Management method and system of high-flow humidification therapy device based on artificial intelligence

Through missing value sorting, local median value and Gaussian naive Bayes algorithm, isolated sample point filtering and Chebishev chaos map optimization, redundant samples, classification difficulties and local optimization in the high-flow humidification therapy instrument management model are solved. The problem improves the generalization ability and performance of the model.

CN117349732BActive Publication Date: 2024-11-05GANZHOU FIFTH PEOPLES HOSPITAL
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311489871.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-11-10
Publication Date
2024-11-05
Estimated Expiration
2043-11-10

AI Technical Summary

Technical Problem

The traditional high-flow humidification therapy instrument management model has problems such as missing value processing algorithms generating redundant samples, underutilizing sample distribution information, poor classification effect of balanced data set algorithms and easily falling into local optimal models.

Method used

Use missing value sorting method and local median value and Gaussian naive Bayes algorithm to fill in missing values, filter isolated sample points, calculate the cluster center of most classes, and generate subclusters. Use Chebischev Chaos Map and Elite Reverse Strategy Optimization Model Initial position, update position based on dynamic weighting factors and Levy flight strategy.

Benefits of technology

It improves data quality and model generalization capabilities, avoids overfitting, enhances data diversity and model performance, and avoids local optimal traps.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117349732B_ABST
    Figure CN117349732B_ABST
Patent Text Reader

Abstract

The present invention discloses a high-flow humidification therapy device management method and system based on artificial intelligence, and the method includes: data collection, missing value processing, balancing data sets, building a high-flow humidification therapy device management model and high-flow humidification therapy device management. The present invention belongs to the field of therapeutic device management technology, and specifically refers to a high-flow humidification therapy device management method and system based on artificial intelligence. This scheme uses local median and Gaussian naive Bayes algorithm to fill in missing values; filter isolated sample points, and determine the weight value of each sub-cluster and the number of new samples to be synthesized by calculating the distance between the majority class cluster center and the sub-cluster, and use random linear interpolation and nearest neighbor samples to generate new samples; use Chebyshev chaos map to generate the initial position and use elite reverse strategy for optimization, update the position based on dynamic weight factor, Levy flight strategy and mutation strategy, and complete the construction of the model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of therapeutic apparatus management, and specifically refers to a high-flow humidification therapeutic apparatus management method and system based on artificial intelligence. Background Art

[0002] High-flow humidification therapy device management refers to management based on the patient's treatment effect to ensure that the patient can receive effective treatment. However, the traditional missing value processing algorithm generates a large number of redundant samples and does not fully utilize the correlation between features in the sample distribution information, which brings noise to the data and generates overlapping data, resulting in model overfitting. The existing balanced data set algorithm is random sampling or only samples specific areas, which has the problem of poor classification effect and difficulty in classification. The traditional high-flow humidification therapy device management model is prone to fall into local optimality, which reduces model performance. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a high-flow humidification therapy device management method and system based on artificial intelligence. In view of the problem that the traditional missing value processing algorithm generates a large number of redundant samples and does not fully utilize the correlation between features in the sample distribution information, which brings noise to the data and generates overlapping data, resulting in model overfitting, this solution uses a missing value sorting method to ensure that features with a small number of missing values ​​are processed first, and uses the local median and Gaussian naive Bayes algorithm to fill in the missing values. Based on the binning and unbinning operations, the data quality is improved, the accuracy and efficiency of filling the missing values ​​are improved, the redundancy and overlap are reduced, the generalization ability of the model is improved, and the overfitting problem is avoided; the existing balanced data set algorithm is random sampling or only samples specific areas, which has the problems of poor classification effect and classification difficulty, This scheme adds the operation of filtering isolated sample points to improve the quality of data, calculates the majority class cluster center and uses the K-means algorithm to generate O clusters for the minority class sample set, so that the sub-clusters close to the majority class cluster center are assigned higher weight values ​​and more synthetic samples, avoiding the overlap between the newly synthesized samples and the majority class samples, ensuring that the newly synthesized samples better cover the distribution of the minority class sample set and improving the diversity of data; in view of the problem that the traditional high-flow humidification therapy device management model is prone to fall into the local optimum and reduce the model performance, this scheme uses the Chebyshev chaos map to generate the initial position and uses the elite reverse strategy for optimization to improve the quality of the initial position, and updates the position based on the dynamic weight factor, Levy flight strategy and mutation strategy to avoid the model from falling into the local optimum and improve the performance and efficiency of the model.

[0004] The technical solution adopted by the present invention is as follows: The high-flow humidification therapy device management method based on artificial intelligence provided by the present invention comprises the following steps:

[0005] Step S1: data collection, collecting data of high-flow humidification therapy device;

[0006] Step S2: missing value processing, sorting features based on the number of missing values, and using local median and Gaussian naive Bayes algorithm to fill missing values ​​for features with a small number of missing values;

[0007] Step S3: Balance the data set, filter out isolated sample points, calculate the majority class cluster center and generate sub-clusters for the minority class sample set, determine the weight value of each sub-cluster and the number of new samples to be synthesized by calculating the distance between the majority class cluster center and the sub-cluster, and generate new samples using random linear interpolation and nearest neighbor samples;

[0008] Step S4: construct a high-flow humidification therapy device management model, use the Chebyshev chaos map to generate the initial position and use the elite reverse strategy for optimization, update the position based on the dynamic weight factor, Levy flight strategy and mutation strategy, and complete the construction of the model;

[0009] Step S5: high-flow humidification therapy device management: managing the high-flow humidification therapy device based on the label output by the high-flow humidification therapy device management model.

