CN105135591A - Train air conditioning unit fault diagnosing method based on multi-classification strategy - Google Patents
Train air conditioning unit fault diagnosing method based on multi-classification strategy Download PDFInfo
- Publication number
- CN105135591A CN105135591A CN201510378576.6A CN201510378576A CN105135591A CN 105135591 A CN105135591 A CN 105135591A CN 201510378576 A CN201510378576 A CN 201510378576A CN 105135591 A CN105135591 A CN 105135591A
- Authority
- CN
- China
- Prior art keywords
- mrow
- mtd
- air conditioning
- conditioning unit
- train air
- 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.)
- Granted
Links
- 238000004378 air conditioning Methods 0.000 title claims abstract description 201
- 238000000034 method Methods 0.000 title claims abstract description 69
- 238000012549 training Methods 0.000 claims abstract description 78
- 238000003745 diagnosis Methods 0.000 claims description 78
- 238000012706 support-vector machine Methods 0.000 claims description 19
- 239000003507 refrigerant Substances 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 15
- 238000005057 refrigeration Methods 0.000 claims description 15
- 239000007789 gas Substances 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 14
- 239000011159 matrix material Substances 0.000 claims description 13
- 238000012163 sequencing technique Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000013145 classification model Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 5
- 230000001174 ascending effect Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000007405 data analysis Methods 0.000 claims description 3
- 238000009434 installation Methods 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 3
- 239000007788 liquid Substances 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 241001391944 Commicarpus scandens Species 0.000 description 1
- 125000002015 acyclic group Chemical group 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011541 reaction mixture Substances 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Air Conditioning Control Device (AREA)
Abstract
The invention discloses a train air conditioning unit fault diagnosing method based on a multi-classification strategy. The diagnosing method comprises the steps of firstly calculating the distribution densities of all categories according to actual normal operation data and various fault operation data of a train air conditioning unit; then ranking based on the junction distribution densities; and finally utilizing the ranking results to establish a binary classification strategy tree based on a PSVM, learning and training to obtain a fault diagnosing model of the train air conditioning unit and utilizing the fault diagnosing model of the train air conditioning unit to diagnose and determine fault types of the train air conditioning unit. The train air conditioning unit fault diagnosing method based on the multi-classification strategy can effectively improve the fault diagnosing accuracy.
Description
Technical Field
The invention belongs to the technical field of train air conditioning unit fault detection methods, and relates to a train air conditioning unit fault diagnosis method based on a multi-classification strategy.
Background
With the rapid development of passenger trains in modern society, train air conditioning units are widely applied to passenger trains. The modern passenger train has better sealing performance, directly causes poor air circulation performance in a carriage, and particularly has higher personnel density in the train at spring peak, so that the air conditioner of the train can work for a long time without failure. However, the train air conditioning unit operates under special conditions for a long time, such as: the air conditioning unit of the train is very easy to break down due to strong vibration when the train runs at high speed, large air state change outside the train and the like, so that the performance of the air conditioner is reduced. Therefore, the method has very important practical significance for realizing the accurate diagnosis of the faults of the train air conditioning unit.
With the high-speed development of computer technology, the faults of the train air-conditioning system can be well identified and diagnosed by combining with the unit data monitored in real time, and can be diagnosed in time and a solution can be given when the faults occur, so that the use efficiency of the train air-conditioning is greatly improved.
A Support Vector Machine (SVM) method proposed by researchers from the Rohao is used for fault diagnosis of train air conditioners, so that reference and guidance functions are provided for practical engineering application, certain judgment is made on feasibility of the support vector machine algorithm adopted in the fault diagnosis, however, a one-to-many multi-class classification strategy adopted in the method needs to be used for very careful adjustment of each sub-classifier, and otherwise the method is easy to be over-learned.
At present, methods for constructing an SVM multi-class classifier mainly include two classes: one is a direct method, which directly modifies an objective function, combines the parameter solution of a plurality of classification surfaces into an optimization problem, and realizes multi-class classification by solving the optimization problem in one step; the method seems to be simple, but has higher computational complexity and is difficult to realize, and is only suitable for small-scale problems. The other is indirect method, which mainly realizes the construction of a multi-classifier by combining a plurality of two classifiers, and the common methods are one-against-one and one-against-all.
a. One-to-many method (one-summary-rest, 1-v-rSVMs for short): during training, samples of a certain category are classified into one category, and other remaining samples are classified into another category, so that k SVM are constructed by k categories of samples; the classification classifies the unknown sample as the class having the largest classification function value.
b. One-to-one method (one-summary-one, abbreviated as 1-v-1 SVMs): the method is that an SVM is designed between any two types of samples, so that k (k-1)/2 SVM samples need to be designed; when an unknown sample is classified, the category with the most votes is the category of the unknown sample; the classification of classes in Libsvm is achieved according to this method.
c. Hierarchical support vector machines (H-SVMs): the hierarchical classification method firstly divides all classes into two subclasses, then further divides the subclasses into two secondary subclasses, and the process is circulated until a single class is obtained.
d. Other multiclass classification methods; in addition to the above methods, there are directed acyclic graphs SVMs (DAG-SVMs for short) and error correction coding SVMs that binary code a category.
Based on the knowledge, the train air conditioning unit fault diagnosis method based on the binary tree adjacent support vector machine multi-classification strategy of distribution density sorting can realize accurate diagnosis of the train air conditioning unit fault, and is suitable for the field of high-speed railway train air conditioning unit fault diagnosis.
Disclosure of Invention
The invention aims to provide a train air conditioning unit fault diagnosis method based on a multi-classification strategy, which can accurately diagnose the train air conditioning unit fault.
The technical scheme adopted by the invention is that the train air conditioning unit fault diagnosis method based on the multi-classification strategy is implemented according to the following steps:
step 1, determining a main fault type of a train air conditioning unit according to a working principle of a train air conditioning refrigeration system and common refrigeration air conditioning faults, and constructing a fault diagnosis model of the train air conditioning refrigeration system;
step 2, performing data acquisition and manual labeling according to the fault diagnosis model of the train air-conditioning refrigeration system constructed in the step 1, and acquiring a training sample set;
step 3, calculating to obtain the distribution density of each class by using the training sample set obtained by preprocessing in the step 2, arranging the classification order according to the sequence of the distribution density from light to heavy to obtain the most reasonable classification order, and constructing a binary classification strategy tree, namely a Huffman tree;
step 4, after the steps 1 to 3 are completed, selecting a kernel function;
step 5, constructing an optimal hyperplane of each internal node in the Huffman tree by using a near support vector machine algorithm PSVM as a two-class learning machine, namely constructing a train air conditioning unit fault diagnosis model based on a binary tree multi-class strategy of distribution density sorting;
and 6, diagnosing by using the train air conditioning unit fault diagnosis model obtained in the step 5 to finish the diagnosis of the train air conditioning unit fault type.
The invention is also characterized in that:
the main fault types of the train air conditioning unit in the step 1 are as follows:
refrigerant leakage, evaporator fouling, condenser fouling, non-condensable gases and compressor shutdown;
when different faults occur, the variation amplitude of the operation parameters of the train air conditioning unit is different, and a variation characteristic corresponding table of the judgment parameters corresponding to the normal and fault states of the train air conditioning unit to each fault is obtained through data analysis;
and selecting the characteristic vector Xi ═ compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity as the input of the fault diagnosis model of the train air-conditioning refrigerating system.
