Summary of the invention
The object of the present invention is to provide a kind of many labels diagnostic method of train air-conditioning fault, can accurately to Train air conditioning unit diagnosing malfunction.
The technical solution adopted in the present invention is, many labels diagnostic method of train air-conditioning fault, specifically implements according to following steps:
Step 1, according to train air conditioner refrigerating system principle of work and common refrigeration air-conditioner fault, determine the major failure type of train air-conditioning, and build train air conditioner refrigerating system fault diagnosis model;
Step 2, build train air conditioner refrigerating system model through step 1 after, carry out data acquisition and artificial mark, and obtain training sample set;
The selection of step 3, kernel function and optimized parameter;
Step 4, formation Train air conditioning unit fault many labels diagnostic model;
Step 5, utilize step 4 to obtain Train air conditioning unit fault many labels diagnostic model to carry out Train air conditioning unit fault diagnosis.
Feature of the present invention is also:
In step 1, the major failure type of train air-conditioning is as follows:
Leakage of refrigerant, evaporator dirt, Fouling in Condenser, there are fouling gas and compressor shutdown;
When there is different faults, the operational factor amplitude of variation of Train air conditioning unit is different, through data analysis, obtains the variation characteristic correspondence of judgement parameter to each fault that Train air conditioning unit is normal and each malfunction is corresponding and shows;
Select feature vector, X i={ suction pressure of compressor, Compressor Discharge Pressure, compressor air suction temperature, compressor exhaust temperature, the actual refrigerating capacity of unit } as the input of Train air conditioning unit fault diagnosis model.
Step 2 is specifically implemented according to following steps:
Step 2.1, Train air conditioning unit arrange many places gather measuring point, simulation leakage of refrigerant, evaporator dirt, Fouling in Condenser, have fouling gas and compressor shutdown five kinds of faults and accidental conditions under, carry out data acquisition;
The data gathered mainly are divided into temperature parameter, pressure parameter and humidity parameter three classes totally 9 parameters, specific as follows:
Temperature parameter 4: Train air conditioning unit inlet air dry-bulb temperature, Train air conditioning unit outlet air dry-bulb temperature, compressor air suction temperature, compressor exhaust temperature;
Pressure parameter 3: suction pressure of compressor, Compressor Discharge Pressure, Train air conditioning unit outlet pressure reduction;
Humidity parameter 2: unit Relative humidity of intake air, machine set outlet relative air humidity;
Step 2.2, list is carried out in the title of all collection measuring points arranged in step 2.1, code name, sensor type and installation site;
Suction pressure of compressor and discharge pressure are obtained by Compressor Inlet Pressure and compressor delivery pressure measuring point respectively;
Compressor air suction temperature and compressor exhaust temperature are obtained by Compressor Inlet Temperature and compressor exit temperature measuring point respectively;
Train air conditioning unit inlet air dry-bulb temperature, Train air conditioning unit Relative humidity of intake air, Train air conditioning unit outlet air dry-bulb temperature and Train air conditioning unit outlet air relative humidity are obtained by Train air conditioning unit inlet air temperature, Train air conditioning unit inlet air humidity, Train air conditioning unit outlet air temperature, Train air conditioning unit exhaust air humidity measuring point respectively; Train air conditioning unit outlet pressure reduction is obtained by air channel pressure reduction measuring point;
Step 2.3, the data acquisition plan combined based on step 2.1 and step 2.2, gather the value of Train air conditioning unit each measuring point under above five kinds of faults and accidental conditions, and the enthalpy of Train air conditioning unit inlet air is calculated by Train air conditioning unit inlet air temperature, Train air conditioning unit inlet air hygrometer, the enthalpy of outlet air is calculated by Train air conditioning unit outlet air temperature, Train air conditioning unit exhaust air humidity; And then the actual refrigerating capacity of air-conditioner set is gone out by air import and export enthalpy and air Wind Coverage Calculation; thus obtain shape as feature vector, X i={ suction pressure of compressor; Compressor Discharge Pressure; compressor air suction temperature, compressor exhaust temperature, the actual refrigerating capacity of unit } sample data; and artificial mark fault type label Yi={1; 2,3,4; 5; 6}, represents six type Yi={ accidental conditions, leakage of refrigerant respectively; evaporator dirt; Fouling in Condenser, has fouling gas, compressor shutdown }.