[0010] Furthermore, in step S1, the data collection is to collect high-flow humidification therapy device data, including non-invasive positive pressure adjustment temperature, high-flow humidified oxygen therapy temperature, HFLOW flow, LFLOW flow, inspiratory positive pressure, expiratory positive pressure, ventilation mode, and output gas oxygen concentration, and the treatment effect level is used as a data classification label.

[0011] Furthermore, in step S2, the missing value processing specifically includes the following steps:

[0012] Step S21: Calculate the number of missing values ​​and sort them. Construct an original feature matrix based on the collected high-flow humidification therapy device data, calculate the number of missing values ​​for each feature of the original feature matrix, sort the features according to the number of missing values, and obtain the index value of the feature. The formula used is as follows:

[0013] sortindex=arg sort(A.isnull().sum());

[0014] Where A is the feature matrix, isnull() is used to determine whether each element is a missing value, sum() is used to calculate the number of missing values ​​for each feature, arg sort() is used to return the sorted index value, and sortindex is the sort index of the number of missing values ​​for each feature in the feature matrix;

[0015] Step S22: Fill missing values. Perform missing value filling processing on the feature column with the least number of missing values. The steps are as follows:

[0016] Step S221: feature processing, boxing and discretizing the original feature matrix to obtain N1 categories;

[0017] Step S222: constructing a feature training set and a feature prediction set, constructing a feature prediction set based on feature columns whose number of missing values ​​is 0, and constructing a feature training set based on feature columns whose number of missing values ​​is not 0;

[0018] Step S223: Process the feature training set, select the feature column with the least number of missing values ​​in the feature training set for processing, and fill the missing values ​​with the median of the feature column;

[0019] Step S224: training a Gaussian Naive Bayes model, using the processed feature training set and data classification labels to train the Gaussian Naive Bayes model;

[0020] Step S225: predict missing values, use the trained Gaussian naive Bayes model to predict the feature prediction set, and obtain the prediction result of the missing value;

[0021] Step S226: Update the original feature matrix, perform an unboxing operation on the prediction results to obtain the original continuous values, which are used to fill the missing values ​​of the feature column with the least number of missing values, and use the filled feature column to update the original feature matrix;

[0022] Step S23: After filling, repeat step S22 and select the next feature column with the least number of missing values ​​for processing until all missing values ​​are filled.

[0023] Furthermore, in step S3, the balancing of the data set specifically includes the following steps:

[0024] Step S31: Filter isolated sample points, construct the original unbalanced data set B, the minority class sample set F and the majority class sample set L based on the data after missing value processing, calculate the K nearest neighbor samples of all sample points in the minority class sample set F based on the KNN algorithm, and treat the sample points whose K nearest neighbor samples are all majority class samples as isolated sample points and delete them;

[0025] Step S32: Calculate the total number of samples that need to be inserted, and pre-set the balance rate M of the data set after sampling. The balance rate is the ratio of the number of minority class samples to the number of majority class samples in the data set. The formula used is as follows:

[0026] f'=M×l;

[0027] f add =f'-(ff 0 );

[0028] In the formula, f add is the total number of samples that need to be inserted, f' is the number of samples in the minority class sample set after sampling, f is the number of samples in the minority class sample set, and f 0 is the number of isolated sample points, l is the number of samples in the majority class sample set;

[0029] Step S33: Calculate the majority cluster center using the following formula:

[0030] ;

[0031] ;

[0032] In the formula, is the average value of the i-th feature of the majority class samples, i is the feature index, x ij is the i-th feature of the j-th sample in the majority class sample set L, j is the sample index, N2 is the number of features present in l samples in the majority class sample set L, is the majority class cluster center sample;

[0033] Step S34: Calculate the distance and use the K-means algorithm to generate O sub-clusters for the minority class sample set F. The cluster center of each sub-cluster is Q. b , calculate the cluster center Q of each subcluster based on the Euclidean distance algorithm b Cluster center with majority class The distance between them is calculated using the following formula:

[0034] ;

[0035] ;

[0036] In the formula, is the i-th feature of the b-th cluster center in the minority class, b is the cluster center index, g b is the bth cluster center in the minority class and the majority class cluster center The distance between

[0037] Step S35: Calculate the weight value of each sub-cluster using the following formula:

[0038] ;

[0039] ;

[0040] Where W b is the weight value of the b-th sub-cluster in the minority class;

[0041] Step S36: Calculate the new samples that need to be synthesized for each sub-cluster. The formula used is as follows:

[0042] h b =W b ×f add , b=1,2,…,O;

[0043] In the formula, h b is the number of new samples that need to be synthesized in the b-th sub-cluster;

[0044] Step S37: Generate a new sample, the steps are as follows:

[0045] Step S371: Set the initial value of the synthesized new sample and randomly select a sub-cluster Q b , from the sub-cluster Q b A sample u is randomly selected from the cluster, and the z nearest neighbor samples of the sample u are determined according to the KNN algorithm. The currently selected sub-cluster Q is pre-set. b The number of new samples synthesized is v b , v b The initial value of is 0;

[0046] Step S372: Select synthesis method. If z≥(h b -v b ), then go to step S3721 and select the first synthesis method to synthesize a new sample; otherwise, go to step S3722 and select the second synthesis method to synthesize a new sample. The steps are as follows:

[0047] Step S3721: The first synthesis method is to select the sample u and its (h b -v b ) nearest neighbor samples to generate a total of (h b -v b ) new samples, then the current sub-cluster Q b The synthesis of the new sample is completed, and the process goes to step S373. The formula used is as follows:

[0048] u new =u+rand(0,1)×(uu z1 );

[0049] In the formula, u new is a new sample generated, rand(0,1) generates a random number between 0 and 1, u z1 is the sample u (h b -v b ) is one of the nearest neighbor samples;

[0050] Step S3722: The second synthesis method generates a total of z new samples by random linear interpolation between the selected sample u and its z nearest neighbor samples, and updates v b The value of the currently selected subcluster Q b A sample w is randomly selected from the remaining samples in the formula, sample w≠sample u, and the process goes to step S372 to reselect the synthesis method;

[0051] Step S373: complete the generation of new samples, repeat steps S371-S372 until the generation of new samples of all sub-clusters is completed;

[0052] Step S38: construct a balanced data set, add the new samples generated by all sub-clusters to the original unbalanced data set B, and obtain a balanced data set with a balance rate of M.