The step 2 is implemented according to the following steps:
step 2.1, arranging a plurality of acquisition measuring points on the train air conditioning unit, and acquiring data under the conditions of simulating five faults and normal operation conditions of refrigerant leakage, evaporator dirt, condenser dirt, non-condensable gas and compressor shutdown;
the acquired data mainly comprises 9 parameters including temperature parameters, pressure parameters and humidity parameters, and the parameters are as follows:
temperature parameters 4: the temperature of the dry bulb of the air at the inlet of the train air conditioning unit, the temperature of the dry bulb of the air at the outlet of the train air conditioning unit, the air suction temperature of a compressor and the exhaust temperature of the compressor;
the pressure parameters are 3: the air suction pressure of the compressor, the exhaust pressure of the compressor and the outlet pressure difference of the train air conditioning unit are measured;
humidity parameters 2: the relative humidity of air at the inlet of the train air conditioning unit and the relative humidity of air at the outlet of the train air conditioning unit;
step 2.2, listing the names, codes, sensor types and installation positions of all the acquisition measuring points set in the step 2.1, and acquiring the following data:
the suction pressure and the exhaust pressure of the compressor are respectively obtained by measuring points of the inlet pressure and the outlet pressure of the compressor;
the air suction temperature and the air discharge temperature of the compressor are respectively obtained by measuring points of the inlet temperature and the outlet temperature of the compressor;
the train air conditioning unit inlet air dry bulb temperature, the train air conditioning unit inlet air relative humidity, the train air conditioning unit outlet air dry bulb temperature and the train air conditioning unit outlet air relative humidity are respectively obtained by train air conditioning unit inlet air temperature, train air conditioning unit inlet air humidity, train air conditioning unit outlet air temperature and train air conditioning unit outlet air humidity measuring points;
the train air conditioning unit outlet pressure difference is obtained by an air duct pressure difference measuring point;
step 2.3, acquiring values of each measuring point of the train air conditioning unit under the five fault working conditions and the normal operation working condition based on the data acquisition scheme combined with the step 2.1 and the step 2.2, calculating an enthalpy value of inlet air of the train air conditioning unit according to the inlet air temperature of the train air conditioning unit and the inlet air humidity of the train air conditioning unit, and calculating an enthalpy value of outlet air according to the outlet air temperature of the train air conditioning unit and the outlet air humidity of the train air conditioning unit; and then, calculating the actual refrigerating capacity of the air conditioning unit by using the enthalpy values of the air inlet and the air outlet and the air volume, thereby obtaining sample data in the form of a characteristic vector Xi ═ { compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity }, and manually marking a fault type label Yi ═ {1,2, 3, 4, 5 and 6}, wherein the sample data respectively represent six types Yi ═ { normal operating condition, refrigerant leakage, evaporator fouling, condenser fouling, non-condensable gas and compressor shutdown }.
Step 3 is specifically implemented according to the following steps:
step 3.1, calculating to obtain the distribution density of each class by using the training sample set obtained by the pretreatment in the step 2;
the method is implemented according to the following steps:
step a, calculating the distribution volume of each class by the following two calculation methods: one is a super-cuboid volume; the other is a hypersphere volume;
for the volume of the super-rectangular body:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the volume of the hyper-cuboid containing all training samples in class S is:
for a hypersphere volume:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the hypersphere volume containing all training samples in class S is:
for super-rectangular bulk density:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the ultracuboid density of all training samples in class S is included:
for the hypersphere density:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen all training in class S is includedThe hypersphere density of the sample was:
combining the formula (1) to the formula (4) to obtain the ultra-rectangular density and the ultra-spherical density of each class;
step 3.2, after the step 3.1, performing ascending arrangement according to the super rectangular density or super spherical density of each class from small to large;
if two or more categories have the same ultra-rectangular density or ultra-spherical density, randomly arranging;
and 3.3, after the step 3.2, taking each class as a leaf node of the binary tree, and constructing the Huffman tree according to the super-rectangular body density or the super-spherical body density of each class.
Step 3.3 is specifically carried out according to the following method:
if m categories are set, the constructed Huffman tree has m leaf nodes, and the weight of each leaf node is the super-rectangular body density of the corresponding categoryOr density of hyper-spheresWherein S belongs to {1, 2.. eta., m };
for ultra-rectangular bulk densityThe construction rule of the huffman tree is as follows:
first, theThe forest is regarded as a forest with m trees, and each tree has only one node;
secondly, selecting and combining two trees with the minimum weight of root nodes in the forest as a left subtree and a right subtree of a new tree, wherein the weight of the root nodes of the new tree is the sum of the weights of the root nodes of the left subtree and the right subtree;
thirdly, deleting two selected trees from the forest and adding the new trees into the forest;
and fourthly, repeating the second step and the third step until only one tree is left in the forest, wherein the tree is the obtained Huffman tree.
The kernel function in step 4 is a gaussian kernel function, which is specifically as follows:
where x and z are input feature vectors.
Step 5 is specifically implemented according to the following steps:
step 5.1, starting from the current inner node, firstly, taking all sample classes in the left subtree as positive classes, taking the sample classes of the right subtree as negative classes, and obtaining the PSVM classifier of the current inner node by utilizing the learning and training of a PSVM algorithm:
firstly, a matrix A is constructed by utilizing training sampleskThe method comprises the following steps:
Ak＝{x1...xi...xN}′(5)；
next, a diagonal matrix D is constructed by using the corresponding category of each samplekThe diagonal elements are the current category corresponding to each sample, and specifically the following are:
order to <math>
<mrow>
<mi>I</mi>
<mo>=</mo>
<msub>
<mfenced open = '(' close = ')'>
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mrow>
<mi>d</mi>
<mo>×</mo>
<mi>d</mi>
</mrow>
</msub>
<mo>,</mo>
</mrow>
</math> Gk＝Dk[K(Ak,A′k)-e]Wherein e ═ 1, 1.. 1)'1×N，
The following algorithm is obtained:
wherein v ∈ R;
finally, the obtained classification model is concretely as follows:
yk＝sign((K(x,A′k)K(Ak,A′k)+e′)Dkvk)(8)；
wherein, <math>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>g</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>{</mo>
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>x</mi>
<mo><</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mrow>
<mi>x</mi>
<mo>≥</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>;</mo>
</mrow>
</mrow>
</math>
step 5.2, when the left subtree is a leaf node, the classifier of the current internal node is the classifier of the leaf node, and all training samples of the class corresponding to the leaf node are removed from the sample set;
if the left subtree is an inner node, returning to the step 5.1;
step 5.3, when the right subtree is a leaf node, the classifier of the current inner node is the classifier of the leaf node, and all training samples of the class corresponding to the leaf node are removed from the sample set;
otherwise, the right subtree is an inner node, and the step 5.1 is returned to;
through the steps 5.1 to 5.3, a plurality of classifiers can be obtained, and the plurality of classifiers form a train air conditioning unit fault diagnosis model based on a binary tree multi-classification strategy of distribution density.