The kernel function adopted in step 3 is gaussian kernel function:
wherein, x and z is input feature value; Optimum punishment parameter C and nuclear parameter σ, utilizes particle cluster algorithm optimizing to obtain.
Step 4 is concrete to be implemented in accordance with the following methods:
Adopt many labels fault diagnosis algorithm of one-to-many, construct six Proximal Support Vector Machine sorters respectively for six kinds of different faults states;
The key step obtaining a kth sorter is as follows:
Step a, will have all training samples of kth class label as positive class, remaining sample, as negative class, utilizes all training samples to build matrix A
k, formula specific as follows:
A
k={x
1…x
i…x
N}′(1);
Step b, the new classification utilizing each sample corresponding build diagonal matrix D
k, wherein diagonal element is the current class that each sample is corresponding, formula specific as follows:
Order
g
k=D
k[K (A
k, A '
k)-e] and wherein e=(1,1 ..., 1) '
1 × N,
Then obtain:
Wherein, v ∈ R;
Step c, the result utilizing step b to obtain, obtain disaggregated model, formula specific as follows:
y
k=sign((K(x,A′
k)K(A
k,A′
k)+e′)D
kv
k)(4);
Wherein,
Reuse above-mentioned three little steps, just can obtain all six sorters; These six sorters just constitute Train air conditioning unit fault many labels diagnostic model.
Step 5 is specifically implemented according to following steps:
Utilize step 4 to obtain Train air conditioning unit fault many labels diagnostic model and carry out Train air conditioning unit fault diagnosis; Diagnostic state comprises: accidental conditions, leakage of refrigerant, evaporator dirt, Fouling in Condenser, have fouling gas and compressor shutdown fault, specifically implements according to following steps:
Step 5.1, utilize the data acquisition plan of step 2 to collect the value of train air-conditioning each measuring point under above operating mode, and calculate the enthalpy of Train air conditioning unit inlet air by Train air conditioning unit inlet air temperature, Train air conditioning unit inlet air hygrometer; The enthalpy of outlet air is calculated by Train air conditioning unit outlet air temperature, Train air conditioning unit exhaust air humidity; And then the actual refrigerating capacity of air-conditioner set is gone out by air import and export enthalpy and air Wind Coverage Calculation, thus obtain shape as proper vector x={ suction pressure of compressor, Compressor Discharge Pressure, compressor air suction temperature, compressor exhaust temperature, the actual refrigerating capacity of unit } input vector;
Step 5.2, the data that step 5.1 obtained are as the input of Train air conditioning unit fault many labels diagnostic model; Six PSVM sorter models are obtained in Train air conditioning unit fault many labels diagnostic model and step 5;
Adopt fault type label y={1,2,3,4,5,6} represents, represent six kinds of operating mode types respectively: accidental conditions, leakage of refrigerant, evaporator dirt, Fouling in Condenser, has fouling gas, compressor shutdown;
The input vector obtained through step 5.1 is input to this six disaggregated model y successively
k=sign ((K (x, A '
k) K (A
k, A '
k)+e ') D
kv
k) in, obtaining diagnostic result vector is (y
1, y
2..., y
6); If wherein certain several y
kwhen being 1, just diagnosing it which has and plant fault.
Beneficial effect of the present invention is:
Many labels diagnostic method of train air-conditioning fault of the present invention is first that basic classification algorithm sets up many labelings mechanism with Proximal Support Vector Machine, then Train air conditioning unit actual normally service data and various faults service data is gathered as training sample, learning training obtains multiple single label sub-classifier, finally utilize Train air conditioning unit actual operating data as the input of these single label sub-classifiers, obtain multiple subclassification result vector, namely this result is the solution of many labelings problem.Many labels diagnostic method of train air-conditioning fault of the present invention not only increases the precision of fault diagnosis, and diagnostic result more comprehensively; Many labels diagnostic method of train air-conditioning fault of the present invention also can be used for High Speed Railway Trains air-conditioner set fault diagnosis field.
Embodiment
Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.
Many labels diagnostic method of train air-conditioning fault of the present invention, specifically implement according to following steps:
Step 1, according to train air conditioner refrigerating system principle of work and common refrigeration air-conditioner fault, determine the major failure type of train air-conditioning, and build train air conditioner refrigerating system fault diagnosis model, specifically according to following steps:
Build the structure of train air conditioner refrigerating system model based on Train air conditioning unit; Existing Train air conditioning unit is connected to form a closed refrigeration system primarily of evaporator, condenser, expansion valve and compressor by pipeline;
The principle of work of existing train air conditioner refrigerating system is: liquid refrigerant is produced by the continuous circulation of refrigeration system circuit, evaporate in evaporator, with cooled air generation exchange heat, steam is vaporized into after absorbing the heat of cooled air, the steam of generation is taken away by compressor from evaporator subsequently, and compressed refrigerant, make it under high pressure be discharged; High temperature after compression, high compressed steam are cooled by ambient air in condenser, condense into highly pressurised liquid, heating power expansion valve is utilized to make highly pressurized liquid throttling, low pressure after throttling, Low Temperature Wet steam enter evaporator, again vaporize, absorb the heat of cooled air, so go round and begin again, as shown in Figure 1.
Based on principle of work and the existing fault occurrences of train air conditioner refrigerating system, can show that train air-conditioning major failure is: leakage of refrigerant, evaporator dirt, Fouling in Condenser, have fouling gas and compressor shutdown.The present invention will select these five kinds of typical faults as research object.
When there is different faults, the operational factor amplitude of variation of Train air conditioning unit is different, through data analysis, obtains the variation characteristic correspondence of judgement parameter to each fault that Train air conditioning unit is normal and each malfunction is corresponding and shows, as shown in table 1; Select feature vector, X i={ suction pressure of compressor, Compressor Discharge Pressure, compressor air suction temperature, compressor exhaust temperature, the actual refrigerating capacity of unit } as the input of train air-conditioning unit fault diagnosis model.
Table 1 fault distinguishing Parameters variation rule list
Symbol in table 1 represents :=represent that parameter is substantially unchanged; ++ represent that parameter obviously increases, alter a great deal;--represent that parameter obviously reduces, alter a great deal; + representing that parameter has increased slightly, change is little;-representing that parameter slightly reduces, change is little.
Step 2, build train air conditioner refrigerating system model through step 1 after, carry out data acquisition and artificial mark, and obtain training sample set, specifically implement according to following steps:
Step 2.1, Train air conditioning unit arrange many places gather measuring point, simulation leakage of refrigerant, evaporator dirt, Fouling in Condenser, have fouling gas and compressor shutdown five kinds of faults and accidental conditions under, carry out data acquisition;
The data gathered mainly are divided into temperature parameter, pressure parameter and humidity parameter three classes totally 9 parameters, specific as follows:
(1) temperature parameter 4: Train air conditioning unit inlet air dry-bulb temperature, Train air conditioning unit outlet air dry-bulb temperature, compressor air suction temperature, compressor exhaust temperature;
(2) pressure parameter 3: suction pressure of compressor, Compressor Discharge Pressure, Train air conditioning unit outlet pressure reduction;
(3) humidity parameter 2: unit Relative humidity of intake air, Train air conditioning unit outlet air relative humidity;
Step 2.2, list is carried out in the title of all collection measuring points arranged in step 2.1, code name, sensor type and installation site, specifically as shown in table 2;
Table 2 test point information
Suction pressure of compressor and discharge pressure are obtained by Compressor Inlet Pressure and compressor delivery pressure measuring point respectively; Compressor air suction temperature and compressor exhaust temperature are obtained by Compressor Inlet Temperature and compressor exit temperature measuring point respectively;
Train air conditioning unit inlet air dry-bulb temperature, Train air conditioning unit Relative humidity of intake air, Train air conditioning unit outlet air dry-bulb temperature and Train air conditioning unit outlet air relative humidity are obtained by Train air conditioning unit inlet air temperature, Train air conditioning unit inlet air humidity, Train air conditioning unit outlet air temperature, Train air conditioning unit exhaust air humidity measuring point respectively;
Train air conditioning unit outlet pressure reduction is obtained by air channel pressure reduction measuring point;
Step 2.3, the data acquisition plan combined based on step 2.1 and step 2.2, gather the value of Train air conditioning unit each measuring point under above five kinds of faults (leakage of refrigerant, evaporator dirt, Fouling in Condenser, have fouling gas and compressor shutdown) and accidental conditions, and the enthalpy of Train air conditioning unit inlet air is calculated by Train air conditioning unit inlet air temperature, Train air conditioning unit inlet air hygrometer, the enthalpy of outlet air is calculated by Train air conditioning unit outlet air temperature, Train air conditioning unit exhaust air humidity; And then the actual refrigerating capacity of air-conditioner set is gone out by air import and export enthalpy and air Wind Coverage Calculation; thus obtain shape as feature vector, X i={ suction pressure of compressor; Compressor Discharge Pressure; compressor air suction temperature, compressor exhaust temperature, the actual refrigerating capacity of unit } sample data; and artificial mark fault type label Yi={1; 2,3,4; 5; 6}, represents six type Yi={ accidental conditions, leakage of refrigerant respectively; evaporator dirt; Fouling in Condenser, has fouling gas, compressor shutdown }.
The selection of step 3, kernel function and optimized parameter:
Relate to kernel method in many labels diagnostic method of train air-conditioning fault of the present invention, kernel method is a kind of effective ways solving non-linear fault diagnosis problem, and kernel method, for the situation of linearly inseparable, uses a nonlinear transformation
input space X is mapped to a high-dimensional feature space F, in feature space, is then become the situation of linear separability, finally use linear classification model to classify to it, so select the performance contributing to improving sorter of suitable kernel function;
Select gaussian kernel function:
wherein, x and z is input feature value; Optimum punishment parameter C and nuclear parameter σ, utilizes particle cluster algorithm optimizing to obtain.
Step 4, formation Train air conditioning unit fault many labels diagnostic model:
Adopt many labels fault diagnosis algorithm of one-to-many, construct six Proximal Support Vector Machine (PSVM) sorters respectively for six kinds of different faults states;
The key step obtaining a kth sorter is as follows:
Step a, will have all training samples of kth class label as positive class, remaining sample, as negative class, utilizes all training samples to build matrix A
k, formula specific as follows:
A
k={x
1…x
i…x
N}′(1);
Step b, the new classification utilizing each sample corresponding build diagonal matrix D
k, wherein diagonal element is the current class that each sample is corresponding, formula specific as follows:
Order
g
k=D
k[K (A
k, A '
k)-e] wherein e=(1,1 ..., 1) '
1 × N,
Then obtain:
Wherein, v ∈ R;
Step c, the result utilizing step b to obtain, obtain disaggregated model, formula specific as follows:
y
k=sign((K(x,A′
k)K(A
k,A′
k)+e′)D
kv
k)(4);
Wherein,
Reuse above-mentioned three little steps (step a is to step c), just can obtain all six sorters;
These six sorters just constitute Train air conditioning unit fault many labels diagnostic model.
Step 5, utilize step 4 to obtain Train air conditioning unit fault many labels diagnostic model to carry out Train air conditioning unit fault diagnosis;
Diagnostic state comprises: accidental conditions, leakage of refrigerant, evaporator dirt, Fouling in Condenser, have fouling gas and compressor shutdown fault, specifically implements according to following steps:
Step 5.1, utilize the data acquisition plan of step 2 to collect the value of train air-conditioning each measuring point under above operating mode, and calculate the enthalpy of Train air conditioning unit inlet air by Train air conditioning unit inlet air temperature, Train air conditioning unit inlet air hygrometer; The enthalpy of outlet air is calculated by Train air conditioning unit outlet air temperature, Train air conditioning unit exhaust air humidity; And then the actual refrigerating capacity of air-conditioner set is gone out by air import and export enthalpy and air Wind Coverage Calculation, thus obtain shape as proper vector x={ suction pressure of compressor, Compressor Discharge Pressure, compressor air suction temperature, compressor exhaust temperature, the actual refrigerating capacity of unit } input vector;
Step 5.2, the data that step 5.1 obtained are as the input of Train air conditioning unit fault many labels diagnostic model;
Six PSVM sorter models are obtained in Train air conditioning unit fault many labels diagnostic model and step 5;
Adopt fault type label y={1,2,3,4,5,6} represents, represent six kinds of operating mode types respectively: accidental conditions, leakage of refrigerant, evaporator dirt, Fouling in Condenser, has fouling gas, compressor shutdown;
The input vector obtained through step 5.1 is input to this six disaggregated model y successively
k=sign ((K (x, A '
k) K (A
k, A '
k)+e ') D
kv
k) in, obtaining diagnostic result vector is (y
1, y
2..., y
6); If wherein certain several y
kwhen being 1, just diagnosing it which has and plant fault.
Embodiment:
Sample data gathers and artificial mark: for the KLD29 series unit formula coach air conditioner unit that China is conventional, according to the scheme in the train air conditioner refrigerating system model in technical scheme at simulation accidental conditions, leakage of refrigerant, evaporator dirt, Fouling in Condenser, has fouling gas and compressor shutdown these six kinds, carry out data acquisition respectively, collect the data as table 3, as the input of algorithm, specifically in table 3;
Table 3 training sample
Artificial mark fault type label y={1,2,3,4,5,6} be representing fault type respectively: accidental conditions, leakage of refrigerant, evaporator dirt, and Fouling in Condenser, has fouling gas, compressor shutdown;
The selection of kernel function and optimized parameter:
Before use PSVM carries out Data classification, suitable kernel function and the most optimized parameter be selected; Adopt gaussian kernel function:
carry out Data classification, wherein x and z is input feature value; At this moment to determine the nuclear parameter of gaussian kernel function and control to divide sample to punish the value of the adjustable parameter of degree to mistake; The classifying quality of different parameters is different, adopts population to seek ginseng and obtains optimum punishment parameter C=0.7 and nuclear parameter σ=0.5;
The generation of Train air conditioning unit fault many labels diagnostic model:
Adopt the many labels fault diagnosis algorithm based on the one-to-many of priori, construct six Proximal Support Vector Machine (PSVM) sorters respectively for six kinds of different faults states, for the 1st kind of fault, specific as follows:
First small step, using class label be the sample of 2 all as positive class, using the sample of non-for class label 2 all as negative class.All training samples are utilized to build matrix A
1, as shown in table 4;
The training sample of the corresponding sorter of table 4 label 2
Second small step, the new classification utilizing each sample corresponding builds diagonal matrix D
1, wherein diagonal element is the current class that each sample is corresponding;
Order
g
1=D
1[K (A
1, A '
1)-e];
Wherein, e=(1,1 ..., 1) '
1 × N, formula (3) can be utilized, calculate v
1;
3rd small step, the result utilizing the second small step to obtain, obtaining disaggregated model is: y
1=sign ((K (x, A '
1) K (A
1, A '
1)+e ') D
1v
1); Wherein
Reuse above-mentioned three little steps, just can obtain all six sorters, these six sorters just constitute Train air conditioning unit fault many labels diagnostic model;
Utilize Train air conditioning unit fault many labels diagnostic model to diagnose the running status of train air-conditioning, diagnostic state comprises: accidental conditions, leakage of refrigerant, evaporator dirt, Fouling in Condenser, have fouling gas and compressor shutdown fault:
First small step, by the train air-conditioning that collects under above operating mode the value of (five kinds of fault conditions and normal operating conditions under) each measuring point, and the enthalpy of Train air conditioning unit inlet air is calculated by Train air conditioning unit inlet air temperature, Train air conditioning unit inlet air hygrometer, the enthalpy of outlet air is calculated by Train air conditioning unit outlet air temperature, Train air conditioning unit exhaust air humidity; And then the actual refrigerating capacity of air-conditioner set is gone out by air import and export enthalpy and air Wind Coverage Calculation, thus obtain shape as proper vector x={ suction pressure of compressor, Compressor Discharge Pressure, compressor air suction temperature, compressor exhaust temperature, the actual refrigerating capacity of unit } input vector, as shown in table 5;
Table 5 test sample book
Sample number |
Pressure of inspiration(Pi)/Bar |
Discharge pressure/Bar |
Suction temperature/DEG C |
Delivery temperature/DEG C |
Refrigerating capacity/kW |
1 |
5.86 |
19.02 |
16.00 |
87.00 |
28.50 |
2 |
5.94 |
20.29 |
13.00 |
84.00 |
28.40 |
3 |
6.02 |
19.59 |
14.00 |
82.00 |
28.00 |
4 |
5.90 |
19.99 |
16.00 |
82.00 |
27.50 |
5 |
6.02 |
19.06 |
16.00 |
85.00 |
27.10 |
6 |
5.44 |
17.76 |
16.50 |
93.50 |
24.45 |
7 |
4.87 |
17.03 |
15.00 |
93.50 |
24.90 |
8 |
6.68 |
13.33 |
26.00 |
72.00 |
11.95 |
9 |
5.48 |
19.38 |
21.00 |
97.00 |
24.70 |
10 |
5.15 |
18.61 |
18.00 |
93.00 |
24.30 |
11 |
4.83 |
17.17 |
15.50 |
91.00 |
24.90 |
12 |
5.94 |
21.44 |
15.00 |
92.50 |
27.00 |
13 |
5.62 |
19.60 |
14.50 |
88.50 |
25.70 |
14 |
5.61 |
20.90 |
12.50 |
89.00 |
26.40 |
15 |
7.06 |
15.61 |
20.00 |
66.50 |
14.25 |
16 |
5.70 |
19.95 |
12.50 |
89.50 |
25.80 |
17 |
7.48 |
16.84 |
22.50 |
71.00 |
13.85 |
18 |
6.27 |
23.00 |
19.50 |
95.50 |
26.35 |
19 |
5.20 |
18.30 |
18.50 |
95.00 |
24.50 |
20 |
7.25 |
19.86 |
22.33 |
79.67 |
18.20 |
21 |
5.34 |
19.81 |
20.50 |
98.50 |
24.95 |
22 |
5.82 |
21.14 |
15.50 |
94.50 |
26.65 |
23 |
6.27 |
23.00 |
19.50 |
95.50 |
26.35 |
24 |
7.67 |
17.69 |
24.00 |
74.00 |
14.20 |
25 |
5.32 |
19.14 |
16.67 |
95.67 |
25.23 |
26 |
6.72 |
13.19 |
25.50 |
74.50 |
11.95 |
27 |
7.26 |
15.41 |
20.00 |
67.00 |
13.90 |
28 |
7.54 |
16.02 |
24.50 |
67.00 |
13.45 |
29 |
7.72 |
17.92 |
24.00 |
76.00 |
14.70 |
30 |
6.26 |
14.87 |
21.00 |
76.33 |
16.67 |
Second small step, using the input of the every a line test data in table 5 as Train air conditioning unit fault many labels diagnostic model, is input to this six disaggregated model y successively
k=sign ((K (x, A '
k) K (A
k, A ' k) and+e ') D
kv
k) in, obtaining diagnostic result vector is (y
1, y
2..., y
6); As wherein certain several y
kwhen being 1, just diagnosing it which has and plant fault.
As shown in table 6;
The diagnostic result of table 6 test sample book
Many labels diagnostic method of train air-conditioning fault of the present invention, first be that basic classification algorithm sets up many labelings mechanism with Proximal Support Vector Machine, then Train air conditioning unit actual normally service data and various faults service data is gathered as training sample, learning training obtains multiple single label sub-classifier, finally utilize Train air conditioning unit actual operating data as the input of these single label sub-classifiers, obtain multiple subclassification result vector, namely this result is the solution of many labelings problem.Many labels diagnostic method of train air-conditioning fault of the present invention not only increases the precision of fault diagnosis, and diagnostic result more comprehensively.