[0053] Further, in step S4, the construction of the high-flow humidification therapy device management model specifically includes the following steps:

[0054] Step S41: construct a training data set and a test data set, randomly select 70% of the sample data from the balanced data set as the training data set, and the remaining 30% of the sample data as the test data set, and use model parameters to represent the position of the individual;

[0055] Step S42: Initialize the position, the steps are as follows:

[0056] Step S421: Generate individual initial positions using the Chebyshev chaos map. The formula used is as follows:

[0057] ;

[0058] In the formula, is the initial position of the ath individual in the dth dimension, LB d is the lower boundary of the dth dimension of the individual search space, UB d is the upper boundary of the dth dimension of the individual search space, c a-1 is the Chebyshev chaos value of the a-1th individual, c 0 is a random number in [0, 1], a is the individual index, cos() is the cosine function, d is the dimension index, N4 is the number of individuals, and D is the individual dimension;

[0059] Step S422: training a high-flow humidification therapy device management model, using Python to import the sklearn library to call the SVM function based on the current parameters, training the high-flow humidification therapy device management model based on the training data set, and using the performance of the high-flow humidification therapy device management model established based on the parameters as the individual fitness value;

[0060] Step S423: Optimize the initial position of the elite individuals, calculate the fitness values ​​of the individuals, arrange them in descending order, select the first N5 individuals as elite individuals, and use the elite reverse strategy to optimize the initial position of the elite individuals. The formula used is as follows:

[0061] ;

[0062] In the formula, is the initial position of the ath individual in the dth dimension after optimization, is the lower boundary of all individuals in the dth dimension after step S421, is the upper boundary of all individuals in the dth dimension after step S421, r1 is a random number in [0, 1], N5 is the number of elite individuals and N5 <N4;

[0063] Step S43: Calculate the optimal fitness value and the global optimal position, update the individual fitness value, and select the highest fitness value as the optimal fitness value E best , the optimal fitness value E best The corresponding individual position is taken as the global optimal position C best ;

[0064] Step S44: Location update, the steps are as follows:

[0065] Step S441: Update individual positions based on dynamic weight factors and randomly generate a random number r3 in [0, 1] for all individuals a , the formula used is as follows:

[0066] ;

[0067] ;

[0068] Where ω is the dynamic weight factor, t is the current number of iterations, T is the maximum number of iterations, RT is the safety threshold and RT∈[0.5, 1], μ is a random number in [0, 0.1], C a,d (t+1) is the position of the ath individual in the dth dimension at the t+1th iteration, C a,d (t) is the position of the ath individual in the dth dimension at the tth iteration, exp() is an exponential function with the natural constant e as the base, r2 is a random number in (0, 1], G is a random number that follows a normal distribution, and H is a 1×D row vector in which all elements are 1;

[0069] Step S442: Optimize the position based on the Levy flight strategy and mutation strategy, update the individual fitness value and the current global optimal position C best (t+1), the formula used is as follows:

[0070] ;

[0071] In the formula, is the position of the ath individual in the dth dimension at the t+1th iteration after optimization, σ is the mutation factor and σ∈[0,1];

[0072] Step S45: Model determination, pre-set fitness value evaluation threshold δ, update the optimal fitness value and the global optimal position, when the optimal fitness value E best When it is higher than the fitness value evaluation threshold δ, a high-flow humidification therapy device management model is constructed based on the current model parameters; otherwise, if the maximum number of iterations T is reached, go to step S42; otherwise, go to step S44.

[0073] Furthermore, in step S5, the high-flow humidification therapy device management is to collect real-time high-flow humidification therapy device data, input it into the high-flow humidification therapy device management model, and manage the high-flow humidification therapy device according to the label output by the high-flow humidification therapy device management model.

[0074] The artificial intelligence-based high-flow humidification therapy device management system provided by the present invention includes a data acquisition module, a missing value processing module, a data set balance module, a high-flow humidification therapy device management model building module and a high-flow humidification therapy device management module;

[0075] The data acquisition module acquires data of the high-flow humidification therapy device and sends the data to the missing value processing module;

[0076] The missing value processing module sorts the features based on the number of missing values, preferentially fills the missing values ​​for the features with a small number of missing values ​​using the local median and Gaussian naive Bayes algorithm, and sends the data to the balanced data set module;

[0077] The balanced data set module filters isolated sample points, calculates the majority class cluster center and generates sub-clusters for the minority class sample set, determines the weight value of each sub-cluster and the number of new samples to be synthesized by calculating the distance between the majority class cluster center and the sub-cluster, generates new samples using random linear interpolation and nearest neighbor samples, constructs a balanced data set, and sends the data to the module for constructing a high-flow humidification therapy device management model;

[0078] The module for constructing a high-flow humidification therapy device management model generates an initial position using a Chebyshev chaos map and optimizes using an elite reverse strategy, updates the position based on a dynamic weight factor, a Levy flight strategy, and a mutation strategy, completes the construction of the model, and sends the data to the high-flow humidification therapy device management module;

[0079] The high-flow humidification therapy device management module collects real-time high-flow humidification therapy device data, inputs it into the high-flow humidification therapy device management model, and manages the high-flow humidification therapy device according to the label output by the high-flow humidification therapy device management model.

[0080] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0081] (1) In view of the problem that traditional missing value processing algorithms generate a large number of redundant samples and fail to fully utilize the correlation between features in the sample distribution information, which brings noise to the data and produces overlapping data, leading to model overfitting, this solution uses a missing value sorting method to ensure that features with fewer missing values ​​are processed first, and uses the local median and Gaussian naive Bayes algorithms to fill in the missing values. It also uses binning and unbinning operations to improve the quality of the data, the accuracy and efficiency of filling missing values, reduce redundancy and overlap, improve the generalization ability of the model, and avoid overfitting.

[0082] (2) In view of the fact that the existing balanced data set algorithms are random sampling or only sampling in specific areas, which has the problems of poor classification effect and classification difficulty, this scheme adds the operation of filtering isolated sample points to improve the quality of the data, calculates the majority class cluster center and uses the K-means algorithm to generate O clusters for the minority class sample set. The sub-clusters close to the majority class cluster center are assigned higher weight values ​​and more synthetic samples, avoiding the overlap between the newly synthesized samples and the majority class samples, ensuring that the newly synthesized samples better cover the distribution of the minority class sample set, and improving the diversity of the data.

[0083] (3) In view of the problem that the traditional high-flow humidification therapy device management model is prone to fall into local optimality, which reduces the model performance, this scheme uses Chebyshev chaos map to generate the initial position and uses elite reverse strategy for optimization to improve the quality of the initial position. The position is updated based on dynamic weight factor, Levy flight strategy and mutation strategy to avoid the model from falling into local optimality and improve the performance and efficiency of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A schematic diagram of the process of the artificial intelligence-based high-flow humidification therapy device management method provided by the present invention;

[0085] Figure 2 A schematic diagram of an artificial intelligence-based high-flow humidification therapy device management system provided by the present invention;

[0086] Figure 3 is a schematic flow chart of step S2;

[0087] Figure 4 is a schematic diagram of the process of step S3;

[0088] Figure 5 It is a schematic diagram of the process of step S4.

[0089] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0090] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0091] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0092] Example 1, see Figure 1 The present invention provides a high-flow humidification therapy device management method based on artificial intelligence, which comprises the following steps:

[0093] Step S1: data collection, collecting data of high-flow humidification therapy device;

[0094] Step S2: missing value processing, sorting features based on the number of missing values, and using local median and Gaussian naive Bayes algorithm to fill missing values ​​for features with a small number of missing values;

[0095] Step S3: Balance the data set, filter out isolated sample points, calculate the majority class cluster center and generate sub-clusters for the minority class sample set, determine the weight value of each sub-cluster and the number of new samples to be synthesized by calculating the distance between the majority class cluster center and the sub-cluster, and generate new samples using random linear interpolation and nearest neighbor samples;

[0096] Step S4: construct a high-flow humidification therapy device management model, use the Chebyshev chaos map to generate the initial position and use the elite reverse strategy for optimization, update the position based on the dynamic weight factor, Levy flight strategy and mutation strategy, and complete the construction of the model;

[0097] Step S5: high-flow humidification therapy device management: managing the high-flow humidification therapy device based on the label output by the high-flow humidification therapy device management model.

[0098] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, data collection is to collect high-flow humidification therapy device data, including non-invasive positive pressure adjustment temperature, high-flow humidification oxygen therapy temperature, HFLOW flow, LFLOW flow, inspiratory positive pressure, expiratory positive pressure, ventilation mode, output gas oxygen concentration, and the treatment effect level is used as a data classification label.

[0099] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the missing value processing specifically includes the following steps:

[0100] Step S21: Calculate the number of missing values ​​and sort them. Construct an original feature matrix based on the collected high-flow humidification therapy device data, calculate the number of missing values ​​for each feature of the original feature matrix, sort the features according to the number of missing values, and obtain the index value of the feature. The formula used is as follows:

[0101] sortindex=arg sort(A.isnull().sum());

[0102] Where A is the feature matrix, isnull() is used to determine whether each element is a missing value, sum() is used to calculate the number of missing values ​​for each feature, arg sort() is used to return the sorted index value, and sortindex is the sort index of the number of missing values ​​for each feature in the feature matrix;

[0103] Step S22: Fill missing values. Perform missing value filling processing on the feature column with the least number of missing values. The steps are as follows:

[0104] Step S221: feature processing, boxing and discretizing the original feature matrix to obtain N1 categories;

[0105] Step S222: constructing a feature training set and a feature prediction set, constructing a feature prediction set based on feature columns whose number of missing values ​​is 0, and constructing a feature training set based on feature columns whose number of missing values ​​is not 0;

[0106] Step S223: Process the feature training set, select the feature column with the least number of missing values ​​in the feature training set for processing, and fill the missing values ​​with the median of the feature column;

[0107] Step S224: training a Gaussian Naive Bayes model, using the processed feature training set and data classification labels to train the Gaussian Naive Bayes model;

[0108] Step S225: predict missing values, use the trained Gaussian naive Bayes model to predict the feature prediction set, and obtain the prediction result of the missing value;

[0109] Step S226: Update the original feature matrix, perform an unboxing operation on the prediction results to obtain the original continuous values, which are used to fill the missing values ​​of the feature column with the least number of missing values, and use the filled feature column to update the original feature matrix;

[0110] Step S23: After filling, repeat step S22 and select the next feature column with the least number of missing values ​​for processing until all missing values ​​are filled.

[0111] By performing the above operations, the traditional missing value processing algorithm generates a large number of redundant samples and does not fully utilize the correlation between features in the sample distribution information, which brings noise to the data and generates overlapping data, resulting in model overfitting. This solution uses a missing value sorting method to ensure that features with fewer missing values ​​are processed first, and uses local median and Gaussian naive Bayes algorithms to fill in missing values. It also uses binning and unbinning operations to improve data quality, improve the accuracy and efficiency of filling missing values, reduce redundancy and overlap, improve the generalization ability of the model, and avoid overfitting problems.

[0112] Example 4, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S3, balancing the data set specifically includes the following steps:

[0113] Step S31: Filter isolated sample points, construct the original unbalanced data set B, the minority class sample set F and the majority class sample set L based on the data after missing value processing, calculate the K nearest neighbor samples of all sample points in the minority class sample set F based on the KNN algorithm, and treat the sample points whose K nearest neighbor samples are all majority class samples as isolated sample points and delete them;

[0114] Step S32: Calculate the total number of samples that need to be inserted, and pre-set the balance rate M of the data set after sampling. The balance rate is the ratio of the number of minority class samples to the number of majority class samples in the data set. The formula used is as follows:

[0115] f'=M×l;

[0116] f add =f'-(ff 0 );

[0117] In the formula, f add is the total number of samples that need to be inserted, f' is the number of samples in the minority class sample set after sampling, f is the number of samples in the minority class sample set, and f 0is the number of isolated sample points, l is the number of samples in the majority class sample set;

[0118] Step S33: Calculate the majority cluster center using the following formula:

[0119] ;

[0120] ;

[0121] In the formula, is the average value of the i-th feature of the majority class samples, i is the feature index, x ij is the i-th feature of the j-th sample in the majority class sample set L, j is the sample index, N2 is the number of features present in l samples in the majority class sample set L, is the majority class cluster center sample;

[0122] Step S34: Calculate the distance and use the K-means algorithm to generate O sub-clusters for the minority class sample set F. The cluster center of each sub-cluster is Q. b , calculate the cluster center Q of each subcluster based on the Euclidean distance algorithm b Cluster center with majority class The distance between them is calculated using the following formula:

[0123] ;

[0124] ;

[0125] In the formula, is the i-th feature of the b-th cluster center in the minority class, b is the cluster center index, g b is the bth cluster center in the minority class and the majority class cluster center The distance between

[0126] Step S35: Calculate the weight value of each sub-cluster using the following formula:

[0127] ;

[0128] ;

[0129] Where W b is the weight value of the b-th sub-cluster in the minority class;

[0130] Step S36: Calculate the new samples that need to be synthesized for each sub-cluster. The formula used is as follows:

[0131] h b =W b ×f add , b=1,2,…,O;

[0132] In the formula, h b is the number of new samples that need to be synthesized in the b-th sub-cluster;

[0133] Step S37: Generate a new sample, the steps are as follows:

[0134] Step S371: Set the initial value of the synthesized new sample and randomly select a sub-cluster Q b , from the sub-cluster Q b A sample u is randomly selected from the cluster, and the z nearest neighbor samples of the sample u are determined according to the KNN algorithm. The currently selected sub-cluster Q is pre-set. b The number of new samples synthesized is v b , v b The initial value of is 0;

[0135] Step S372: Select synthesis method. If z≥(h b -v b ), then go to step S3721 and select the first synthesis method to synthesize a new sample; otherwise, go to step S3722 and select the second synthesis method to synthesize a new sample. The steps are as follows:

[0136] Step S3721: The first synthesis method is to select the sample u and its (h b -v b ) nearest neighbor samples to generate a total of (h b -v b ) new samples, then the current sub-cluster Q b The synthesis of the new sample is completed, and the process goes to step S373. The formula used is as follows:

[0137] u new =u+rand(0,1)×(uu z1 );

[0138] In the formula, u new is a new sample generated, rand(0,1) generates a random number between 0 and 1, u z1 is the sample u (h b -v b ) is one of the nearest neighbor samples;

[0139] Step S3722: The second synthesis method generates a total of z new samples by random linear interpolation between the selected sample u and its z nearest neighbor samples, and updates v b The value of the currently selected subcluster Q b A sample w is randomly selected from the remaining samples in the formula, sample w≠sample u, and the process goes to step S372 to reselect the synthesis method;

[0140] Step S373: complete the generation of new samples, repeat steps S371-S372 until the generation of new samples of all sub-clusters is completed;

[0141] Step S38: construct a balanced data set, add the new samples generated by all sub-clusters to the original unbalanced data set B, and obtain a balanced data set with a balance rate of M.

[0142] By performing the above operations, the existing balanced data set algorithm is randomly sampled or only samples specific areas, which has the problems of poor classification effect and classification difficulty. This scheme adds the operation of filtering isolated sample points to improve the quality of the data, calculates the majority class cluster center and uses the K-means algorithm to generate O clusters for the minority class sample set, so that the sub-clusters close to the majority class cluster center are assigned higher weight values ​​and more synthetic samples, avoiding the overlap between the newly synthesized samples and the majority class samples, ensuring that the newly synthesized samples better cover the distribution of the minority class sample set, and improving the diversity of the data.

[0143] Example 5, see Figure 1 and Figure 5 This embodiment is based on the above embodiment. In step S4, building a high-flow humidification therapy device management model specifically includes the following steps:

[0144] Step S41: construct a training data set and a test data set, randomly select 70% of the sample data from the balanced data set as the training data set, and the remaining 30% of the sample data as the test data set, and use model parameters to represent the position of the individual;

[0145] Step S42: Initialize the position, the steps are as follows:

[0146] Step S421: Generate individual initial positions using the Chebyshev chaos map. The formula used is as follows:

[0147] ;

[0148] In the formula, is the initial position of the ath individual in the dth dimension, LB d is the lower boundary of the dth dimension of the individual search space, UB d is the upper boundary of the dth dimension of the individual search space, c a-1 is the Chebyshev chaos value of the a-1th individual, c 0 is a random number in [0, 1], a is the individual index, cos() is the cosine function, d is the dimension index, N4 is the number of individuals, and D is the individual dimension;

[0149] Step S422: training a high-flow humidification therapy device management model, using Python to import the sklearn library to call the SVM function based on the current parameters, training the high-flow humidification therapy device management model based on the training data set, and using the performance of the high-flow humidification therapy device management model established based on the parameters as the individual fitness value;

[0150] Step S423: Optimize the initial position of the elite individuals, calculate the fitness values ​​of the individuals, arrange them in descending order, select the first N5 individuals as elite individuals, and use the elite reverse strategy to optimize the initial position of the elite individuals. The formula used is as follows:

[0151] ;

[0152] In the formula, is the initial position of the ath individual in the dth dimension after optimization, is the lower boundary of all individuals in the dth dimension after step S421, is the upper boundary of all individuals in the dth dimension after step S421, r1 is a random number in [0, 1], N5 is the number of elite individuals and N5 <N4;

[0153] Step S43: Calculate the optimal fitness value and the global optimal position, update the individual fitness value, and select the highest fitness value as the optimal fitness value E best , the optimal fitness value E best The corresponding individual position is taken as the global optimal position C best ;

[0154] Step S44: Location update, the steps are as follows:

[0155] Step S441: Update individual positions based on dynamic weight factors and randomly generate a random number r3 in [0, 1] for all individuals a , the formula used is as follows:

[0156] ;

[0157] ;

[0158] Where ω is the dynamic weight factor, t is the current number of iterations, T is the maximum number of iterations, RT is the safety threshold and RT∈[0.5, 1], μ is a random number in [0, 0.1], C a,d (t+1) is the position of the ath individual in the dth dimension at the t+1th iteration, C a,d(t) is the position of the ath individual in the dth dimension at the tth iteration, exp() is an exponential function with the natural constant e as the base, r2 is a random number in (0, 1], G is a random number that follows a normal distribution, and H is a 1×D row vector in which all elements are 1;

[0159] Step S442: Optimize the position based on the Levy flight strategy and mutation strategy, update the individual fitness value and the current global optimal position C best (t+1), the formula used is as follows:

[0160] ;

[0161] In the formula, is the position of the ath individual in the dth dimension at the t+1th iteration after optimization, σ is the mutation factor and σ∈[0,1];

[0162] Step S45: Model determination, pre-set fitness value evaluation threshold δ, update the optimal fitness value and the global optimal position, when the optimal fitness value E best When it is higher than the fitness value evaluation threshold δ, a high-flow humidification therapy device management model is constructed based on the current model parameters; otherwise, if the maximum number of iterations T is reached, go to step S42; otherwise, go to step S44.

[0163] By performing the above operations, the traditional high-flow humidification therapy device management model is prone to fall into local optimality, which reduces the performance of the model. This solution uses the Chebyshev chaos map to generate the initial position and uses the elite reverse strategy for optimization to improve the quality of the initial position. The position is updated based on the dynamic weight factor, Levy flight strategy and mutation strategy to avoid the model from falling into local optimality and improve the performance and efficiency of the model.

[0164] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the high-flow humidification therapy device management is to collect real-time high-flow humidification therapy device data, input it into the high-flow humidification therapy device management model, and manage the high-flow humidification therapy device according to the label output by the high-flow humidification therapy device management model.

[0165] Embodiment 7, see Figure 2 This embodiment is based on the above embodiment. The artificial intelligence-based high-flow humidification therapy device management system provided by the present invention includes a data acquisition module, a missing value processing module, a balanced data set module, a high-flow humidification therapy device management model construction module and a high-flow humidification therapy device management module;

[0166] The data acquisition module acquires data of the high-flow humidification therapy device and sends the data to the missing value processing module;

[0167] The missing value processing module sorts the features based on the number of missing values, preferentially fills the missing values ​​for the features with a small number of missing values ​​using the local median and Gaussian naive Bayes algorithm, and sends the data to the balanced data set module;

[0168] The balanced data set module filters isolated sample points, calculates the majority class cluster center and generates sub-clusters for the minority class sample set, determines the weight value of each sub-cluster and the number of new samples to be synthesized by calculating the distance between the majority class cluster center and the sub-cluster, generates new samples using random linear interpolation and nearest neighbor samples, constructs a balanced data set, and sends the data to the module for constructing a high-flow humidification therapy device management model;

[0169] The module for constructing a high-flow humidification therapy device management model generates an initial position using a Chebyshev chaos map and optimizes using an elite reverse strategy, updates the position based on a dynamic weight factor, a Levy flight strategy, and a mutation strategy, completes the construction of the model, and sends the data to the high-flow humidification therapy device management module;

[0170] The high-flow humidification therapy device management module collects real-time high-flow humidification therapy device data, inputs it into the high-flow humidification therapy device management model, and manages the high-flow humidification therapy device according to the label output by the high-flow humidification therapy device management model.

[0171] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0172] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0173] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. Management method of high-flow humidification therapy device based on artificial intelligence, Features: The method comprises the following steps: Step S1: data collection, collecting data of high-flow humidification therapy device; Step S2: missing value processing, sorting features based on the number of missing values, and using local median and Gaussian naive Bayes algorithm to fill missing values ​​for features with a small number of missing values; Step S3: Balance the data set, filter out isolated sample points, calculate the majority class cluster center and generate sub-clusters for the minority class sample set, determine the weight value of each sub-cluster and the number of new samples to be synthesized by calculating the distance between the majority class cluster center and the sub-cluster, and generate new samples using random linear interpolation and nearest neighbor samples; Step S4: construct a high-flow humidification therapy device management model, use the Chebyshev chaos map to generate the initial position and use the elite reverse strategy for optimization, update the position based on the dynamic weight factor, Levy flight strategy and mutation strategy, and complete the construction of the model; Step S5: high-flow humidification therapy device management, managing the high-flow humidification therapy device based on the label output by the high-flow humidification therapy device management model; In step S4, the construction of the high-flow humidification therapy device management model specifically includes the following steps: Step S41: construct a training data set and a test data set, randomly select 70% of the sample data from the balanced data set as the training data set, and the remaining 30% of the sample data as the test data set, and use model parameters to represent the position of the individual; Step S42: Initialize the position, the steps are as follows: Step S421: Generate individual initial positions using the Chebyshev chaos map. The formula used is as follows: ; In the formula, is the initial position of the ath individual in the dth dimension, LB d is the lower boundary of the dth dimension of the individual search space, UB d is the upper boundary of the dth dimension of the individual search space, c a-1 is the Chebyshev chaos value of the a-1th individual, c 0 is a random number in [0, 1], a is the individual index, cos() is the cosine function, d is the dimension index, N4 is the number of individuals, and D is the individual dimension; Step S422: training a high-flow humidification therapy device management model, using Python to import the sklearn library to call the SVM function based on the current parameters, training the high-flow humidification therapy device management model based on the training data set, and using the performance of the high-flow humidification therapy device management model established based on the parameters as the individual fitness value; Step S423: Optimize the initial position of the elite individuals, calculate the fitness values ​​of the individuals, arrange them in descending order, select the first N5 individuals as elite individuals, and use the elite reverse strategy to optimize the initial position of the elite individuals. The formula used is as follows: ; In the formula, is the initial position of the ath individual in the dth dimension after optimization, is the lower boundary of all individuals in the dth dimension after step S421, is the upper boundary of all individuals in the dth dimension after step S421, r1 is a random number in [0, 1], N5 is the number of elite individuals and N5 <N4; Step S43: Calculate the optimal fitness value and the global optimal position, update the individual fitness value, and select the highest fitness value as the optimal fitness value E best , the optimal fitness value E best The corresponding individual position is taken as the global optimal position C best ; Step S44: Location update, the steps are as follows: Step S441: Update individual positions based on dynamic weight factors and randomly generate a random number r3 in [0, 1] for all individuals a , the formula used is as follows: ; ; Where ω is the dynamic weight factor, t is the current number of iterations, T is the maximum number of iterations, RT is the safety threshold and RT∈[0.5, 1], μ is a random number in [0, 0.1], C a,d (t+1) is the position of the ath individual in the dth dimension at the t+1th iteration, C a,d (t) is the position of the ath individual in the dth dimension at the tth iteration, exp() is an exponential function with the natural constant e as the base, r2 is a random number in (0, 1], G is a random number that follows a normal distribution, and H is a 1×D row vector in which all elements are 1; Step S442: Optimize the position based on the Levy flight strategy and mutation strategy, update the individual fitness value and the current global optimal position C best (t+1), the formula used is as follows: ; In the formula, is the position of the ath individual in the dth dimension at the t+1th iteration after optimization, σ is the mutation factor and σ∈[0,1]; Step S45: Model determination, pre-set fitness value evaluation threshold δ, update the optimal fitness value and the global optimal position, when the optimal fitness value E best When it is higher than the fitness value evaluation threshold δ, a high-flow humidification therapy device management model is constructed based on the current model parameters; otherwise, if the maximum number of iterations T is reached, go to step S42; otherwise, go to step S44; In step S3, the balancing of the data set specifically includes the following steps: Step S31: Filter isolated sample points, construct the original unbalanced data set B, the minority class sample set F and the majority class sample set L based on the data after missing value processing, calculate the K nearest neighbor samples of all sample points in the minority class sample set F based on the KNN algorithm, and treat the sample points whose K nearest neighbor samples are all majority class samples as isolated sample points and delete them; Step S32: Calculate the total number of samples that need to be inserted, and pre-set the balance rate M of the data set after sampling. The balance rate is the ratio of the number of minority class samples to the number of majority class samples in the data set. The formula used is as follows: f'=M×l; f add =f’-(f-f 0 ); In the formula, f add is the total number of samples that need to be inserted, f' is the number of samples in the minority class sample set after sampling, f is the number of samples in the minority class sample set, and f 0 is the number of isolated sample points, l is the number of samples in the majority class sample set; Step S33: Calculate the majority cluster center using the following formula: ; ; In the formula, is the average value of the i-th feature of the majority class samples, i is the feature index, x ij is the i-th feature of the j-th sample in the majority class sample set L, j is the sample index, N2 is the number of features present in l samples in the majority class sample set L, is the majority class cluster center sample; Step S34: Calculate the distance and use the K-means algorithm to generate O sub-clusters for the minority class sample set F. The cluster center of each sub-cluster is Q. b , calculate the cluster center Q of each subcluster based on the Euclidean distance algorithm b Cluster center with majority class The distance between them is calculated using the following formula: ; ; In the formula, is the i-th feature of the b-th cluster center in the minority class, b is the cluster center index, g b is the bth cluster center in the minority class and the majority class cluster center The distance between Step S35: Calculate the weight value of each sub-cluster using the following formula: ; ; Where W b is the weight value of the b-th sub-cluster in the minority class; Step S36: Calculate the new samples that need to be synthesized for each sub-cluster. The formula used is as follows: h b =W b ×f add ,b=1,2,…,O; In the formula, h b is the number of new samples that need to be synthesized in the b-th sub-cluster; Step S37: Generate a new sample, the steps are as follows: Step S371: Set the initial value of the synthesized new sample and randomly select a sub-cluster Q b , from the sub-cluster Q b A sample u is randomly selected from the cluster, and the z nearest neighbor samples of the sample u are determined according to the KNN algorithm. The currently selected sub-cluster Q is pre-set. b The number of new samples synthesized is v b , v b The initial value of is 0; Step S372: Select synthesis method. If z≥(h b -v b ), then go to step S3721 and select the first synthesis method to synthesize a new sample; otherwise, go to step S3722 and select the second synthesis method to synthesize a new sample. The steps are as follows: Step S3721: The first synthesis method is to select the sample u and its (h b -v b ) nearest neighbor samples to generate a total of (h b -v b ) new samples, then the current sub-cluster Q b The synthesis of the new sample is completed, and the process goes to step S373. The formula used is as follows: u new =u+rand(0,1)×(uu z1 ) In the formula, u new is a new sample generated, rand(0,1) generates a random number between 0 and 1, u z1 is the sample u (h b -v b ) is one of the nearest neighbor samples; Step S3722: The second synthesis method generates a total of z new samples by random linear interpolation between the selected sample u and its z nearest neighbor samples, and updates v b The value of the currently selected subcluster Q b A sample w is randomly selected again from the remaining samples in , sample w≠sample u, and the process goes to step S372 to reselect the synthesis method; Step S373: complete the generation of new samples, repeat steps S371-S372 until the generation of new samples of all sub-clusters is completed; Step S38: construct a balanced data set, add the new samples generated by all sub-clusters to the original unbalanced data set B, and obtain a balanced data set with a balance rate of M; In step S2, the missing value processing specifically includes the following steps: Step S21: Calculate the number of missing values ​​and sort them. Construct an original feature matrix based on the collected high-flow humidification therapy device data, calculate the number of missing values ​​for each feature of the original feature matrix, sort the features according to the number of missing values, and obtain the index value of the feature. The formula used is as follows: sortindex=arg sort(A.isnull().sum()); Where A is the feature matrix, isnull() is used to determine whether each element is a missing value, sum() is used to calculate the number of missing values ​​for each feature, arg sort() is used to return the sorted index value, and sortindex is the sort index of the number of missing values ​​for each feature in the feature matrix; Step S22: Fill missing values. Perform missing value filling processing on the feature column with the least number of missing values. The steps are as follows: Step S221: feature processing, boxing and discretizing the original feature matrix to obtain N1 categories; Step S222: constructing a feature training set and a feature prediction set, constructing a feature prediction set based on feature columns whose number of missing values ​​is 0, and constructing a feature training set based on feature columns whose number of missing values ​​is not 0; Step S223: Process the feature training set, select the feature column with the least number of missing values ​​in the feature training set for processing, and fill the missing values ​​with the median of the feature column; Step S224: training a Gaussian Naive Bayes model, using the processed feature training set and data classification labels to train the Gaussian Naive Bayes model; Step S225: predict missing values, use the trained Gaussian naive Bayes model to predict the feature prediction set, and obtain the prediction result of the missing value; Step S226: Update the original feature matrix, perform an unboxing operation on the prediction results to obtain the original continuous values, which are used to fill the missing values ​​of the feature column with the least number of missing values, and use the filled feature column to update the original feature matrix; Step S23: After filling, repeat step S22 and select the next feature column with the least number of missing values ​​for processing until all missing values ​​are filled.

2. The high-flow humidification therapy device management method based on artificial intelligence according to claim 1, Features: In step S1, the data collection is to collect high-flow humidification therapy device data, including non-invasive positive pressure adjustment temperature, high-flow humidified oxygen therapy temperature, HFLOW flow, LFLOW flow, inspiratory positive pressure, expiratory positive pressure, ventilation mode, output gas oxygen concentration, and the treatment effect level is used as a data classification label.

3. The artificial intelligence-based high-flow humidification therapy device management method according to claim 1, Features: In step S5, the high-flow humidification therapy device management is to collect real-time high-flow humidification therapy device data, input it into the high-flow humidification therapy device management model, and manage the high-flow humidification therapy device according to the label output by the high-flow humidification therapy device management model.

4. A high-flow humidification therapy device management system based on artificial intelligence, used to implement a high-flow humidification therapy device management method based on artificial intelligence as described in any one of claims 1 to 3, Features: It includes a data acquisition module, a missing value processing module, a data set balance module, a high-flow humidification therapy device management model building module and a high-flow humidification therapy device management module.

5. The artificial intelligence-based high-flow humidification therapy device management system according to claim 4, Features: The data acquisition module acquires data of the high-flow humidification therapy device and sends the data to the missing value processing module; The missing value processing module sorts the features based on the number of missing values, preferentially fills the missing values ​​for the features with a small number of missing values ​​using the local median and Gaussian naive Bayes algorithm, and sends the data to the balanced data set module; The balanced data set module filters isolated sample points, calculates the majority class cluster center and generates sub-clusters for the minority class sample set, determines the weight value of each sub-cluster and the number of new samples to be synthesized by calculating the distance between the majority class cluster center and the sub-cluster, generates new samples using random linear interpolation and nearest neighbor samples, constructs a balanced data set, and sends the data to the module for constructing a high-flow humidification therapy device management model; The module for constructing a high-flow humidification therapy device management model generates an initial position using a Chebyshev chaos map and optimizes using an elite reverse strategy, updates the position based on a dynamic weight factor, a Levy flight strategy, and a mutation strategy, completes the construction of the model, and sends the data to the high-flow humidification therapy device management module; The high-flow humidification therapy device management module collects real-time high-flow humidification therapy device data, inputs it into the high-flow humidification therapy device management model, and manages the high-flow humidification therapy device according to the label output by the high-flow humidification therapy device management model.