Step 6 is implemented according to the following steps:
diagnosing the running state of the train air conditioner by using the train air conditioner set fault diagnosis model based on the binary tree multi-classification strategy of the distribution density, which is obtained in the step 5, wherein the diagnosis state comprises the following steps: the method comprises the following steps of (1) carrying out normal operation working conditions, refrigerant leakage, evaporator dirt, condenser dirt, non-condensable gas and compressor shutdown faults according to the following specific method:
6.1, calculating an enthalpy value of inlet air of the train air conditioning unit according to the temperature of the inlet air of the train air conditioning unit and the humidity of the inlet air of the train air conditioning unit by using the values of the measuring points of the train air conditioning unit under six working conditions, which are acquired in the step 2;
calculating an enthalpy value of outlet air of the train air conditioning unit according to outlet air temperature of the train air conditioning unit and outlet air humidity of the train air conditioning unit, and further calculating actual refrigerating capacity of the air conditioning unit according to the inlet and outlet enthalpy values and air volume of the air, so as to obtain data in the form of a characteristic vector x ═ compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity;
step 6.2, taking the data obtained in the step 6.1 as an input matrix of a train air conditioning unit fault diagnosis model based on a binary tree multi-classification strategy of distribution density sequencing:
firstly, classifying by using a classifier corresponding to a root node;
then, when the classification result is 1, classifying by using a classifier of the left sub-tree, and if the classification result is-1, classifying by using a classifier of the right sub-tree until the left sub-tree is a leaf node or the right sub-tree is a leaf node;
finally, the category corresponding to the leaf node is the current diagnosis result, and the diagnosis of the train air conditioning unit fault is completed.
The invention has the beneficial effects that:
(1) the train air conditioning unit fault diagnosis method based on the multi-classification strategy firstly calculates the distribution density of each class according to the distribution characteristics of the actual normal operation data and the multiple fault operation data of the train air conditioning unit, then carries out sequencing based on the silent junction distribution density, finally constructs a two-branch classification strategy tree based on PSVM (particle swarm optimization) by utilizing the sequencing result, and obtains a train air conditioning unit fault diagnosis model through learning and training;
(2) the train air conditioning unit fault diagnosis method based on the multi-classification strategy can effectively improve the precision of train air conditioning unit fault diagnosis and bring great convenience to maintenance and repair agents of train air conditioners.
(3) The train air conditioning unit fault diagnosis method based on the multi-classification strategy adopts a train air conditioning unit fault diagnosis method based on a binary tree adjacent support vector machine multi-classification strategy of distribution density sorting, and is very suitable for the field of train air conditioning unit fault diagnosis of high-speed railways.
Drawings
FIG. 1 is a schematic view of the operating principle of the refrigeration cycle of the air conditioning unit of the train;
FIG. 2 is a diagram of a hypersphere density Huffman tree involved in the train air conditioning unit fault diagnosis method based on a multi-classification strategy.
Detailed Description
The invention relates to a train air conditioning unit fault diagnosis method based on a multi-classification strategy, which is implemented according to the following steps:
the method comprises the following steps of 1, determining main fault types of a train air conditioning unit according to the working principle of a train air conditioning and refrigerating system and common refrigerating and air conditioning faults, and constructing a fault diagnosis model of the train air conditioning and refrigerating system, wherein the method specifically comprises the following steps:
constructing a train air-conditioning refrigeration system model based on a structure of a train air-conditioning unit;
the existing train air conditioning unit mainly comprises an evaporator, a condenser, an expansion valve and a compressor which are connected through pipelines to form a closed refrigeration system;
the working principle of the existing train air-conditioning refrigeration system is as follows: the liquid refrigerant is generated by the continuous circulation of the refrigerating system loop, is evaporated in the evaporator, exchanges heat with cooled air, absorbs the heat of the cooled air and is vaporized into vapor, and then the compressor extracts the generated vapor from the evaporator and compresses the refrigerant to be discharged under high pressure; the compressed high-temperature and high-pressure vapor is cooled by surrounding air in the condenser and condensed into high-pressure liquid, the high-pressure liquid is throttled by a thermal expansion valve, the throttled low-pressure and low-temperature wet vapor enters the evaporator and is vaporized again to absorb the heat of the cooled air, and the process is repeated, as shown in fig. 1.
Based on the working principle of the train air-conditioning refrigeration system and the existing fault occurrence condition, the main faults of the train air-conditioning can be obtained as follows: refrigerant leakage, evaporator fouling, condenser fouling, non-condensable gases and compressor shutdown; these five typical faults will be selected as the study object.
When different faults occur, the variation amplitude of the operation parameters of the train air conditioning unit is different, and a variation characteristic corresponding table of the judgment parameters corresponding to the normal and fault states of the train air conditioning unit to each fault is obtained through data analysis, wherein the variation characteristic corresponding table is shown in table 1; according to table 1, in the train air conditioning unit fault diagnosis method based on the classification strategy, the characteristic vector Xi ═ compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity is selected as the input of the train air conditioning refrigeration system fault diagnosis model.
TABLE 1 Fault determination parameter variation rule Table
The symbols in table 1 indicate: denotes that the parameter is essentially unchanged; + indicates a significant increase in the parameter, with a large variation; -indicates a significant decrease in the parameter, a great change; + indicates a slight increase in the parameter, with little change; the expression parameter is slightly reduced and does not vary much.
Step 2, performing data acquisition and manual labeling according to the fault diagnosis model of the train air-conditioning refrigeration system constructed in the step 1, and acquiring a training sample set, wherein the method is implemented according to the following steps:
step 2.1, arranging a plurality of acquisition measuring points on the train air conditioning unit, and acquiring data under the conditions of simulating five faults and normal operation conditions of refrigerant leakage, evaporator dirt, condenser dirt, non-condensable gas and compressor shutdown;
the acquired data mainly comprises 9 parameters including temperature parameters, pressure parameters and humidity parameters, and the parameters are as follows:
(1) temperature parameters 4: the temperature of the dry bulb of the air at the inlet of the train air conditioning unit, the temperature of the dry bulb of the air at the outlet of the train air conditioning unit, the air suction temperature of a compressor and the exhaust temperature of the compressor;
(2) the pressure parameters are 3: the air suction pressure of the compressor, the exhaust pressure of the compressor and the outlet pressure difference of the train air conditioning unit are measured;
(3) humidity parameters 2: the relative humidity of air at the inlet of the train air conditioning unit and the relative humidity of air at the outlet of the train air conditioning unit;
step 2.2, listing the names, code numbers, sensor types and installation positions of all the acquisition measuring points set in the step 2.1, wherein the list is specifically shown in a table 2;
table 2 test point information
The suction pressure and the exhaust pressure of the compressor are respectively obtained by measuring points of the inlet pressure and the outlet pressure of the compressor;
the air suction temperature and the air discharge temperature of the compressor are respectively obtained by measuring points of the inlet temperature and the outlet temperature of the compressor;
the train air conditioning unit inlet air dry bulb temperature, the train air conditioning unit inlet air relative humidity, the train air conditioning unit outlet air dry bulb temperature and the train air conditioning unit outlet air relative humidity are respectively obtained by train air conditioning unit inlet air temperature, train air conditioning unit inlet air humidity, train air conditioning unit outlet air temperature and train air conditioning unit outlet air humidity measuring points;
the train air conditioning unit outlet pressure difference is obtained by an air duct pressure difference measuring point;
step 2.3, based on the data acquisition scheme combined with the step 2.1 and the step 2.2, acquiring values of each measuring point of the train air conditioning unit under the five faults (refrigerant leakage, evaporator dirt, condenser dirt, non-condensable gas and compressor stop) and normal operation conditions, calculating an enthalpy value of inlet air of the train air conditioning unit according to the inlet air temperature of the train air conditioning unit and the inlet air humidity of the train air conditioning unit, and calculating an enthalpy value of outlet air according to the outlet air temperature of the train air conditioning unit and the outlet air humidity of the train air conditioning unit; and then, calculating the actual refrigerating capacity of the air conditioning unit by using the enthalpy values of the air inlet and the air outlet and the air volume, thereby obtaining sample data in the form of a characteristic vector Xi ═ { compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity }, and manually marking a fault type label Yi ═ {1,2, 3, 4, 5 and 6}, wherein the sample data respectively represent six types Yi ═ { normal operating condition, refrigerant leakage, evaporator fouling, condenser fouling, non-condensable gas and compressor shutdown }.
Step 3, utilizing the training sample set obtained by preprocessing in the step 2, obtaining the distribution density of each class through calculation, arranging the classification order according to the sequence of the distribution density from light to heavy to obtain the most reasonable classification order, and constructing a binary classification strategy tree, namely a Huffman tree, and specifically implementing according to the following steps:
step 3.1, obtaining the distribution density of each class by calculating the training sample set obtained by preprocessing in the step 2, and specifically implementing according to the following steps:
step a, calculating the distribution volume of each class by the following two calculation methods: one is a super-cuboid volume; the other is a hypersphere volume;
for the volume of the super-rectangular body:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the volume of the hyper-cuboid containing all training samples in class S is:
for a hypersphere volume:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the hypersphere volume containing all training samples in class S is:
for super-rectangular bulk density:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the ultracuboid density of all training samples in class S is included:
for the hypersphere density:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the hypersphere density of all training samples in class S is:
by combining the formulas (1) to (4), the ultra-rectangular density and the ultra-spherical density of each class can be obtained;
step 3.2, after the step 3.1, performing ascending arrangement according to the super rectangular density or super spherical density of each class from small to large;
if two or more categories have the same ultra-rectangular density or ultra-spherical density, randomly arranging;
such as: the resulting permutation was: n is1,n2,...,nmWhere is the total number of m classes, then niE {1, 2.., m } is a class label;
step 3.3, after the step 3.2, taking each class as a leaf node of the binary tree, and constructing a Huffman tree according to the super-rectangular body density or the super-spherical body density of each class;
if m categories are set, the constructed Huffman tree has m leaf nodes, and the weight of each leaf node is the super-rectangular body density of the corresponding categoryOr density of hyper-spheresWherein S belongs to {1, 2.. eta., m };
in ultra-rectangular bulk densityFor example, the construction rule of the huffman tree is as follows:
(1) will be provided withA forest viewed as having m trees (each tree has only one node);
(2) selecting and combining two trees with the smallest weight of root nodes in a forest as a left subtree and a right subtree of a new tree, wherein the weight of the root nodes of the new tree is the sum of the weights of the root nodes of the left subtree and the right subtree;
(3) deleting two selected trees from the forest, and adding the new trees into the forest;
(4) repeating the steps (2) and (3) until only one tree is left in the forest, wherein the tree is the obtained Huffman tree.
And 4, after the steps 1 to 3 are completed, selecting a kernel function:
the invention relates to a multi-classification strategy-based train air conditioning unit fault diagnosis method, which relates to a core method, wherein the core method is an effective method for solving the problem of nonlinear fault diagnosis, and the core method is used for the linear inseparable condition: mapping an input space X to a high-dimensional feature space F by using a nonlinear transformation phi, then changing the input space X into a linear separable condition in the feature space, and finally classifying the input space X by using a linear classification model, so that the performance of a classifier is improved by selecting a proper kernel function;
the train air conditioning unit fault diagnosis method based on the multi-classification strategy selects the Gaussian kernel function:where x and z are input feature vectors.
Step 5, constructing an optimal hyperplane of each internal node in the Huffman tree by using a proximity support vector machine algorithm PSVM as a two-class learning machine, namely constructing a train air conditioning unit fault diagnosis model based on a binary tree multi-class strategy of distribution density sequencing, and specifically implementing the following steps:
step 5.1, starting from the current inner node, firstly, taking all sample classes in the left subtree as positive classes, taking the sample classes of the right subtree as negative classes, and obtaining the PSVM classifier of the current inner node by utilizing the learning training of a PSVM algorithm, wherein the specific method comprises the following steps:
firstly, a matrix A is constructed by utilizing training sampleskThe method comprises the following steps:
Ak＝{x1...xi...xN}′(5)；
next, a diagonal matrix D is constructed by using the corresponding category of each samplekThe diagonal elements are the current category corresponding to each sample, and specifically the following are:
order to <math>
<mrow>
<mi>I</mi>
<mo>=</mo>
<msub>
<mfenced open = '(' close = ')'>
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mrow>
<mi>d</mi>
<mo>×</mo>
<mi>d</mi>
</mrow>
</msub>
<mo>,</mo>
</mrow>
</math> Gk＝Dk[K(Ak,A′k)-e]Wherein e ═ 1, 1.. 1)'1×N，
The following algorithm is obtained:
in the formula (7), v ∈ R;
finally, the obtained classification model is concretely as follows:
yk＝sign((K(x,A′k)K(Ak,A′k)+e′)Dkvk)(8)；
in the formula (8), the reaction mixture is, <math>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>g</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>{</mo>
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>x</mi>
<mo><</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mrow>
<mi>x</mi>
<mo>≥</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>;</mo>
</mrow>
</mrow>
</math>
step 5.2, when the left subtree is a leaf node, the classifier of the current internal node is the classifier of the leaf node, and all training samples of the class corresponding to the leaf node are removed from the sample set;
if the left subtree is an inner node, returning to the step 5.1;
step 5.3, when the right subtree is a leaf node, the classifier of the current inner node is the classifier of the leaf node, and all training samples of the class corresponding to the leaf node are removed from the sample set;
otherwise, the right subtree is an inner node, and the step 5.1 is returned to;
through the steps 5.1 to 5.3, a plurality of classifiers can be obtained, and the plurality of classifiers form a train air conditioning unit fault diagnosis model based on a binary tree multi-classification strategy of distribution density.
Step 6, diagnosing by using the train air conditioning unit fault diagnosis model obtained in the step 5 to finish the diagnosis of the train air conditioning unit fault type;
diagnosing the running state of the train air conditioner by using the train air conditioner unit fault diagnosis model based on the binary tree multi-classification strategy of the distribution density, which is obtained in the step 5, wherein the diagnosis state comprises the following steps: the method comprises the following steps of (1) carrying out normal operation working conditions, refrigerant leakage, evaporator dirt, condenser dirt, non-condensable gas and compressor shutdown faults according to the following specific method:
6.1, calculating an enthalpy value of inlet air of the train air conditioning unit according to the temperature of the inlet air of the train air conditioning unit and the humidity of the inlet air of the train air conditioning unit by using the values of the measuring points of the train air conditioning unit under six working conditions, which are acquired in the step 2; calculating an enthalpy value of outlet air of the train air conditioning unit according to outlet air temperature of the train air conditioning unit and outlet air humidity of the train air conditioning unit, and further calculating actual refrigerating capacity of the air conditioning unit according to the inlet and outlet enthalpy values and air volume of the air, so as to obtain data in the form of a characteristic vector x ═ compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity;
step 6.2, taking the data obtained in the step 6.1 as an input matrix of a train air conditioning unit fault diagnosis model based on a binary tree multi-classification strategy of distribution density sequencing:
firstly, classifying by using a classifier corresponding to a root node;
then, when the classification result is 1, classifying by using a classifier of the left sub-tree, and if the classification result is-1, classifying by using a classifier of the right sub-tree until the left sub-tree is a leaf node or the right sub-tree is a leaf node;
finally, the category corresponding to the leaf node is the current diagnosis result.
The train air conditioning unit fault diagnosis method based on the multi-classification strategy firstly calculates the distribution density of each class according to the distribution characteristics of the actual normal operation data and the multiple fault operation data of the train air conditioning unit, then carries out sequencing based on the silent junction distribution density, finally constructs a two-branch classification strategy tree based on PSVM (particle swarm optimization) by using the sequencing result, and obtains a train air conditioning unit fault diagnosis model by learning and training, thereby diagnosing by using the train air conditioning unit fault diagnosis model and determining the fault type of the train air conditioning unit.
Example (b):
take a volume of a hyper-sphere as an example.
Volume of the hyper-sphere:
30 5-dimensional training samples in the class 1 training sample set are calculated according to a formula (2) to obtain V1 2Class 1 hypersphere density according to equation (4) of 3.44
30 5-dimensional training samples in the class 2 training sample set are calculated according to the formula (2) to obtain V2 2Class 2 hypersphere according to equation (4) 2.24Density is rho2 2＝13.39；
30 5-dimensional training samples in the class 3 training sample set are calculated according to the formula (2) to obtain V3 2Class 3 hypersphere density is ρ, according to equation (4), 1.53 2＝20；
30 5-dimensional training samples in the class 4 training sample set are calculated according to the formula (2) to obtain V4 2Class 4 hypersphere density is ρ, according to equation (4), 2.324 2＝12.93；
30 5-dimensional training samples in the class 5 training sample set are calculated according to the formula (2) to obtain V5 2Class 5 hypersphere density according to equation (4) of 2.78 ρ5 2＝10.79；
30 5-dimensional training samples in the class 6 training sample set are obtained through calculation according to the formula (2)The density of 6-like hypersphere is ρ according to equation (4)6 2＝13.89；
The results of the ascending order arrangement according to the density of the hyper-spheres of each class are {8.72, 10.79, 12.93, 13.39, 13.89, 20}, i.e. the results are obtained
Taking each class as a leaf node of the binary tree, and constructing a Huffman tree according to the hypersphere density of each class; in this embodiment, there are 6 categories, the constructed huffman tree has 6 leaf nodes, the weight of each leaf node is the hypersphere density of the category corresponding to the leaf node, and the obtained huffman tree is shown in fig. 2 according to the construction rule of the huffman tree;
before classification, selecting a proper kernel function and an optimization parameter, and adopting a Gaussian kernel function:carrying out data classification, wherein x and z are input feature vectors, and the kernel parameters of a Gaussian kernel function and the values of adjustable parameters for controlling the punishment degree of the misclassification samples need to be determined;
the classification effect of different parameters is different, in this embodiment, the penalty parameter C is selected to be 0.8, and the kernel parameter is 0.1;
constructing an optimal hyperplane of each internal node in a Huffman tree by using a proximity support vector machine algorithm (PSVM) as a two-classification learning machine, namely an optimal classifier:
starting from a current inner node, firstly, setting all sample classes in a left subtree as positive classes, setting the sample classes of a right subtree as negative classes, and obtaining a PSVM classifier of the current inner node by utilizing a PSVM algorithm learning training, wherein the method specifically comprises the following steps:
first, a matrix A is constructed by using training sampleskIn a matrix AkThe first 5 columns of data in Table 3 are given as examples; then, the corresponding category of each sample is utilized to construct a diagonal matrix DkWherein the diagonal element is the current class for each sample, i.e. DkColumn 6 data with diagonal elements as table 3 is obtained;
order to <math>
<mrow>
<mi>I</mi>
<mo>=</mo>
<msub>
<mfenced open = '(' close = ')'>
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mrow>
<mi>d</mi>
<mo>×</mo>
<mi>d</mi>
</mrow>
</msub>
<mo>,</mo>
</mrow>
</math> Gk＝Dk[K(Ak,A′k)-e]；
Wherein e ═ 1, 1.. 1)'1×NThen, obtaining:
the obtained classification model is specifically as follows:
yk＝sign((K(x,A′k)K(Ak,A′k)+e′)Dkvk)；
wherein, <math>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>g</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfenced open = '{' close = ''>
<mtable>
<mtr>
<mtd>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>x</mi>
<mo><</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mrow>
<mi>x</mi>
<mo>≥</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>;</mo>
</mrow>
</math>
TABLE 3 training samples
And secondly, when the left subtree is a leaf node, the classifier of the current internal node is the classifier of the leaf node, and all training samples of the class corresponding to the leaf node are removed from the sample set. If the left subtree is an inner node, returning to the first step;
thirdly, when the right subtree is a leaf node, the classifier of the current inner node is the classifier of the leaf node, and all training samples of the class corresponding to the leaf node are removed from the sample set; otherwise, the right subtree is an inner node and returns to the first step;
a plurality of classifiers can be obtained through the first step to the third step, and the plurality of classifiers form a train air conditioning unit fault diagnosis model based on a binary tree multi-classification strategy of distribution density;
diagnosing the running state of the train air conditioner by using a train air conditioner set fault diagnosis model based on a binary tree multi-classification strategy of distribution density, wherein the diagnosis state comprises the following steps: normal operating conditions, refrigerant leakage, evaporator fouling, condenser fouling, non-condensable gases and compressor shutdown failure:
the enthalpy value of the inlet air of the train air conditioning unit is calculated according to the inlet air temperature of the train air conditioning unit and the inlet air humidity of the train air conditioning unit, the enthalpy value of the outlet air of the train air conditioning unit is calculated according to the outlet air temperature of the train air conditioning unit and the outlet air humidity of the train air conditioning unit, and the actual refrigerating capacity of the air conditioning unit is calculated according to the inlet and outlet enthalpy values and the air volume, so that data in the form of a characteristic vector x ═ compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity } are obtained, and test samples are shown in a table 4;
TABLE 4 test specimens
Taking the obtained data as an input matrix of a train air conditioning unit fault diagnosis model based on a binary tree multi-classification strategy of distribution density sequencing: and classifying by using a classifier corresponding to the root node, then classifying by using a classifier of the left sub-tree when the classification result is 1, and classifying by using a classifier of the right sub-tree when the classification result is-1 until the left sub-tree is a leaf node or the right sub-tree is a leaf node. And finally, the category corresponding to the leaf node is the current diagnosis result. Experiments prove that when the punishment parameter C is 0.8 and the nuclear parameter is 0.1, the classification effect is the best and is 100%.
The train air conditioning unit fault diagnosis method based on the multi-classification strategy adopts the train air conditioning unit fault diagnosis method based on the binary tree adjacent support vector machine multi-classification strategy of distribution density sequencing, can realize accurate diagnosis of the train air conditioning unit fault, and is very suitable for the field of train air conditioning unit fault diagnosis of high-speed railways.
Claims (8)
1. The train air conditioning unit fault diagnosis method based on the multi-classification strategy is characterized by being implemented according to the following steps:
step 1, determining a main fault type of a train air conditioning unit according to a working principle of a train air conditioning refrigeration system and common refrigeration air conditioning faults, and constructing a fault diagnosis model of the train air conditioning refrigeration system;
step 2, performing data acquisition and manual labeling according to the fault diagnosis model of the train air-conditioning refrigeration system constructed in the step 1, and acquiring a training sample set;
step 3, calculating to obtain the distribution density of each class by using the training sample set obtained by preprocessing in the step 2, arranging the classification order according to the sequence of the distribution density from light to heavy to obtain the most reasonable classification order, and constructing a binary classification strategy tree, namely a Huffman tree;
step 4, after the steps 1 to 3 are completed, selecting a kernel function;
step 5, constructing an optimal hyperplane of each internal node in the Huffman tree by using a near support vector machine algorithm PSVM as a two-class learning machine, namely constructing a train air conditioning unit fault diagnosis model based on a binary tree multi-class strategy of distribution density sorting;
and 6, diagnosing by using the train air conditioning unit fault diagnosis model obtained in the step 5 to finish the diagnosis of the train air conditioning unit fault type.
