: collecting data by taking the flow rate of the gas in the pipeline of the breathing machine, the aperture of the leakage gas, the tidal volume of inspiration and the tidal volume of expiration as indexes; taking the preprocessed collected data as training data, and training the super network by combining the training data; searching a neural network structure of the trained super network by utilizing a genetic algorithm to obtain a breathing machine air leakage detection neural network model; the method comprises the steps of carrying out data preprocessing on index data of a to-be-detected breathing machine, taking the preprocessed data as input of a breathing machine air leakage detection neural network model, and judging whether the breathing machine has air leakage potential safety hazards or not by utilizing the breathing machine air leakage detection neural network model. According to the method, the ventilator safety detection based on the ventilator air leakage is realized by establishing the ventilator air leakage detection neural network model, and the real-time detection of the ventilator is realized.
Description
Technical Field
The invention relates to the technical field of ventilator safety detection, in particular to a ventilator safety detection method.
Background
The breathing machine is one of the indispensable medical equipment of each hospital, plays a vital role in rescuing and treating patients, and the treatment object of the breathing machine is mainly critical illness patients, and is widely applied to the fields of intensive care, emergency recovery and the like. How to implement ventilator safety detection is a hot topic of the current research field.
The ventilator has the following safety risks in the using process, the display value of the ventilator is inconsistent with the set value, and the deviation between the display value such as tidal volume, inhalation-exhalation ratio and the like and the set value of a doctor is large; alarm failure, such as air source pressure alarm failure and the like; the main factor responsible for these safety risks comes from ventilator leaks.
In view of the above, the invention provides a ventilator safety detection method, which realizes ventilator safety detection based on ventilator air leakage by establishing a ventilator air leakage detection neural network model, and realizes real-time detection of a ventilator.
Disclosure of Invention
The invention provides a ventilator safety detection method, which aims at (1) realizing ventilator safety detection based on ventilator air leakage; (2) realizing real-time safety detection of the breathing machine.
The invention provides a safety detection method of a breathing machine, which comprises the following steps:
s1: collecting data by taking the flow rate of the gas in the pipeline of the breathing machine, the aperture of the leakage gas, the tidal volume of inspiration, the tidal volume of expiration and the gas leakage as indexes, and preprocessing the collected data;
s2: taking the preprocessed collected data as training data, and training the super network by combining the training data;
s3: searching a neural network structure of the trained super network by utilizing a genetic algorithm to obtain an optimized breathing machine air leakage detection neural network model;
s4: and carrying out data preprocessing on index data of the breathing machine to be detected, taking the preprocessed data as input of a breathing machine air leakage detection neural network model, and judging whether the breathing machine has air leakage potential safety hazards or not according to an output result of the breathing machine air leakage detection neural network model.
As a further improvement of the present invention:
and in the step S1, collecting the breathing machine index data, which comprises the following steps:
dividing the ventilator leakage aperture into 5 aperture levels [1,2,3,4,5], and setting 10 groups of different ventilator pipeline gas flow rates;
selecting a breathing machine with an air leakage aperture level of h, wherein the initial value of h is 1, adjusting the gas flow rate in a breathing machine pipeline according to the set 10 groups of different breathing machine pipeline gas flow rates, and acquiring the data of the inspiration tidal volume, expiration tidal volume and air leakage volume of the breathing machine under the current funnel aperture level and the gas flow rate;
setting the air leakage aperture level h to be h+1, and repeating the steps until h=5;
the collected ventilator index data are:
Data={(h 1 ,v 1 ,s 1,1 ,c 1,1 ,e 1,1 ),…,(h 1 ,v 10 ,s 1,10 ,c 1,10 ,e 1,10 ),…,(h 5 ,v 10 ,s 5,10 ,c 5,10 ,e 5,10 )}
wherein:
h i represents the funnel pore size class i, i.e. [1,2,3,4,5]];
v j Representing the gas flow rate of the pipeline of the arranged j group suction machine;
s i,j the method comprises the steps that the aperture level i of a funnel is represented, and when the gas flow rate in a pipeline of a breathing machine is the j-th set of gas flow rates, the acquired inspiration tidal volume is obtained;
c i,j indicating the aperture level i of the funnel, and collecting the expiratory tidal volume when the gas flow rate in the pipeline of the breathing machine is the j-th set of gas flow rates;
e i,j the method comprises the steps that the aperture level i of a funnel is represented, and when the gas flow rate in a pipeline of a breathing machine is the j-th set of gas flow rate, the collected air leakage is detected;
(h i ,v j ,s i,j ,c i,j ,e i,j ) For a set of ventilator index data collected, in one embodiment of the present invention, 50 sets of ventilator index data are collected;
data is the collected ventilator index dataset.
In the step S1, data preprocessing is performed on the collected ventilator index data, and the method comprises the following steps:
carrying out Data preprocessing on the collected ventilator index Data to obtain a preprocessed ventilator index Data set Data', wherein the Data preprocessing flow is as follows:
wherein:
x k the kth Data of an index x in the breathing machine index Data set Data is represented, wherein the index x to be preprocessed comprises the breathing machine pipeline gas flow rate, the air leakage aperture, the inspiration tidal volume and the expiration tidal volume;
x′ k kth preprocessed Data representing the index x in the ventilator index Data set Data;
x min representing the minimum value of the index x in the ventilator index Data set Data;
x max represents the maximum value of the index x in the ventilator index Data set Data.
In the step S2, training the super network by using the preprocessed data as training data, where training obtains the super network, and the method includes:
constructing a plurality of three-layer neural network models, wherein the three-layer structure of the neural network models is an input layer, a hidden layer and a full-connection layer;
constructing three super network layers, wherein a first layer of the super network is an input layer of a plurality of nerve network models in parallel, a second layer of the super network is a hidden layer of the plurality of nerve network models in parallel, and a third layer of the super network is a full connection layer of the plurality of nerve network models in parallel;
input layer F1 for any neural network model in the first layer of the super network i Sequentially connecting the neural network model hidden layers in the second layer of the super network to obtain N F1-based neural network models i An input layer-hidden layer model that is an input layer, where N represents the number of neural network models constructed;
input layer F2 for any neural network model in the second layer of the super network j Sequentially connecting all the connecting layers of the neural network model in the third layer of the super network to obtainUp to N with F2 j A hidden layer-fully connected layer model that is a hidden layer, wherein N represents the number of neural network models constructed; taking any input layer-hidden layer-full connection layer model obtained by connection as a sub-network in the super-network;
the search space S of the pre-defined super-network parameters is used for inputting the ventilator pipeline gas flow rate, the air leakage aperture, the inspiration tidal volume and the expiration tidal volume in the ventilator index Data set Data' into the first layer of the super-network, and setting the loss function of the super-network training as follows:
wherein:
representing a loss function of the sub-network g;
L net representing a loss function of the super network training;
w g a weight representing the sub-network g;
e i,j the method comprises the steps that the aperture level i of a funnel is represented, and when the gas flow rate in a pipeline of a breathing machine is the j-th set of gas flow rate, the collected air leakage is detected;
when the aperture level i of the funnel is represented and the gas flow rate in the pipeline of the breathing machine is the j-th set of gas flow rate, the sub-network g inputs the predicted air leakage in the full-connection layer;
g represents the number of subnetworks in the super network;
the training formula for the super network is:
wherein:
NET represents a super network;
w represents a weight combination of subnetworks in the super network;
W S representing the super-network weight combination obtained by training;
inputting training data into the super network until the training obtained super network weight combination W S And (5) convergence.
And in the step S3, searching the neural network structure of the trained super network by utilizing a genetic algorithm, wherein the method comprises the following steps:
and searching the neural network structure of the trained super network by utilizing a genetic algorithm, wherein the neural network structure searching flow based on the genetic algorithm is as follows:
1) Recording the loss function value of each sub-network in the super-network training process:
wherein:
L g a loss function value indicating a sub-network g;
selecting m sub-networks with the minimum loss function value as an initial population of a genetic algorithm, initializing the maximum iteration word number Max of the genetic algorithm, and carrying out coding treatment on a sub-network structure of the initial population, wherein in a specific embodiment of the invention, the sub-network structure is coded by using a single-heat coding method, and the obtained sub-network structure coding result is as follows:
{z 1 z 2 z 3 }
wherein:
z 1 representing the binary encoding result of the input layer in the sub-network structure;
z 2 representing the binary coding result of the hidden layer in the sub-network structure;
z 2 representing full connectivity in a subnetwork structureA binary encoding result of the layer;
2) Calculating a loss function value of each individual in the population, and taking a sub-network with the minimum loss function value in the population as a winning individual a, wherein each individual in the population is a sub-network;
3) Replacing and recombining partial structures of the winning individual a to generate a new individual, wherein the replacing and recombining of the winning individual structure is to replace and recombine an input layer, a hidden layer and a full connection layer in a sub-network; the replacement recombination probability of the winning individual is as follows:
wherein:
L avg representing population loss function averages;
L a a loss function value representing a winning individual a;
L min representing a population loss function minimum;
b 1 ,b 2 is [0,1]Constant of b 1 =0.4,b 2 =0.6;
4) Repeating the steps 2) -3) until the maximum iteration number Max is reached, wherein the recombined sub-network structure in the population is the neural network structure obtained by searching.
In the step S3, selecting a sub-network with the optimal structural search in the super-network as an optimized breathing machine air leakage detection neural network model, wherein the method comprises the following steps:
calculating the loss function values of the sub-network individuals in the population after the genetic algorithm is iterated, and selecting K sub-network individuals with the minimum loss function values as training individuals; combining the K sub-network individuals into a super-network, and calculating the weight of each sub-network individual by using a super-network training formula, wherein the super-network training formula is as follows:
wherein:
NET' represents the reorganized super network;
w represents a weight combination of subnetworks in the super network;
W S representing the super-network weight combination obtained by training;
inputting training data into the super network until the training obtained super network weight combination W S Converging, and selecting the sub-network with the largest weight as an optimized breathing machine air leakage detection neural network model.
And S4, selecting the breathing machine index data to be detected for data preprocessing to obtain the input value of the breathing machine air leakage detection neural network model, wherein the method comprises the following steps:
acquiring index data of a to-be-detected breathing machine, wherein the acquired index data comprise the flow rate of the breathing machine pipeline gas, the air leakage aperture, the inspiration tidal volume and the expiration tidal volume; and supplementing the training set data to the index data of the breathing machine to be detected to obtain a supplementary data set, carrying out data preprocessing on the index data of the breathing machine to be detected in the supplementary data set, and taking the index data after the data preprocessing as an input value of a breathing machine air leakage detection neural network model.
In the step S4, an input value of the ventilator air leakage detection neural network model is input into the ventilator air leakage detection neural network model, and a judgment result of whether the ventilator has an air leakage potential safety hazard or not is input, including:
and taking the preprocessed data as input of a breathing machine air leakage detection neural network model, outputting the air leakage of the current detection breathing machine by the breathing machine air leakage detection neural network model, setting an air leakage threshold Q, and if the detected air leakage amount of the breathing machine is higher than the air leakage threshold Q, indicating that the current detection breathing machine has air leakage potential safety hazards.
Compared with the prior art, the invention provides a safety detection method of a breathing machine, which has the following advantages:
firstly, the proposal provides a super network training method, by constructing a three-layer super network, all possible neural network models are used as sub-graphs of the super network, namely, the super network training method is constructedThe method comprises the steps of constructing a plurality of three-layer neural network models, and constructing three super network layers, wherein a first layer of the super network is an input layer of the plurality of parallel neural network models, a second layer of the super network is a hidden layer of the plurality of parallel neural network models, and a third layer of the super network is a full connection layer of the plurality of parallel neural network models; input layer F1 for any neural network model in the first layer of the super network i Sequentially connecting the neural network model hidden layers in the second layer of the super network to obtain N F1-based neural network models i An input layer-hidden layer model that is an input layer, where N represents the number of neural network models constructed; input layer F2 for any neural network model in the second layer of the super network j Sequentially connecting all the connecting layers of the neural network model in the third layer of the super network to obtain N F2-based data j A hidden layer-fully connected layer model that is a hidden layer, wherein N represents the number of neural network models constructed; taking any input layer-hidden layer-full connection layer model obtained by connection as a sub-network in the super-network; the constructed N neural network models are converted into N subgraphs in the super network, so that models with different neural network structures can be trained, and the robustness of the final ventilator air leakage detection neural network model is improved.
Meanwhile, the scheme utilizes a genetic algorithm to search the neural network structure of the trained super network, and the neural network structure searching flow based on the genetic algorithm is as follows: recording a loss function value of each sub-network in the super-network training process, selecting m sub-networks with the minimum loss function value as an initial population of a genetic algorithm, initializing the maximum iteration word number Max of the genetic algorithm, carrying out coding processing on the sub-network structure of the initial population, and taking the sub-network with the minimum loss function value in the population as a winning individual a by calculating the loss function value of each individual in the population, wherein each individual in the population is the sub-network; replacing and recombining partial structures of the winning individual a to generate a new individual, wherein the replacing and recombining of the winning individual structure is that an input layer, a hidden layer and a full-connection layer in a sub-network are replaced and recombined, so that the optimization of the neural network structure is realized in algorithm iteration; the replacement recombination probability of the winning individual is as follows:
wherein: l (L) avg Representing population loss function averages; l (L) a A loss function value representing a winning individual a; l (L) min Representing a population loss function minimum; b 1 ,b 2 Is [0,1]Constant of b 1 =0.4,b 2 =0.6; repeating the iterative steps until the maximum iterative times are reached, wherein the recombined sub-network structure in the population is the neural network structure obtained by searching. Calculating the loss function value of population sub-network individuals after the iteration of a genetic algorithm, selecting K sub-network individuals with the minimum loss function value as training individuals, retraining the K sub-network individuals with the best performance by selecting the K sub-networks, combining the K sub-network individuals into a super-network, and calculating the weight of each sub-network individual by using a super-network training formula, thereby effectively avoiding bad sub-network performance with high weight in the pre-trained super-network, wherein the super-network training formula is as follows:
wherein: NET' represents the reorganized super network; w represents a weight combination of subnetworks in the super network; w (W) S Representing the super-network weight combination obtained by training; inputting training data into the super network until the training obtained super network weight combination W S Converging, and selecting the sub-network with the largest weight as an optimized breathing machine air leakage detection neural network model. And taking the index data of the ventilator to be detected after pretreatment as the input of a ventilator air leakage detection neural network model, outputting the air leakage of the current detected ventilator by the ventilator air leakage detection neural network model, setting an air leakage threshold Q, and if the detected ventilator air leakage is higher than the air leakage threshold Q, indicating that the current detected ventilator has air leakage potential safety hazards, thereby realizing the safety detection of the ventilator based on the air leakage of the ventilator.
Drawings
FIG. 1 is a schematic flow chart of a method for detecting safety of a ventilator according to an embodiment of the present invention;
the achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
S1: collecting data by taking the flow rate of the gas in the pipeline of the breathing machine, the aperture of the leakage gas, the tidal volume of inspiration, the tidal volume of expiration and the gas leakage as indexes, and preprocessing the collected data.
And in the step S1, collecting the breathing machine index data, which comprises the following steps:
dividing the ventilator leakage aperture into 5 aperture levels [1,2,3,4,5], and setting 10 groups of different ventilator pipeline gas flow rates;
selecting a breathing machine with an air leakage aperture level of h, wherein the initial value of h is 1, adjusting the gas flow rate in a breathing machine pipeline according to the set 10 groups of different breathing machine pipeline gas flow rates, and acquiring the data of the inspiration tidal volume, expiration tidal volume and air leakage volume of the breathing machine under the current funnel aperture level and the gas flow rate;
setting the air leakage aperture level h to be h+1, and repeating the steps until h=5;
the collected ventilator index data are:
Data={(h 1 ,v 1 ,s 1,1 ,c 1,1 ,e 1,1 ),…,(h 1 ,v 10 ,s 1,10 ,c 1,10 ,e 1,10 ),…,(h 5 ,v 10 ,s 5,10 ,c 5,10 ,e 5,10 )}
wherein:
h i represents the funnel pore size class i, i.e. [1,2,3,4,5]];
v j Representing the gas flow rate of the pipeline of the arranged j group suction machine;
s i,j indicating the aperture of the funnelThe level i, and when the gas flow rate in the pipeline of the breathing machine is the j-th set of gas flow rate, the acquired inspiration tidal volume;
c i,j indicating the aperture level i of the funnel, and collecting the expiratory tidal volume when the gas flow rate in the pipeline of the breathing machine is the j-th set of gas flow rates;
e i,j the method comprises the steps that the aperture level i of a funnel is represented, and when the gas flow rate in a pipeline of a breathing machine is the j-th set of gas flow rate, the collected air leakage is detected;
(h i ,v j ,s i,j ,c i,j ,e i,j ) For a set of ventilator index data collected, in one embodiment of the present invention, 50 sets of ventilator index data are collected;
data is the collected ventilator index dataset.
In the step S1, data preprocessing is performed on the collected ventilator index data, and the method comprises the following steps:
carrying out Data preprocessing on the collected ventilator index Data to obtain a preprocessed ventilator index Data set Data', wherein the Data preprocessing flow is as follows:
wherein:
x k the kth Data of an index x in the breathing machine index Data set Data is represented, wherein the index x to be preprocessed comprises the breathing machine pipeline gas flow rate, the air leakage aperture, the inspiration tidal volume and the expiration tidal volume;
x′ k kth preprocessed Data representing the index x in the ventilator index Data set Data;
x min representing the minimum value of the index x in the ventilator index Data set Data;
x max represents the maximum value of the index x in the ventilator index Data set Data.
S2: and taking the preprocessed collected data as training data, and training the super network by combining the training data.
In the step S2, training the super network by using the preprocessed data as training data, where training obtains the super network, and the method includes:
constructing a plurality of three-layer neural network models, wherein the three-layer structure of the neural network models is an input layer, a hidden layer and a full-connection layer;
constructing three super network layers, wherein a first layer of the super network is an input layer of a plurality of nerve network models in parallel, a second layer of the super network is a hidden layer of the plurality of nerve network models in parallel, and a third layer of the super network is a full connection layer of the plurality of nerve network models in parallel;
input layer F1 for any neural network model in the first layer of the super network i Sequentially connecting the neural network model hidden layers in the second layer of the super network to obtain N F1-based neural network models i An input layer-hidden layer model that is an input layer, where N represents the number of neural network models constructed;
input layer F2 for any neural network model in the second layer of the super network j Sequentially connecting all the connecting layers of the neural network model in the third layer of the super network to obtain N F2-based data j A hidden layer-fully connected layer model that is a hidden layer, wherein N represents the number of neural network models constructed; taking any input layer-hidden layer-full connection layer model obtained by connection as a sub-network in the super-network;
the search space S of the pre-defined super-network parameters is used for inputting the ventilator pipeline gas flow rate, the air leakage aperture, the inspiration tidal volume and the expiration tidal volume in the ventilator index Data set Data' into the first layer of the super-network, and setting the loss function of the super-network training as follows:
wherein:
representing a loss function of the sub-network g;
L net representing a loss function of the super network training;
w g a weight representing the sub-network g;
e i,j the method comprises the steps that the aperture level i of a funnel is represented, and when the gas flow rate in a pipeline of a breathing machine is the j-th set of gas flow rate, the collected air leakage is detected;
when the aperture level i of the funnel is represented and the gas flow rate in the pipeline of the breathing machine is the j-th set of gas flow rate, the sub-network g inputs the predicted air leakage in the full-connection layer;
g represents the number of subnetworks in the super network;
the training formula for the super network is:
wherein:
NET represents a super network;
w represents a weight combination of subnetworks in the super network;
W S representing the super-network weight combination obtained by training;
inputting training data into the super network until the training obtained super network weight combination W S And (5) convergence.
S3: and searching a neural network structure of the trained super network by utilizing a genetic algorithm to obtain an optimized breathing machine air leakage detection neural network model.
And in the step S3, searching the neural network structure of the trained super network by utilizing a genetic algorithm, wherein the method comprises the following steps:
and searching the neural network structure of the trained super network by utilizing a genetic algorithm, wherein the neural network structure searching flow based on the genetic algorithm is as follows:
1) Recording the loss function value of each sub-network in the super-network training process:
wherein:
L g a loss function value indicating a sub-network g;
selecting m sub-networks with the minimum loss function value as an initial population of a genetic algorithm, initializing the maximum iteration word number Max of the genetic algorithm, and carrying out coding treatment on a sub-network structure of the initial population, wherein in a specific embodiment of the invention, the sub-network structure is coded by using a single-heat coding method, and the obtained sub-network structure coding result is as follows:
{z 1 z 2 z 3 }
wherein:
z 1 representing the binary encoding result of the input layer in the sub-network structure;
z 2 representing the binary coding result of the hidden layer in the sub-network structure;
z 2 representing the binary coding result of the full connection layer in the sub-network structure;
2) Calculating a loss function value of each individual in the population, and taking a sub-network with the minimum loss function value in the population as a winning individual a, wherein each individual in the population is a sub-network;
3) Replacing and recombining partial structures of the winning individual a to generate a new individual, wherein the replacing and recombining of the winning individual structure is to replace and recombine an input layer, a hidden layer and a full connection layer in a sub-network; the replacement recombination probability of the winning individual is as follows:
wherein:
L avg representing population loss function averages;
L a a loss function value representing a winning individual a;
L min representing a population loss function minimum;
b 1 ,b 2 is [0,1]Constant of b 1 =0.4,b 2 =0.6;
4) Repeating the steps 2) -3) until the maximum iteration number Max is reached, wherein the recombined sub-network structure in the population is the neural network structure obtained by searching.
In the step S3, selecting a sub-network with the optimal structural search in the super-network as an optimized breathing machine air leakage detection neural network model, wherein the method comprises the following steps:
calculating the loss function values of the sub-network individuals in the population after the genetic algorithm is iterated, and selecting K sub-network individuals with the minimum loss function values as training individuals; combining the K sub-network individuals into a super-network, and calculating the weight of each sub-network individual by using a super-network training formula, wherein the super-network training formula is as follows:
wherein:
NET' represents the reorganized super network;
w represents a weight combination of subnetworks in the super network;
W S representing the super-network weight combination obtained by training;
inputting training data into the super network until the training obtained super network weight combination W S Converging, and selecting the sub-network with the largest weight as an optimized breathing machine air leakage detection neural network model.
S4: and carrying out data preprocessing on index data of the breathing machine to be detected, taking the preprocessed data as input of a breathing machine air leakage detection neural network model, and judging whether the breathing machine has air leakage potential safety hazards or not according to an output result of the breathing machine air leakage detection neural network model.
And S4, selecting the breathing machine index data to be detected for data preprocessing to obtain the input value of the breathing machine air leakage detection neural network model, wherein the method comprises the following steps:
acquiring index data of a to-be-detected breathing machine, wherein the acquired index data comprise the flow rate of the breathing machine pipeline gas, the air leakage aperture, the inspiration tidal volume and the expiration tidal volume; and supplementing the training set data to the index data of the breathing machine to be detected to obtain a supplementary data set, carrying out data preprocessing on the index data of the breathing machine to be detected in the supplementary data set, and taking the index data after the data preprocessing as an input value of a breathing machine air leakage detection neural network model.
In the step S4, an input value of the ventilator air leakage detection neural network model is input into the ventilator air leakage detection neural network model, and a judgment result of whether the ventilator has an air leakage potential safety hazard or not is input, including:
and taking the preprocessed data as input of a breathing machine air leakage detection neural network model, outputting the air leakage of the current detection breathing machine by the breathing machine air leakage detection neural network model, setting an air leakage threshold Q, and if the detected air leakage amount of the breathing machine is higher than the air leakage threshold Q, indicating that the current detection breathing machine has air leakage potential safety hazards.
It should be noted that, the foregoing reference numerals of the embodiments of the present invention are merely for describing the embodiments, and do not represent the advantages and disadvantages of the embodiments. And the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, apparatus, article or method that comprises the element.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as described above, comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.
Claims (7)
1. A method of ventilator safety detection, the method comprising:
s1: collecting data by taking the flow rate of the gas in the pipeline of the breathing machine, the aperture of the leakage gas, the tidal volume of inspiration, the tidal volume of expiration and the gas leakage as indexes, and preprocessing the collected data;
s2: taking the preprocessed collected data as training data, and training the super network by combining the training data;
s3: searching a neural network structure of the trained super network by utilizing a genetic algorithm to obtain an optimized breathing machine air leakage detection neural network model;
s4: the method comprises the steps of carrying out data preprocessing on index data of a to-be-detected breathing machine, taking the preprocessed data as input of a breathing machine air leakage detection neural network model, and judging whether an air leakage potential safety hazard exists in the breathing machine or not according to an output result of the breathing machine air leakage detection neural network model;
and in the step S3, searching the neural network structure of the trained super network by utilizing a genetic algorithm, wherein the method comprises the following steps:
the neural network structure search is carried out on the trained super network by utilizing a genetic algorithm, and the neural network structure search flow based on the genetic algorithm is as follows:
1) Recording the loss function value of each sub-network in the super-network training process:
wherein:
L g a loss function value indicating a sub-network g;
when the aperture level i of the funnel is represented and the gas flow rate in the pipeline of the breathing machine is the j-th set of gas flow rate, the sub-network g inputs the predicted air leakage in the full-connection layer;
the method comprises the steps that when the aperture level i of a funnel is represented, and the gas flow rate in a pipeline of a breathing machine is the j-th set of gas flow rate, the sub-network g inputs collected air leakage in a full-connection layer;
selecting m sub-networks with the minimum loss function values as an initial population of a genetic algorithm, initializing the maximum iteration number Max of the genetic algorithm, and carrying out coding treatment on the sub-network structure of the initial population;
2) Calculating a loss function value of each individual in the population, and taking a sub-network with the minimum loss function value in the population as a winning individual a, wherein each individual in the population is a sub-network;
3) Replacing and recombining partial structures of the winning individual a to generate a new individual, wherein the replacing and recombining of the winning individual structure is to replace and recombine an input layer, a hidden layer and a full connection layer in a sub-network; the replacement recombination probability of the winning individual is as follows:
wherein:
L avg representing population loss function averages;
L a a loss function value representing a winning individual a;
L min representing a population loss function minimum;
b 1 ,b 2 is [0,1]Constant of b 1 =0.4,b 2 =0.6;
4) Repeating the steps 2) -3) until the maximum iteration number Max is reached, wherein the recombined sub-network structure in the population is the neural network structure obtained by searching.
2. The method for detecting safety of a ventilator according to claim 1, wherein said S1: collecting data by taking the flow rate of the gas in the pipeline of the breathing machine, the aperture of the leakage gas, the tidal volume of inspiration, the tidal volume of expiration and the gas leakage volume as indexes, and preprocessing the data of the collected data, wherein the method comprises the following steps of:
dividing the ventilator leakage aperture into 5 aperture levels [1,2,3,4,5], and setting 10 groups of different ventilator pipeline gas flow rates;
selecting a breathing machine with an air leakage aperture level of h, wherein the initial value of h is 1, adjusting the gas flow rate in a breathing machine pipeline according to the set 10 groups of different breathing machine pipeline gas flow rates, and acquiring the data of the inspiration tidal volume, expiration tidal volume and air leakage volume of the breathing machine under the current funnel aperture level and the gas flow rate;
setting the air leakage aperture level h to be h+1, and repeating the steps until h=5;
the collected ventilator index data are:
Data={(h 1 ,v 1 ,s 1,1 ,c 1,1 ,e 1,1 ),...,(h 1 ,v 10 ,s 1,10 ,c 1,10 ,e 1,10 ),...,(h 5 ,v 10 ,s 5,10 ,c 5,10 ,e 5,10 )}
wherein:
h i represents the funnel pore size class i, i.e. [1,2,3,4,5]];
v j Representing the gas flow rate of the pipeline of the arranged j group suction machine;
s i,j the method comprises the steps that the aperture level i of a funnel is represented, and when the gas flow rate in a pipeline of a breathing machine is the j-th set of gas flow rates, the acquired inspiration tidal volume is obtained;
c i,j indicating the aperture level i of the funnel, and collecting the expiratory tidal volume when the gas flow rate in the pipeline of the breathing machine is the j-th set of gas flow rates;
e i,j the method comprises the steps that the aperture level i of a funnel is represented, and when the gas flow rate in a pipeline of a breathing machine is the j-th set of gas flow rate, the collected air leakage is detected;
(h i ,v j ,s i,j ,c i,j ,e i,j ) For the acquired set of ventilator index data;
data is the collected ventilator index dataset.
3. The method for detecting safety of a ventilator according to claim 2, wherein said S1: collecting data by taking the flow rate of the gas in the pipeline of the breathing machine, the aperture of the leakage gas, the tidal volume of inspiration, the tidal volume of expiration and the gas leakage volume as indexes, and preprocessing the data of the collected data, wherein the method comprises the following steps of:
carrying out Data preprocessing on the collected ventilator index Data to obtain a preprocessed ventilator index Data set Data', wherein the Data preprocessing flow is as follows:
wherein:
x k the kth Data representing the index x in the ventilator index Data set Data, wherein the index x to be preprocessed comprises ventilator conduit gas flow rate, leakage aperture, inspiration tidal volumeAn expiratory tidal volume;
x′ k kth preprocessed Data representing the index x in the ventilator index Data set Data;
x min representing the minimum value of the index x in the ventilator index Data set Data;
x max represents the maximum value of the index x in the ventilator index Data set Data.
4. A method of ventilator safety detection according to claim 3, wherein S2: taking the preprocessed collected data as training data, and training the super network by combining the training data, wherein the method comprises the following steps:
constructing a plurality of three-layer neural network models, wherein the three-layer structure of the neural network models is an input layer, a hidden layer and a full-connection layer;
constructing three super network layers, wherein a first layer of the super network is an input layer of a plurality of nerve network models in parallel, a second layer of the super network is a hidden layer of the plurality of nerve network models in parallel, and a third layer of the super network is a full connection layer of the plurality of nerve network models in parallel;
input layer F1 for any neural network model in the first layer of the super network i Sequentially connecting the neural network model hidden layers in the second layer of the super network to obtain N F1-based neural network models i An input layer-hidden layer model that is an input layer, where N represents the number of neural network models constructed;
input layer F2 for any neural network model in the second layer of the super network j Sequentially connecting all the connecting layers of the neural network model in the third layer of the super network to obtain N F2-based data j A hidden layer-fully connected layer model that is a hidden layer, wherein N represents the number of neural network models constructed; taking any input layer-hidden layer-full connection layer model obtained by connection as a sub-network in the super-network;
the search space S of the pre-defined super-network parameters is used for inputting the ventilator pipeline gas flow rate, the air leakage aperture, the inspiration tidal volume and the expiration tidal volume in the ventilator index Data set Data' into the first layer of the super-network, and setting the loss function of the super-network training as follows:
wherein:
representing a loss function of the sub-network g;
L net representing a loss function of the super network training;
w g a weight representing the sub-network g;
the method comprises the steps that when the aperture level i of a funnel is represented, and the gas flow rate in a pipeline of a breathing machine is the j-th set of gas flow rate, the sub-network g inputs collected air leakage in a full-connection layer;
when the aperture level i of the funnel is represented and the gas flow rate in the pipeline of the breathing machine is the j-th set of gas flow rate, the sub-network g inputs the predicted air leakage in the full-connection layer;
g represents the number of subnetworks in the super network;
the training formula for the super network is:
wherein:
NET represents a super network;
w represents a weight combination of subnetworks in the super network;
W S representing the super-network weight combination obtained by training;
inputting training data into the super network until the training obtained super network weight combination W S And (5) convergence.
5. The method for detecting safety of a ventilator according to claim 4, wherein said S3: searching the neural network structure of the trained super network by utilizing a genetic algorithm to obtain an optimized breathing machine air leakage detection neural network model, wherein the method comprises the following steps:
calculating the loss function values of the sub-network individuals in the population after the genetic algorithm is iterated, and selecting K sub-network individuals with the minimum loss function values as training individuals; combining the K sub-network individuals into a super network, and calculating the weight of each sub-network individual by using a super network training formula;
inputting training data into the super network until the training obtained super network weight combination W S Converging, and selecting the sub-network with the largest weight as an optimized breathing machine air leakage detection neural network model.
6. The method for detecting safety of a ventilator according to claim 5, wherein said S4: the method comprises the steps of preprocessing index data of a to-be-detected breathing machine, taking the preprocessed data as input of a breathing machine air leakage detection neural network model, and judging whether an air leakage potential safety hazard exists in the breathing machine or not by using an output result of the breathing machine air leakage detection neural network model, wherein the method comprises the following steps:
acquiring index data of a to-be-detected breathing machine, wherein the acquired index data comprise the flow rate of the breathing machine pipeline gas, the air leakage aperture, the inspiration tidal volume and the expiration tidal volume; and supplementing the training set data to the index data of the breathing machine to be detected to obtain a supplementary data set, carrying out data preprocessing on the index data of the breathing machine to be detected in the supplementary data set, and taking the index data after the data preprocessing as an input value of a breathing machine air leakage detection neural network model.
7. The method for detecting safety of a ventilator according to claim 6, wherein said S4: the method comprises the steps of preprocessing index data of a to-be-detected breathing machine, taking the preprocessed data as input of a breathing machine air leakage detection neural network model, and judging whether an air leakage potential safety hazard exists in the breathing machine or not by using an output result of the breathing machine air leakage detection neural network model, wherein the method comprises the following steps:
and taking the preprocessed data as input of a breathing machine air leakage detection neural network model, outputting the air leakage of the current detection breathing machine by the breathing machine air leakage detection neural network model, setting an air leakage threshold Q, and if the detected air leakage amount of the breathing machine is higher than the air leakage threshold Q, indicating that the current detection breathing machine has air leakage potential safety hazards.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111292272.XA CN114121250B (en) | 2021-11-03 | 2021-11-03 | Safety detection method for breathing machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111292272.XA CN114121250B (en) | 2021-11-03 | 2021-11-03 | Safety detection method for breathing machine |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114121250A CN114121250A (en) | 2022-03-01 |
CN114121250B true CN114121250B (en) | 2023-12-19 |
Family
ID=80380306
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111292272.XA Active CN114121250B (en) | 2021-11-03 | 2021-11-03 | Safety detection method for breathing machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114121250B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102114290A (en) * | 2009-12-31 | 2011-07-06 | 北京谊安医疗系统股份有限公司 | Method, device and system for detecting breathing machine |
CN108619597A (en) * | 2017-03-17 | 2018-10-09 | 深圳市大雅医疗技术有限公司 | A kind of pressure compensating method of lung ventilator runner |
KR20190094692A (en) * | 2018-02-05 | 2019-08-14 | 문경훈 | Method of estimating flow rate and of detecting leak of wide area water using recurrent analysis, recurrent neural network and deep neural network |
CN111275172A (en) * | 2020-01-21 | 2020-06-12 | 复旦大学 | Feedforward neural network structure searching method based on search space optimization |
CN111582453A (en) * | 2020-05-09 | 2020-08-25 | 北京百度网讯科技有限公司 | Method and device for generating neural network model |
-
2021
- 2021-11-03 CN CN202111292272.XA patent/CN114121250B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102114290A (en) * | 2009-12-31 | 2011-07-06 | 北京谊安医疗系统股份有限公司 | Method, device and system for detecting breathing machine |
CN108619597A (en) * | 2017-03-17 | 2018-10-09 | 深圳市大雅医疗技术有限公司 | A kind of pressure compensating method of lung ventilator runner |
KR20190094692A (en) * | 2018-02-05 | 2019-08-14 | 문경훈 | Method of estimating flow rate and of detecting leak of wide area water using recurrent analysis, recurrent neural network and deep neural network |
CN111275172A (en) * | 2020-01-21 | 2020-06-12 | 复旦大学 | Feedforward neural network structure searching method based on search space optimization |
CN111582453A (en) * | 2020-05-09 | 2020-08-25 | 北京百度网讯科技有限公司 | Method and device for generating neural network model |
Also Published As
Publication number | Publication date |
---|---|
CN114121250A (en) | 2022-03-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105868108B (en) | The unrelated binary code similarity detection method of instruction set based on neural network | |
CN105574098B (en) | The generation method and device of knowledge mapping, entity control methods and device | |
CN109036553A (en) | A kind of disease forecasting method based on automatic extraction Medical Technologist's knowledge | |
CN109543745A (en) | Feature learning method and image-recognizing method based on condition confrontation autoencoder network | |
CN106650650B (en) | Cross-age face recognition method | |
CN109741826B (en) | Anesthesia evaluation decision tree construction method and equipment | |
CN108509790A (en) | A kind of password strength assessment method based on group | |
CN108875907B (en) | Fingerprint identification method and device based on deep learning | |
CN110533549A (en) | One kind optimizing RS-BP neural network electric network fault localization method based on GA | |
US9345413B2 (en) | Heart rate extraction using neural wavelet adaptive gain control and neural pattern processing | |
CN105930447B (en) | A method of tree-like nested data is converted into panel data table | |
CN112765370B (en) | Entity alignment method and device of knowledge graph, computer equipment and storage medium | |
CN111428848A (en) | Molecular intelligent design method based on self-encoder and 3-order graph convolution | |
CN105031784B (en) | Alarm method of respirator and alarm system of respirator | |
CN106897265A (en) | Term vector training method and device | |
CN112148997B (en) | Training method and device for multi-modal countermeasure model for disaster event detection | |
CN110584654A (en) | Multi-mode convolutional neural network-based electrocardiosignal classification method | |
CN115346345B (en) | Intelligent toxic and harmful gas alarm system for preparing hexafluorobutadiene | |
CN114121250B (en) | Safety detection method for breathing machine | |
CN109508689B (en) | Face recognition method for strengthening confrontation | |
CN108536825A (en) | A method of whether identification source of houses data repeat | |
CN106644035B (en) | Vibration source identification method and system based on time-frequency transformation characteristics | |
Xu et al. | Hybrid deep-semantic matrix factorization for tag-aware personalized recommendation | |
CN103793747B (en) | A kind of sensitive information template construction method in network content security management | |
CN110537917A (en) | mechanical ventilation intelligent monitoring system and monitoring method based on respiratory mechanics |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |