CN111310781A - Refrigeration equipment refrigerant leakage identification method based on PCA-ANN - Google Patents

Refrigeration equipment refrigerant leakage identification method based on PCA-ANN Download PDF

Info

Publication number
CN111310781A
CN111310781A CN201911397048.XA CN201911397048A CN111310781A CN 111310781 A CN111310781 A CN 111310781A CN 201911397048 A CN201911397048 A CN 201911397048A CN 111310781 A CN111310781 A CN 111310781A
Authority
CN
China
Prior art keywords
data
leakage
principal component
characteristic
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201911397048.XA
Other languages
Chinese (zh)
Inventor
巫江虹
于仙毅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201911397048.XA priority Critical patent/CN111310781A/en
Publication of CN111310781A publication Critical patent/CN111310781A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F25—REFRIGERATION OR COOLING; COMBINED HEATING AND REFRIGERATION SYSTEMS; HEAT PUMP SYSTEMS; MANUFACTURE OR STORAGE OF ICE; LIQUEFACTION SOLIDIFICATION OF GASES
    • F25B—REFRIGERATION MACHINES, PLANTS OR SYSTEMS; COMBINED HEATING AND REFRIGERATION SYSTEMS; HEAT PUMP SYSTEMS
    • F25B49/00—Arrangement or mounting of control or safety devices
    • F25B49/005—Arrangement or mounting of control or safety devices of safety devices
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • 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
    • G06N3/045—Combinations of networks
    • 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

Abstract

The invention discloses a refrigeration equipment refrigerant leakage identification method based on PCA-ANN, which comprises the following steps: 1. collecting operation data on refrigeration equipment, and acquiring original data simulating leakage faults, other faults and normal operation; 2. screening and removing abnormal, vacant and other invalid data, retaining steady-state valid data, and normalizing to establish a training sample data matrix containing all collected test characteristic quantities; 3. training sample data principal component analysis to obtain a plurality of principal component features with the variance contribution rate decreasing in sequence, and selecting the first n principal component features with the cumulative variance contribution rate reaching 85% as principal component features for leakage fault mode identification; 4. constructing and training a two-classification type neural network pattern recognition model capable of recognizing leakage and non-leakage by using the principal component characteristics as a neural network input layer; 5. real-time operation data of the refrigeration equipment is collected, and the data is input into an identification model to judge whether leakage occurs.

Description

Refrigeration equipment refrigerant leakage identification method based on PCA-ANN
Technical Field
The invention belongs to the field of mode identification and diagnosis of refrigeration equipment for refrigerant leakage, and particularly relates to a refrigerant leakage identification method based on PCA (principal component analysis) and ANN (neural network algorithm).
Background
The refrigeration equipment has rich and various use scenes and different running conditions, and is easy to cause equipment aging, control failure and equipment in a fault running state after long-term running. When the fault is not discharged in time, the refrigeration equipment deviates from the normal operation working condition, the system operation energy efficiency attenuation is reduced, the energy consumption waste is increased, the comfort is reduced, and the equipment system is easy to damage due to serious faults. Therefore, the research aiming at the fault diagnosis and detection of the refrigeration equipment is a powerful measure for ensuring the safe, efficient and continuous operation of the refrigeration equipment.
The refrigerant leakage fault is one of common faults of refrigeration equipment, and the refrigerant leakage is easily caused by the factors of non-standard installation operation, bad environment of a connecting pipe interface, long-term operation and the like. After the equipment leaks, the operation energy efficiency of the equipment is reduced, the leaked refrigerant is discharged to cause environmental protection problems such as greenhouse effect and the like, and the safety problem is also easily caused by the leakage of some flammable refrigerants (R290, R32 and the like) of the equipment. Refrigerant leaks can be classified into sudden faults and gradual faults, wherein the leakage time is short, the system operation influence and the phenomenon when the leakage occurs are rapidly distinguished in a short period, the leakage time is long, the influence on the system is long-term and fine, and the system has coupling performance between the operation characteristic change caused by the slow leakage of the refrigerant and the influence change of other faults, so that the fault type is difficult to accurately distinguish from a single characteristic.
For slow refrigerant leakage faults, the existing identification and diagnosis methods mainly comprise the following aspects: through experimental tests, the state parameter change of a leakage system of the refrigeration equipment is researched, and the parameter change is counted and analyzed; and fault diagnosis of the refrigeration equipment is carried out by combining a data mining method. In the above diagnostic method for the leakage fault, the data in the absence of the actual leakage condition is used for diagnostic identification of the slow leakage fault, and the state of the system after the refrigerant leakage occurs is simulated by the change of the refrigerant charge of the equipment. The refrigeration equipment is a nonlinear system, strong correlation exists among different parameters, and a simple system characteristic statistical analysis method judges that leakage faults have great limitation; the leak recognition characteristic parameters established using the experimental data of the refrigerant charge amount decrease also differ depending on the diagnostic model.
Disclosure of Invention
Aiming at the limitation of the existing leakage identification and diagnosis technical method, the invention provides a novel refrigerant leakage pattern identification method based on principal component analysis and neural network algorithm, thereby solving the problems of characteristic coupling of the existing statistical analysis method, leakage identification characteristic diversity of the data mining method and the like.
The invention is realized by at least one of the following technical schemes.
A refrigeration equipment refrigerant leakage identification method based on PCA-ANN comprises the following steps:
(1) collecting operation data on refrigeration equipment, and acquiring an original data matrix simulating leakage faults, other interference faults and normal operation;
(2) screening and removing abnormal and vacant invalid data, retaining steady-state valid data, and normalizing to establish a training sample data matrix containing all collected test characteristic quantities;
(3) training sample data principal component analysis to obtain a plurality of principal component characteristics with the variance contribution rate decreasing in sequence, and selecting the first n principal component characteristics with the cumulative variance contribution rate reaching 85% as principal component characteristics of leakage fault mode identification;
(4) constructing and training a two-classification type neural network pattern recognition model capable of recognizing leakage and non-leakage by using the principal component characteristics as a neural network input layer;
(5) real-time operation data of the refrigeration equipment are collected, principal component characteristics of the operation data are obtained through principal component characteristic load matrix calculation, and input into an identification model to judge whether leakage occurs or not.
Further, the step (1) comprises the following steps:
(1.1) carrying out an experiment for simulating a leakage fault working condition and other interference faults and normal working conditions on refrigeration equipment needing leakage identification and diagnosis, wherein the other interference faults comprise condenser blockage and evaporator fan blockage;
(1.2) continuously repeating the experiment to obtain parameters including system operation characteristic parameters, system performance and the like;
and (1.3) sorting data according to the temperature, the pressure, the flow, the electrical parameters, the characteristic energy efficiency, the temperature difference and the pressure difference in the test characteristics to obtain an original characteristic data matrix.
Further, the step (2) includes:
(2.1) screening abnormal data in the original characteristic data matrix, and rejecting unsteady state operation data, wherein the abnormal data comprises system shutdown state acquisition data, starting fluctuation state data, abnormal test of a test point and vacancy test state data;
and (2.2) normalizing the screened steady-state original data matrix by adopting standard deviation normalization, wherein the average number of characteristic quantities of the data matrix after standard normalization is 0, and the standard deviation is 1.
Further, the standard deviation normalization in the step (2.2) can convert the data of different characteristic parameters into pure quantities without units, half of the parameter values are smaller than 0, and the other half of the parameter values are larger than 0, and the influence of dimension units on the variables can be eliminated after the standard normalization.
Further, the step (3) comprises the following steps:
(3.1) carrying out standardization treatment on the screened original characteristic data matrix to obtain a pure quantity data matrix which is formed by n characteristic variables and is shown in the formula (1);
Figure BDA0002346592370000031
in the formula XnIs a feature vector, xNnSpecific parameter values in the feature vector are obtained, N is the feature number in the data matrix, and N is the principal component feature;
(3.2) calculating the covariance of the data matrix in the step (3.1) to obtain a covariance matrix, wherein the covariance calculation formula of the two vectors is shown as the formula (2), calculating the covariance of every two vectors of the data matrix to obtain a covariance matrix shown as the formula (3), and calculatingEigenvalues λ of covariance matrix1，λ2，λ3，……，λnThe feature vector corresponding to each feature value is shown as an equation (4);
Figure BDA0002346592370000032
in the formula Xa,XbRepresenting two different feature vectors, xiaIs a feature vector XaOf the ith parameter value, xjbFeature vector XbThe jth parameter value;
Figure BDA0002346592370000033
Figure BDA0002346592370000034
in the formula anAn eigenvector corresponding to the nth eigenvalue of the covariance matrix, annSpecific values of covariance matrix eigenvectors;
(3.3) arranging the eigenvalues from large to small, and arranging the eigenvectors according to the corresponding eigenvalue sorting mode;
and (3.4) selecting the first i principal elements with the cumulative contribution rate of 85% as principal element characteristics after characteristic extraction according to the variance contribution rate corresponding to the characteristic vector.
Further, the step (4) is specifically as follows:
(4.1) taking the first i principal component characteristics obtained in the step (3) as the input of a neural network model;
and (4.2) selecting a structural model of the neural network model, determining the number of hidden layers and the number of output layers of the neural network, and selecting 2 layers as the output layers to respectively represent leakage and non-leakage because the leakage identification is targeted to leakage faults.
Furthermore, the structure of the neural network model is mainly formed by combining a plurality of neuron models in the shape of y ═ f (wx + b), wherein w represents a neural network hierarchy structure formed by a plurality of neurons, and the neuron model is arranged in the neuron network hierarchy structureIn the structure, a connection weight is expressed, b represents a threshold, a functional expression of y ═ f (wx + b) represents an input/output structure of one neuron, f is called an activation function of the neuron, a typical activation function is a sigmoid function, and an expression is sigmoid (x) ═ 1/(1+ e)-x) And x is the input to the neuron activation function.
The deviation between model prediction and reality is continuously corrected among the multiple layers of neurons through mechanisms such as an inverse error propagation method and the like, so that a classification recognition mathematical model with high judgment precision can be formed. The neural network model has the biggest characteristic that partial parameters are input, the model with the error rate smaller than the threshold value is obtained through training, the function value for classification, prediction and other purposes can be output, and the input and output values among different neurons are used as black box models and do not need to be processed one by one. When the same type of parameter data is input again, the model can obtain the same output result with small error rate. The neural network model for leakage fault diagnosis uses principal component characteristics as input, and an output layer is a classification model of two output layers respectively representing leakage classes and non-leakage classes.
Further, the step (5) comprises:
(5.1) acquiring running state parameter data on the refrigeration equipment system to be identified and diagnosed in real time;
(5.2) preprocessing the acquired data in real time and carrying out normalization processing, and generating principal component characteristics for leakage identification when the load matrix is finished according to principal component analysis of the training sample;
and (5.3) inputting the principal component characteristic quantity into the trained neural network recognition model, judging the data type in real time, determining whether the system belongs to the leakage mode according to the step (4.2), and completing the leakage mode recognition diagnosis of the refrigeration equipment.
Compared with the prior art, the technology of the invention can obtain the following beneficial effects:
(1) the refrigerant leakage mode identification method based on principal component analysis and neural network algorithm provided by the invention adopts principal component analysis to perform dimensionality reduction processing on system characteristics, obtains principal component characteristics with higher correlation with leakage faults, and eliminates the coupling of the original system operation characteristics in different faults.
(2) The refrigerant leakage mode identification method based on principal component analysis and neural network algorithm provided by the invention adopts data of actual leakage working conditions to train and establish an identification model, and eliminates the deviation of leakage characteristics of the diagnosis leakage fault by adopting a data mining method based on other data.
(3) According to the refrigerant leakage pattern recognition method based on principal component analysis and neural network algorithm, the recognition model established by the neural network classification algorithm is adopted, the result with high recognition accuracy can be obtained, the input leakage characteristic parameters are only 4 principal component characteristics after principal component analysis, the required model is simple and efficient, and the actual application of refrigeration equipment is facilitated.
Drawings
FIG. 1 is a schematic flow chart of a PCA-ANN-based refrigerant leakage identification method for a refrigeration equipment in the embodiment;
FIG. 2 is a flow chart of leakage characteristic extraction in the principal component analysis method according to the present embodiment;
FIG. 3 is a diagram of a neural network model for leak pattern recognition according to an embodiment;
FIG. 4 is a result diagram of the principal component analysis method for extracting leakage characteristics in this embodiment;
FIG. 5 is a graph of the error rate variation of the neural network training iterative process of the present embodiment;
FIG. 6 is a gradient change diagram of the neural network iterative process of the embodiment;
FIG. 7 is a diagram of effective change inspection of the neural network iterative process of the embodiment;
FIG. 8 is a ROC graph of the neural network training process of the present embodiment;
FIG. 9 is a diagram of a neural network model training set recognition result of the leakage recognition in this embodiment;
FIG. 10 is a diagram of the neural network model validation set identification result of the leakage identification of the embodiment;
FIG. 11 is a diagram of a neural network model test set identification result of leakage identification in the present embodiment;
fig. 12 is a diagram of the neural network model overall data set identification result of the leakage identification of the embodiment.
Detailed Description
In order to make the objects and technical solutions of the present invention and their advantages more clear, the present invention is further described below with reference to the accompanying drawings and examples. It should be understood that the specific examples are intended to be illustrative only and are not intended to be limiting.
As shown in fig. 1, a PCA-ANN based refrigerant leak identification method for a refrigeration apparatus includes the following steps:
(1) collecting operation data on refrigeration equipment, and acquiring original data simulating leakage faults, other faults and normal operation; specifically, the method comprises the following steps:
(1.1) designing and carrying out experiments for simulating leakage fault working conditions and other interference faults and normal working conditions on refrigeration equipment needing leakage identification and diagnosis, wherein the interference faults can be condenser blockage, evaporator fan blockage and the like;
(1.2) repeating the experiment for many times to obtain system operation characteristic parameters as many as possible, and calculating calculation parameters such as refrigeration energy efficiency and the like according to the test parameters;
and (1.3) according to temperature, pressure, flow and electric parameters in the test characteristics and the integral data of calculated characteristic energy efficiency, temperature difference and pressure difference, obtaining an original characteristic data matrix.
(2) Invalid data such as abnormity, vacancy and the like are screened and removed, steady-state valid data are reserved, a training sample data matrix containing all collected test characteristic quantities is established in a normalized mode, and the method specifically comprises the following steps:
(2.1) carrying out data preprocessing on the original characteristic data matrix, namely screening abnormal data and removing unsteady state operation data, wherein the abnormal data comprises system shutdown state acquisition data, starting fluctuation state data and test point abnormal test or vacancy test state data;
(2.2) normalizing the preprocessed steady-state original data matrix, wherein the normalization method adopts standard deviation normalization, the average number of characteristic quantities of the data matrix after standard normalization is 0, and the standard deviation is 1;
(3) training sample data principal component analysis to obtain a plurality of principal component features with the variance contribution rate decreasing in sequence, and selecting the first n principal component features with the cumulative variance contribution rate of 85% as principal component features for leakage fault mode identification; the specific steps are shown in fig. 2, and the steps include:
(3.1) carrying out standardization treatment on the preprocessed original characteristic data matrix to obtain a pure number of data matrices; a pure number of data matrices consisting of n characteristic variables, as in formula (1);
Figure BDA0002346592370000061
in the formula XnIs a feature vector, xNnThe specific parameter values in the feature vector are shown, N is the feature number in the data matrix, and N is the principal component feature.
(3.2) calculating the covariance of the data matrix in the step (3.1) to obtain a covariance matrix, wherein the covariance calculation formula of the two vectors is shown as the formula (2), calculating the covariance of every two vectors of the data matrix to obtain a covariance matrix shown as the formula (3), and calculating the eigenvalue lambda of the covariance matrix1，λ2，λ3，……，λnThe feature vector corresponding to each feature value is shown as a formula (4);
Figure BDA0002346592370000062
in the formula Xa,XbRepresenting two different feature vectors, xiaIs a feature vector XaOf the ith parameter value, xjbFeature vector XbThe jth parameter value.
Figure BDA0002346592370000063
Figure BDA0002346592370000071
In the formula anAn eigenvector corresponding to the nth eigenvalue of the covariance matrix, annSpecific values of the covariance matrix eigenvectors.
(3.3) arranging the eigenvalues from large to small, and arranging the eigenvectors according to the corresponding eigenvalue sorting mode;
and (3.4) selecting the first i principal elements with the cumulative contribution rate of 85% as principal element characteristics after characteristic extraction according to the variance contribution rate corresponding to the characteristic vector.
(4) Constructing and training a two-classification type neural network pattern recognition model capable of recognizing leakage and non-leakage by using the principal component characteristics as a neural network input layer; the specific steps are shown in fig. 3, and the steps comprise:
(4.1) taking the first i principal component characteristics obtained in the step 3 as the input of a neural network model;
and (4.2) selecting a structural model of the neural network, determining the number of hidden layers and the number of output layers of the neural network, wherein the output layers can select 2 layers to represent leakage and non-leakage respectively because the leakage identification is targeted to leakage faults.
The neural network structure is formed by combining a plurality of neuron models (w represents a neural network hierarchical structure formed by a plurality of neurons, the neuron structure represents connection weights, b represents a threshold, a functional formula of y ═ f (wx + b) represents an input/output structure of one neuron, f is called an activation function of the neuron, a typical activation function is a sigmoid function, and an expression is sigmoid (x) 1/(1+ e)-x) And x is the input to the neuron activation function). The deviation between model prediction and reality is continuously corrected among the multiple layers of neurons through mechanisms such as an inverse error propagation method and the like, so that a classification recognition mathematical model with high judgment precision can be formed. The neural network model has the biggest characteristic that partial parameters are input, the model with the error rate smaller than the threshold value is obtained through training, the function value for classification, prediction and other purposes can be output, and the input and output values among different neurons are used as black box models and do not need to be processed one by one. When the same type of parameter data is input again, the model can obtain the same output result with small error rate. The neural network model for leakage fault diagnosis uses principal component characteristics as input, and an output layer is a classification model of two output layers respectively representing leakage class and non-leakage class.
(5) Collecting real-time operation data of the refrigeration equipment, calculating by using a principal component characteristic load matrix to obtain principal component characteristics of the operation data, and inputting an identification model to judge whether leakage occurs or not; specifically, the method comprises the following steps:
(5.1) acquiring running state parameter data on the refrigeration equipment system to be identified and diagnosed in real time;
(5.2) preprocessing the acquired data in real time and carrying out normalization processing, and generating principal component characteristics for leakage identification when the load matrix is finished according to principal component analysis of the training sample; the training sample comprises a screened and normalized data matrix containing the characteristics of test characteristic parameters, performance parameters and the like obtained in the steps (1) and (2).
And (5.3) inputting the principal component characteristic quantity into a neural network identification model, judging the data type in real time, determining whether the system belongs to a leakage mode, and finishing the identification and diagnosis of the leakage mode of the refrigeration equipment.
In the embodiment, an R134a heat pump hot water system is taken as a leakage diagnosis mode recognition object, the heating capacity of the system is about 1kw, R134a refrigerant is adopted, the charging amount is 170g, and a double-pipe heat exchanger and a micro-channel evaporator are respectively taken as a condenser and an evaporator of the heat pump system. Since the air source heat pump water heater is a typical refrigeration device, research on refrigerant leakage identification diagnosis on the device system has certain representativeness.
The refrigerant leakage mode identification method based on principal component analysis and neural network algorithm provided by the embodiment of the invention is specifically applied to the system as follows:
(1) the method comprises the steps of carrying out experiments on the refrigeration equipment for simulating other faults such as leakage fault, condenser dirt, evaporator dirt and the like and normal operation working conditions, testing to obtain original data, obtaining 11 groups of normal working conditions, 18 groups of interference working conditions and 12 groups of leakage working conditions. The experiment measures 22 temperatures, 8 pressures and 1 compressor power consumption, and 10 calculated quantities such as the degree of superheat and the degree of supercooling of the system are calculated according to the refrigeration cycle theory to obtain an original data matrix with the original characteristic quantity of 41.
(2) Screening and removing abnormal, vacant and other invalid data, retaining steady-state valid data, normalizing and establishing a training sample data matrix containing all collected test characteristic quantities to obtain a data matrix with the size of 41 × 3213;
(3) training sample data principal component analysis to obtain a plurality of principal component features with the variance contribution rate decreasing in sequence, and selecting the first n principal component features with the cumulative variance contribution rate of 85% as principal component features for leakage fault mode identification to obtain a leakage feature principal component analysis result shown in fig. 4.
From the result of fig. 4, it is known that the cumulative variance contribution rate of the first 4 principal elements reaches 87%, four principal elements with principal element numbers of 1, 2, 3 and 4 are selected as the leakage features after feature extraction, and compared with the original 41 features, the extracted principal element feature quantity is greatly reduced, which is beneficial to simplifying the leakage identification model.
(4) Selecting the 4 principal component features extracted in the step (3) as a neural network input layer, constructing and training a neural network pattern recognition model of a binary type capable of recognizing leakage and non-leakage, wherein a hidden layer is 10, an output layer is 2, and a training process and a training result are shown in fig. 5, 6, 7 and 8; after the neural network model sets structural parameters, the neural network model can be continuously iteratively optimized to a stable model with small error after inputting training set data, and only whether the final result meets the requirements or not needs to be seen. The best verify point in FIG. 5 is 0.13284 at iteration 31, and the gradient at iteration 37 in FIG. 6 is 0.021567.
(5) In order to verify the performance of a leakage recognition model, the operation data acquired on the air source heat pump system divides a 3213 preprocessed set of experimental data into three parts, wherein 70% of the data is used for training, 15% of the data is used for verifying, and 15% of the data is used for testing the final model. The test set comprises 482 groups of data, the principal component characteristic of the test set data is obtained by calculation of a principal component characteristic load matrix, the identification model is input to judge whether leakage occurs or not, the identification result is shown in fig. 9, fig. 10, fig. 11 and fig. 12, namely 153 groups which have leakage and are correctly identified exist in the confusion matrix of the test set, the number of groups which have leakage and are incorrectly identified is 18 groups, the specific identification performance is shown in table 1, namely, the identification accuracy of the leakage identification model in the test set is up to 89.2%, and the identification hit rate of the leakage working condition is 89.5%.
Table 1 test set data leak identification performance results
Figure BDA0002346592370000091
Wherein: acc: the accuracy rate represents the ratio of the correct classification number to the total sample number;
mcr is the misclassification rate which represents the ratio of misclassification samples to the total samples;
hit ratio, which represents the ratio of samples that occur and are correctly predicted to total occurring samples for a given class;
false alarm rate, which represents the ratio of samples that do not occur but are predicted to occur to the total number of samples that do not occur for a given class.
The above description is only an example of the present invention, and is not intended to limit the present invention in any way. Any modification, equivalent replacement, and improvement made by those skilled in the art within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.

Claims (8)

1. A refrigeration equipment refrigerant leakage identification method based on PCA-ANN, which is characterized by comprising the following steps:
(1) collecting operation data on refrigeration equipment, and acquiring an original data matrix simulating leakage faults, other interference faults and normal operation;
(2) screening and removing abnormal and vacant invalid data, retaining steady-state valid data, and normalizing to establish a training sample data matrix containing all collected test characteristic quantities;
(3) training sample data principal component analysis to obtain a plurality of principal component characteristics with the variance contribution rate decreasing in sequence, and selecting the first n principal component characteristics with the cumulative variance contribution rate reaching 85% as principal component characteristics of leakage fault mode identification;
(4) constructing and training a two-classification type neural network pattern recognition model capable of recognizing leakage and non-leakage by using the principal component characteristics as a neural network input layer;
(5) real-time operation data of the refrigeration equipment are collected, principal component characteristics of the operation data are obtained through principal component characteristic load matrix calculation, and input into an identification model to judge whether leakage occurs or not.
2. A PCA-ANN based refrigeration appliance refrigerant leak identification method as recited in claim 1 wherein said step (1) comprises the steps of:
(1.1) carrying out an experiment for simulating a leakage fault working condition and other interference faults and normal working conditions on refrigeration equipment needing leakage identification and diagnosis, wherein the other interference faults comprise condenser blockage and evaporator fan blockage;
(1.2) continuously repeating experiments to obtain system operation characteristic parameters and system performance parameters;
and (1.3) sorting data according to the temperature, the pressure, the flow, the electrical parameters, the characteristic energy efficiency, the temperature difference and the pressure difference in the test characteristics to obtain an original characteristic data matrix.
3. A PCA-ANN based refrigeration appliance refrigerant leak identification method as recited in claim 1 wherein said step (2) comprises:
(2.1) screening abnormal data in the original characteristic data matrix, and rejecting unsteady state operation data, wherein the abnormal data comprises system shutdown state acquisition data, starting fluctuation state data, abnormal test of a test point and vacancy test state data;
and (2.2) normalizing the screened steady-state original data matrix by adopting standard deviation normalization, wherein the average number of characteristic quantities of the data matrix after standard normalization is 0, and the standard deviation is 1.
4. A PCA-ANN based refrigeration apparatus refrigerant leak identification method as recited in claim 3 wherein the standard deviation normalization of step (2.2) transforms data of different characteristic parameters into pure quantities without units, half of the parameters having values less than 0 and the other half having values greater than 0, the standard normalization being effective to eliminate the influence of dimensional units on the variables.
5. A PCA-ANN based refrigeration appliance refrigerant leak identification method as recited in claim 1 wherein said step (3) comprises the steps of:
(3.1) carrying out standardization treatment on the screened original characteristic data matrix to obtain a pure quantity data matrix which is formed by n characteristic variables and is shown in the formula (1);
Figure FDA0002346592360000021
in the formula XnIs a feature vector, xNnSpecific parameter values in the feature vector are obtained, N is the feature number in the data matrix, and N is the principal component feature;
(3.2) calculating the covariance of the data matrix in the step (3.1) to obtain a covariance matrix, wherein the covariance calculation formula of the two vectors is shown as the formula (2), calculating the covariance of every two vectors of the data matrix to obtain a covariance matrix shown as the formula (3), and calculating the eigenvalue lambda of the covariance matrix1，λ2，λ3，……，λnThe feature vector corresponding to each feature value is shown as a formula (4);
Figure FDA0002346592360000022
in the formula Xa,XbRepresenting two different feature vectors, xiaIs a feature vector XaOf the ith parameter value, xjbFeature vector XbThe jth parameter value;
Figure FDA0002346592360000023
Figure FDA0002346592360000024
in the formula anAn eigenvector corresponding to the nth eigenvalue of the covariance matrix, annCharacteristic direction of covariance matrixA specific value of the amount;
(3.3) arranging the eigenvalues from large to small, and arranging the eigenvectors according to the corresponding eigenvalue sorting mode;
and (3.4) selecting the first i principal elements with the cumulative contribution rate of 85% as principal element characteristics after characteristic extraction according to the variance contribution rate corresponding to the characteristic vector.
6. A PCA-ANN based refrigeration appliance refrigerant leak identification method as recited in claim 1 wherein the step (4) is embodied as follows:
(4.1) taking the first i principal component characteristics obtained in the step (3) as the input of a neural network model;
and (4.2) selecting a structural model of the neural network model, determining the number of hidden layers and the number of output layers of the neural network, and selecting 2 layers as the output layers to respectively represent leakage and non-leakage because the leakage identification is targeted to leakage faults.
7. The PCA-ANN based refrigerant leakage identification method for the refrigeration equipment as claimed in claim 6, wherein the neural network model mainly comprises a combination of several neuron models shaped as y ═ f (wx + b), wherein w represents a neural network hierarchy structure composed of several neurons, connection weights are represented in the neuron structure, b represents a threshold, a functional formula of y ═ f (wx + b) represents an input-output structure of one neuron, f is called an activation function of the neuron, and a typical activation function is a sigmoid function expressed by sigmoid (x) ═ 1/(1+ e)-x) And x is the input to the neuron activation function.
8. A PCA-ANN based refrigeration appliance refrigerant leak identification method as recited in claim 1 wherein said step (5) comprises:
(5.1) acquiring running state parameter data on the refrigeration equipment system to be identified and diagnosed in real time;
(5.2) preprocessing the acquired data in real time and carrying out normalization processing, and generating principal component characteristics for leakage identification in real time according to the principal component analysis load matrix result of the training sample;
and (5.3) inputting the principal component characteristic quantity into the trained neural network recognition model, judging the data type in real time, determining whether the system belongs to the leakage mode according to the step (4.2), and completing the leakage mode recognition diagnosis of the refrigeration equipment.
CN201911397048.XA 2019-12-30 2019-12-30 Refrigeration equipment refrigerant leakage identification method based on PCA-ANN Pending CN111310781A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911397048.XA CN111310781A (en) 2019-12-30 2019-12-30 Refrigeration equipment refrigerant leakage identification method based on PCA-ANN

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911397048.XA CN111310781A (en) 2019-12-30 2019-12-30 Refrigeration equipment refrigerant leakage identification method based on PCA-ANN

Publications (1)

Publication Number Publication Date
CN111310781A true CN111310781A (en) 2020-06-19

Family

ID=71161455

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911397048.XA Pending CN111310781A (en) 2019-12-30 2019-12-30 Refrigeration equipment refrigerant leakage identification method based on PCA-ANN

Country Status (1)

Country Link
CN (1) CN111310781A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113834184A (en) * 2021-08-18 2021-12-24 青岛海尔空调器有限总公司 Control method and device for air conditioner and server

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105160416A (en) * 2015-07-31 2015-12-16 国家电网公司 Transformer area reasonable line loss prediction method based on principal component analysis and neural network
CN108898120A (en) * 2018-07-04 2018-11-27 四川泰立智汇科技有限公司 Water cooler method for diagnosing faults based on hybrid neural networks
CN109033632A (en) * 2018-07-26 2018-12-18 北京航空航天大学 A kind of trend forecasting method based on depth quantum nerve network
CN110543932A (en) * 2019-08-12 2019-12-06 珠海格力电器股份有限公司 air conditioner performance prediction method and device based on neural network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105160416A (en) * 2015-07-31 2015-12-16 国家电网公司 Transformer area reasonable line loss prediction method based on principal component analysis and neural network
CN108898120A (en) * 2018-07-04 2018-11-27 四川泰立智汇科技有限公司 Water cooler method for diagnosing faults based on hybrid neural networks
CN109033632A (en) * 2018-07-26 2018-12-18 北京航空航天大学 A kind of trend forecasting method based on depth quantum nerve network
CN110543932A (en) * 2019-08-12 2019-12-06 珠海格力电器股份有限公司 air conditioner performance prediction method and device based on neural network

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113834184A (en) * 2021-08-18 2021-12-24 青岛海尔空调器有限总公司 Control method and device for air conditioner and server

Similar Documents

Publication Publication Date Title
Li et al. An improved fault detection method for incipient centrifugal chiller faults using the PCA-R-SVDD algorithm
Li et al. A data-driven strategy for detection and diagnosis of building chiller faults using linear discriminant analysis
CN107505133B (en) The probability intelligent diagnosing method of rolling bearing fault based on adaptive M RVM
Li et al. A novel semi-supervised data-driven method for chiller fault diagnosis with unlabeled data
Wang et al. Enhanced chiller fault detection using Bayesian network and principal component analysis
Yan et al. ARX model based fault detection and diagnosis for chillers using support vector machines
Gao et al. Sensor drift fault diagnosis for chiller system using deep recurrent canonical correlation analysis and k-nearest neighbor classifier
Rossi et al. A statistical, rule-based fault detection and diagnostic method for vapor compression air conditioners
CN107844799B (en) Water chilling unit fault diagnosis method of integrated SVM (support vector machine) mechanism
Liu et al. A robust online refrigerant charge fault diagnosis strategy for VRF systems based on virtual sensor technique and PCA-EWMA method
CN104712542B (en) A kind of reciprocating compressor sensitive features based on Internet of Things are extracted and method for diagnosing faults
CN106528975B (en) A kind of prognostic and health management method applied to Circuits and Systems
Sun et al. A hybrid ICA-BPNN-based FDD strategy for refrigerant charge faults in variable refrigerant flow system
Zhou et al. A comparison study of basic data-driven fault diagnosis methods for variable refrigerant flow system
CN113792762A (en) Water chilling unit fault diagnosis method, system and medium based on Bayesian optimization LightGBM
Gao et al. Enhanced chiller faults detection and isolation method based on independent component analysis and k-nearest neighbors classifier
CN115187832A (en) Energy system fault diagnosis method based on deep learning and gram angular field image
Xia et al. Fault detection for centrifugal chillers using a Kernel Entropy Component Analysis (KECA) method
Li et al. Interpretation of convolutional neural network-based building HVAC fault diagnosis model using improved layer-wise relevance propagation
Gao et al. Comprehensive study on sensitive parameters for chiller fault diagnosis
Castro Performance evaluation of a reciprocating chiller using experimental data and model predictions for fault detection and diagnosis/Discussion
CN111310781A (en) Refrigeration equipment refrigerant leakage identification method based on PCA-ANN
CN114444663A (en) Water supply pipe network leakage detection and positioning method based on time convolution network
Reddy Application of a generic evaluation methodology to assess four different chiller FDD methods (RP-1275)
Chen et al. Generalizability of machine learning-based fault classification for residential air-conditioners

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