2. The train air conditioning unit fault diagnosis method based on the multi-classification strategy as claimed in claim 1, wherein the main fault types of the train air conditioning unit in the step 1 are as follows:
refrigerant leakage, evaporator fouling, condenser fouling, non-condensable gases and compressor shutdown;
when different faults occur, the variation amplitude of the operation parameters of the train air conditioning unit is different, and a variation characteristic corresponding table of the judgment parameters corresponding to the normal and fault states of the train air conditioning unit to each fault is obtained through data analysis;
and selecting the characteristic vector Xi ═ compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity as the input of the fault diagnosis model of the train air-conditioning refrigerating system.
3. The train air conditioning unit fault diagnosis method based on the multi-classification strategy as claimed in claim 1, wherein the step 2 is implemented specifically according to the following steps:
step 2.1, arranging a plurality of acquisition measuring points on the train air conditioning unit, and acquiring data under the conditions of simulating five faults and normal operation conditions of refrigerant leakage, evaporator dirt, condenser dirt, non-condensable gas and compressor shutdown;
the acquired data mainly comprises 9 parameters including temperature parameters, pressure parameters and humidity parameters, and the parameters are as follows:
temperature parameters 4: the temperature of the dry bulb of the air at the inlet of the train air conditioning unit, the temperature of the dry bulb of the air at the outlet of the train air conditioning unit, the air suction temperature of a compressor and the exhaust temperature of the compressor;
the pressure parameters are 3: the air suction pressure of the compressor, the exhaust pressure of the compressor and the outlet pressure difference of the train air conditioning unit are measured;
humidity parameters 2: the relative humidity of air at the inlet of the train air conditioning unit and the relative humidity of air at the outlet of the train air conditioning unit;
step 2.2, listing the names, codes, sensor types and installation positions of all the acquisition measuring points set in the step 2.1, and acquiring the following data:
the suction pressure and the exhaust pressure of the compressor are respectively obtained by measuring points of the inlet pressure and the outlet pressure of the compressor;
the air suction temperature and the air discharge temperature of the compressor are respectively obtained by measuring points of the inlet temperature and the outlet temperature of the compressor;
the train air conditioning unit inlet air dry bulb temperature, the train air conditioning unit inlet air relative humidity, the train air conditioning unit outlet air dry bulb temperature and the train air conditioning unit outlet air relative humidity are respectively obtained by train air conditioning unit inlet air temperature, train air conditioning unit inlet air humidity, train air conditioning unit outlet air temperature and train air conditioning unit outlet air humidity measuring points;
the train air conditioning unit outlet pressure difference is obtained by an air duct pressure difference measuring point;
step 2.3, acquiring values of each measuring point of the train air conditioning unit under the five fault working conditions and the normal operation working condition based on the data acquisition scheme combined with the step 2.1 and the step 2.2, calculating an enthalpy value of inlet air of the train air conditioning unit according to the inlet air temperature of the train air conditioning unit and the inlet air humidity of the train air conditioning unit, and calculating an enthalpy value of outlet air according to the outlet air temperature of the train air conditioning unit and the outlet air humidity of the train air conditioning unit; and then, calculating the actual refrigerating capacity of the air conditioning unit by using the enthalpy values of the air inlet and the air outlet and the air volume, thereby obtaining sample data in the form of a characteristic vector Xi ═ { compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity }, and manually marking a fault type label Yi ═ {1,2, 3, 4, 5 and 6}, wherein the sample data respectively represent six types Yi ═ { normal operating condition, refrigerant leakage, evaporator fouling, condenser fouling, non-condensable gas and compressor shutdown }.
4. The train air conditioning unit fault diagnosis method based on the multi-classification strategy as claimed in claim 1, wherein the step 3 is implemented specifically according to the following steps:
step 3.1, calculating to obtain the distribution density of each class by using the training sample set obtained by the pretreatment in the step 2;
the method is implemented according to the following steps:
step a, calculating the distribution volume of each class by the following two calculation methods: one is a super-cuboid volume; the other is a hypersphere volume;
for the volume of the super-rectangular body:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the volume of the hyper-cuboid containing all training samples in class S is:
for a hypersphere volume:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the hypersphere volume containing all training samples in class S is:
for super-rectangular bulk density:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the ultracuboid density of all training samples in class S is included:
for the hypersphere density:
let N training samples x with d dimensions in training sample set of class S1,x2,...,xN∈RdThen the hypersphere density of all training samples in class S is:
combining the formula (1) to the formula (4) to obtain the ultra-rectangular density and the ultra-spherical density of each class;
step 3.2, after the step 3.1, performing ascending arrangement according to the super rectangular density or super spherical density of each class from small to large;
if two or more categories have the same ultra-rectangular density or ultra-spherical density, randomly arranging;
and 3.3, after the step 3.2, taking each class as a leaf node of the binary tree, and constructing the Huffman tree according to the super-rectangular body density or the super-spherical body density of each class.
5. The train air conditioning unit fault diagnosis method based on the multi-classification strategy as claimed in claim 4, wherein the step 3.3 is implemented according to the following method:
if m categories are set, the constructed Huffman tree has m leaf nodes, and the weight of each leaf node is the super-rectangular body density of the corresponding categoryOr density of hyper-spheresWherein S belongs to {1, 2.. eta., m };
for ultra-rectangular bulk densityThe construction rule of the huffman tree is as follows:
first, the…，The forest is regarded as a forest with m trees, and each tree has only one node;
secondly, selecting and combining two trees with the minimum weight of root nodes in the forest as a left subtree and a right subtree of a new tree, wherein the weight of the root nodes of the new tree is the sum of the weights of the root nodes of the left subtree and the right subtree;
thirdly, deleting two selected trees from the forest and adding the new trees into the forest;
and fourthly, repeating the second step and the third step until only one tree is left in the forest, wherein the tree is the obtained Huffman tree.
6. The train air conditioning unit fault diagnosis method based on the multi-classification strategy according to claim 1, wherein the kernel function in the step 4 is a gaussian kernel function, and specifically comprises the following steps:
where x and z are input feature vectors.
7. The train air conditioning unit fault diagnosis method based on the multi-classification strategy as claimed in claim 1, wherein the step 5 is implemented specifically according to the following steps:
step 5.1, starting from the current inner node, firstly, taking all sample classes in the left subtree as positive classes, taking the sample classes of the right subtree as negative classes, and obtaining the PSVM classifier of the current inner node by utilizing the learning and training of a PSVM algorithm:
firstly, a matrix A is constructed by utilizing training sampleskThe method comprises the following steps:
Ak＝{x1…xi…xN}′(5)；
next, a diagonal matrix D is constructed by using the corresponding category of each samplekThe diagonal elements are the current category corresponding to each sample, and specifically the following are:
order to <math>
<mrow>
<mi>I</mi>
<mo>=</mo>
<msub>
<mfenced open = '(' close = ')'>
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mrow>
<mi>d</mi>
<mo>×</mo>
<mi>d</mi>
</mrow>
</msub>
<mo>,</mo>
</mrow>
</math> Gk＝Dk[K(Ak,A′k)-e]Wherein e ═ 1, 1.. 1)'1×N，
The following algorithm is obtained:
wherein v ∈ R;
finally, the obtained classification model is concretely as follows:
yk＝sign((K(x,A′k)K(Ak,A′k)+e′)Dkvk)(8)；
wherein, <math>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>g</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>{</mo>
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>x</mi>
<mo><</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mrow>
<mi>x</mi>
<mo>≥</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>;</mo>
</mrow>
</mrow>
</math>
step 5.2, when the left subtree is a leaf node, the classifier of the current internal node is the classifier of the leaf node, and all training samples of the class corresponding to the leaf node are removed from the sample set;
if the left subtree is an inner node, returning to the step 5.1;
step 5.3, when the right subtree is a leaf node, the classifier of the current inner node is the classifier of the leaf node, and all training samples of the class corresponding to the leaf node are removed from the sample set;
otherwise, the right subtree is an inner node, and the step 5.1 is returned to;
through the steps 5.1 to 5.3, a plurality of classifiers can be obtained, and the plurality of classifiers form a train air conditioning unit fault diagnosis model based on a binary tree multi-classification strategy of distribution density.
8. The train air conditioning unit fault diagnosis method based on the multi-classification strategy as claimed in claim 1, wherein the step 6 is implemented specifically according to the following steps:
diagnosing the running state of the train air conditioner by using the train air conditioner set fault diagnosis model based on the binary tree multi-classification strategy of the distribution density, which is obtained in the step 5, wherein the diagnosis state comprises the following steps: the method comprises the following steps of (1) carrying out normal operation working conditions, refrigerant leakage, evaporator dirt, condenser dirt, non-condensable gas and compressor shutdown faults according to the following specific method:
6.1, calculating an enthalpy value of inlet air of the train air conditioning unit according to the temperature of the inlet air of the train air conditioning unit and the humidity of the inlet air of the train air conditioning unit by using the values of the measuring points of the train air conditioning unit under six working conditions, which are acquired in the step 2;
calculating an enthalpy value of outlet air of the train air conditioning unit according to outlet air temperature of the train air conditioning unit and outlet air humidity of the train air conditioning unit, and further calculating actual refrigerating capacity of the air conditioning unit according to the inlet and outlet enthalpy values and air volume of the air, so as to obtain data in the form of a characteristic vector x ═ compressor suction pressure, compressor discharge pressure, compressor suction temperature, compressor discharge temperature and unit actual refrigerating capacity;
step 6.2, taking the data obtained in the step 6.1 as an input matrix of a train air conditioning unit fault diagnosis model based on a binary tree multi-classification strategy of distribution density sequencing:
firstly, classifying by using a classifier corresponding to a root node;
then, when the classification result is 1, classifying by using a classifier of the left sub-tree, and if the classification result is-1, classifying by using a classifier of the right sub-tree until the left sub-tree is a leaf node or the right sub-tree is a leaf node;
finally, the category corresponding to the leaf node is the current diagnosis result, and the diagnosis of the train air conditioning unit fault is completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510378576.6A CN105135591B (en) | 2015-07-01 | 2015-07-01 | Train air conditioning unit method for diagnosing faults based on more classification policys |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510378576.6A CN105135591B (en) | 2015-07-01 | 2015-07-01 | Train air conditioning unit method for diagnosing faults based on more classification policys |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105135591A true CN105135591A (en) | 2015-12-09 |
| CN105135591B CN105135591B (en) | 2018-04-10 |
Family
ID=54721043
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510378576.6A Active CN105135591B (en) | 2015-07-01 | 2015-07-01 | Train air conditioning unit method for diagnosing faults based on more classification policys |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105135591B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105703777A (en) * | 2016-02-01 | 2016-06-22 | 合肥华凌股份有限公司 | Refrigerator reported data compressing method and device |
| CN106600047A (en) * | 2016-12-09 | 2017-04-26 | 上海铁路局 | Method for fault identification and early warning of air conditioners in passenger compartments of motor train unit based on BP neural network |
| CN107255353A (en) * | 2017-07-28 | 2017-10-17 | 广东美的暖通设备有限公司 | Multiple on-line system and its abnormal diagnostic method of delivery temperature detection, device |
| CN107421071A (en) * | 2017-07-28 | 2017-12-01 | 广东美的暖通设备有限公司 | Diagnostic method, the device of multiple on-line system and its delivery temperature detection exception |
| CN108830840A (en) * | 2018-05-31 | 2018-11-16 | 江苏华鹰光电科技有限公司 | A kind of active intelligent detecting method of circuit board defect and its application |
| CN110390671A (en) * | 2019-07-10 | 2019-10-29 | 杭州依图医疗技术有限公司 | A kind of method and device of Breast Calcifications detection |
| EP3531036A4 (en) * | 2016-11-21 | 2019-10-30 | Mitsubishi Electric Corporation | Deterioration diagnosis apparatus, deterioration diagnosis method, and deterioration diagnosis program |
| CN110455560A (en) * | 2019-07-19 | 2019-11-15 | 中车青岛四方机车车辆股份有限公司 | The repair method of vehicle, device and system |
| CN111486552A (en) * | 2020-04-24 | 2020-08-04 | 辽宁工程技术大学 | Method for identifying water supply temperature strategy of chilled water of air conditioner based on subentry metering data |
| CN111546854A (en) * | 2020-06-18 | 2020-08-18 | 中南大学 | On-road identification and diagnosis method for intelligent train air conditioning unit |
| CN111637045A (en) * | 2020-05-09 | 2020-09-08 | 哈尔滨工程大学 | Fault diagnosis method for air compressor of ocean platform |
| CN111829137A (en) * | 2020-07-29 | 2020-10-27 | 四川长虹空调有限公司 | Method and system for identifying mixed air fault of air conditioning system |
| CN113503960A (en) * | 2021-07-21 | 2021-10-15 | 美核电气（济南）股份有限公司 | Mechanical system fault diagnosis method and system based on visual image |
| CN114881110A (en) * | 2022-04-02 | 2022-08-09 | 西安交通大学 | Real-time detection method for total pressure change mode in on-orbit spacecraft cabin |
| CN116975502A (en) * | 2023-09-21 | 2023-10-31 | 天津津轨汇海科技发展有限公司 | Energy efficiency monitoring management system for electromechanical equipment of subway train air conditioning system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5337371A (en) * | 1991-08-09 | 1994-08-09 | Matsushita Electric Industrial Co., Ltd. | Pattern classification system |
| US20120185728A1 (en) * | 2010-12-24 | 2012-07-19 | Commonwealth Scientific And Industrial Research Organisation | System and method for detecting and/or diagnosing faults in multi-variable systems |
| CN104573740A (en) * | 2014-12-22 | 2015-04-29 | 山东鲁能软件技术有限公司 | SVM classification model-based equipment fault diagnosing method |
| CN104697765A (en) * | 2014-08-26 | 2015-06-10 | 安徽工程大学 | Method and system for detecting faults of automotive air conditioner |
-
2015
- 2015-07-01 CN CN201510378576.6A patent/CN105135591B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5337371A (en) * | 1991-08-09 | 1994-08-09 | Matsushita Electric Industrial Co., Ltd. | Pattern classification system |
| US20120185728A1 (en) * | 2010-12-24 | 2012-07-19 | Commonwealth Scientific And Industrial Research Organisation | System and method for detecting and/or diagnosing faults in multi-variable systems |
| CN104697765A (en) * | 2014-08-26 | 2015-06-10 | 安徽工程大学 | Method and system for detecting faults of automotive air conditioner |
| CN104573740A (en) * | 2014-12-22 | 2015-04-29 | 山东鲁能软件技术有限公司 | SVM classification model-based equipment fault diagnosing method |
Non-Patent Citations (4)
| Title |
|---|
| 姜锐: "一种基于支持向量机集成的决策树分类算法", 《科技通报》 * |
| 滕少华: "支持向量机与哈夫曼树实现多分类的研究", 《江西师范大学学报（自然科学版）》 * |
| 王凯: "基于多分类支持向量机的有杆抽油泵故障诊断研究", 《西安石油大学学报（自然科学版）》 * |
| 罗浩: "列车空调机组故障检修中SVM的应用", 《中国学位论文全文数据库》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105703777B (en) * | 2016-02-01 | 2019-02-12 | 合肥美的智能科技有限公司 | A kind of method and apparatus for compressing refrigerator reported data |
| CN105703777A (en) * | 2016-02-01 | 2016-06-22 | 合肥华凌股份有限公司 | Refrigerator reported data compressing method and device |
| EP3531036A4 (en) * | 2016-11-21 | 2019-10-30 | Mitsubishi Electric Corporation | Deterioration diagnosis apparatus, deterioration diagnosis method, and deterioration diagnosis program |
| CN106600047A (en) * | 2016-12-09 | 2017-04-26 | 上海铁路局 | Method for fault identification and early warning of air conditioners in passenger compartments of motor train unit based on BP neural network |
| CN107255353B (en) * | 2017-07-28 | 2020-09-04 | 广东美的暖通设备有限公司 | Multi-split system and method and device for diagnosing exhaust temperature detection abnormality of multi-split system |
| CN107255353A (en) * | 2017-07-28 | 2017-10-17 | 广东美的暖通设备有限公司 | Multiple on-line system and its abnormal diagnostic method of delivery temperature detection, device |
| CN107421071A (en) * | 2017-07-28 | 2017-12-01 | 广东美的暖通设备有限公司 | Diagnostic method, the device of multiple on-line system and its delivery temperature detection exception |
| CN107421071B (en) * | 2017-07-28 | 2019-12-03 | 广东美的暖通设备有限公司 | Diagnostic method, the device of multi-line system and its delivery temperature detection exception |
| CN108830840A (en) * | 2018-05-31 | 2018-11-16 | 江苏华鹰光电科技有限公司 | A kind of active intelligent detecting method of circuit board defect and its application |
| CN110390671A (en) * | 2019-07-10 | 2019-10-29 | 杭州依图医疗技术有限公司 | A kind of method and device of Breast Calcifications detection |
| CN110390671B (en) * | 2019-07-10 | 2021-11-30 | 杭州依图医疗技术有限公司 | Method and device for detecting mammary gland calcification |
| CN110455560A (en) * | 2019-07-19 | 2019-11-15 | 中车青岛四方机车车辆股份有限公司 | The repair method of vehicle, device and system |
| CN111486552A (en) * | 2020-04-24 | 2020-08-04 | 辽宁工程技术大学 | Method for identifying water supply temperature strategy of chilled water of air conditioner based on subentry metering data |
| CN111637045A (en) * | 2020-05-09 | 2020-09-08 | 哈尔滨工程大学 | Fault diagnosis method for air compressor of ocean platform |
| CN111637045B (en) * | 2020-05-09 | 2021-09-10 | 哈尔滨工程大学 | Fault diagnosis method for air compressor of ocean platform |
| CN111546854B (en) * | 2020-06-18 | 2021-08-24 | 中南大学 | On-road identification and diagnosis method for intelligent train air conditioning unit |
| CN111546854A (en) * | 2020-06-18 | 2020-08-18 | 中南大学 | On-road identification and diagnosis method for intelligent train air conditioning unit |
| CN111829137A (en) * | 2020-07-29 | 2020-10-27 | 四川长虹空调有限公司 | Method and system for identifying mixed air fault of air conditioning system |
| CN111829137B (en) * | 2020-07-29 | 2021-10-15 | 四川长虹空调有限公司 | Method and system for identifying mixed air fault of air conditioning system |
| CN113503960A (en) * | 2021-07-21 | 2021-10-15 | 美核电气（济南）股份有限公司 | Mechanical system fault diagnosis method and system based on visual image |
| CN114881110A (en) * | 2022-04-02 | 2022-08-09 | 西安交通大学 | Real-time detection method for total pressure change mode in on-orbit spacecraft cabin |
| CN116975502A (en) * | 2023-09-21 | 2023-10-31 | 天津津轨汇海科技发展有限公司 | Energy efficiency monitoring management system for electromechanical equipment of subway train air conditioning system |
| CN116975502B (en) * | 2023-09-21 | 2023-12-19 | 天津津轨汇海科技发展有限公司 | Energy efficiency monitoring management system for electromechanical equipment of subway train air conditioning system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105135591B (en) | 2018-04-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105135591B (en) | Train air conditioning unit method for diagnosing faults based on more classification policys | |
| CN105046210B (en) | The multi-tag diagnostic method of train air-conditioning failure | |
| CN105160345B (en) | A method of diagnosis Train air conditioning unit failure | |
| CN110162014A (en) | A kind of breakdown of refrigeration system diagnostic method of integrated multi-intelligence algorithm | |
| CN106845717B (en) | Energy efficiency evaluation method based on multi-model fusion strategy | |
| CN114484731B (en) | Central air conditioner fault diagnosis method and device based on stacking fusion algorithm | |
| Ertunc et al. | Comparative analysis of an evaporative condenser using artificial neural network and adaptive neuro-fuzzy inference system | |
| Li et al. | Diagnosis for multiple faults of chiller using ELM-KNN model enhanced by multi-label learning and specific feature combinations | |
| Li et al. | Review on fault detection and diagnosis feature engineering in building heating, ventilation, air conditioning and refrigeration systems | |
| Zhang et al. | Imbalanced data based fault diagnosis of the chiller via integrating a new resampling technique with an improved ensemble extreme learning machine | |
| Rashidi et al. | Analysis and optimization of a transcritical power cycle with regenerator using artificial neural networks and genetic algorithms | |
| CN110162013A (en) | A kind of breakdown of refrigeration system diagnostic method | |
| Wang et al. | Liquid floodback detection for scroll compressor in a VRF system under heating mode | |
| Yu et al. | A VRF charge fault diagnosis method based on expert modification C5. 0 decision tree | |
| Li et al. | Across working conditions fault diagnosis for chillers based on IoT intelligent agent with deep learning model | |
| CN104700147A (en) | Cooling tower water outlet temperature predicating method | |
| CN114353261A (en) | Air conditioning unit fault analysis method and device, terminal device and storage medium | |
| Xia et al. | Incipient chiller fault diagnosis using an optimized least squares support vector machine with gravitational search algorithm | |
| Rashidi et al. | Exergetic optimisation of a multi-stage compression transcritical refrigeration cycle | |
| Masdoua et al. | Fault detection and diagnosis in AHU system with data driven approaches | |
| Sathesh et al. | Optimized deep learning-based prediction model for chiller performance prediction | |
| Deng et al. | A novel semi-supervised fault diagnosis method for chillers based on neighbor-optimized graph convolutional network | |
| Wang et al. | Fault detection and diagnosis of HVAC system based on federated learning | |
| Li et al. | An interpretable graph convolutional neural network based fault diagnosis method for building energy systems | |
| Liu et al. | Fault Diagnosis Based on Prior Knowledge for Train Air-Conditioning Unit |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |