CN110986407A - Fault diagnosis method for centrifugal water chilling unit - Google Patents

Fault diagnosis method for centrifugal water chilling unit Download PDF

Info

Publication number
CN110986407A
CN110986407A CN201911088733.4A CN201911088733A CN110986407A CN 110986407 A CN110986407 A CN 110986407A CN 201911088733 A CN201911088733 A CN 201911088733A CN 110986407 A CN110986407 A CN 110986407A
Authority
CN
China
Prior art keywords
data
fault
sample
samples
water chilling
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201911088733.4A
Other languages
Chinese (zh)
Inventor
潘进
丁强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Dianzi University
Original Assignee
Hangzhou Dianzi University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hangzhou Dianzi University filed Critical Hangzhou Dianzi University
Priority to CN201911088733.4A priority Critical patent/CN110986407A/en
Publication of CN110986407A publication Critical patent/CN110986407A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F25—REFRIGERATION OR COOLING; COMBINED HEATING AND REFRIGERATION SYSTEMS; HEAT PUMP SYSTEMS; MANUFACTURE OR STORAGE OF ICE; LIQUEFACTION SOLIDIFICATION OF GASES
    • F25B—REFRIGERATION MACHINES, PLANTS OR SYSTEMS; COMBINED HEATING AND REFRIGERATION SYSTEMS; HEAT PUMP SYSTEMS
    • F25B1/00—Compression machines, plants or systems with non-reversible cycle
    • F25B1/04—Compression machines, plants or systems with non-reversible cycle with compressor of rotary type
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F25—REFRIGERATION OR COOLING; COMBINED HEATING AND REFRIGERATION SYSTEMS; HEAT PUMP SYSTEMS; MANUFACTURE OR STORAGE OF ICE; LIQUEFACTION SOLIDIFICATION OF GASES
    • F25B—REFRIGERATION MACHINES, PLANTS OR SYSTEMS; COMBINED HEATING AND REFRIGERATION SYSTEMS; HEAT PUMP SYSTEMS
    • F25B49/00—Arrangement or mounting of control or safety devices
    • F25B49/02—Arrangement or mounting of control or safety devices for compression type machines, plants or systems
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M99/00—Subject matter not provided for in other groups of this subclass
    • G01M99/005—Testing of complete machines, e.g. washing-machines or mobile phones
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/15—Correlation function computation including computation of convolution operations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • Thermal Sciences (AREA)
  • Mechanical Engineering (AREA)
  • Computing Systems (AREA)
  • Operations Research (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention discloses a fault diagnosis method for a centrifugal water chilling unit, which comprises the steps of designing a fault simulation scheme, simulating common faults of the centrifugal water chilling unit, and collecting and storing real-time state monitoring data of the unit; screening out steady-state operation data of the centrifugal unit from the collected data, and processing missing values and abnormal values according to a deletion variable method; marking data according to the running state of the unit, calculating the sampling rate of the data according to the quantity difference between different states, and performing over-sampling on training samples according to an LDMOTE algorithm to obtain a new data set with balanced sample quantity; setting the sensitivity weight of different faults; training the new data set by using an XGboost algorithm to obtain a fault diagnosis model; and finally, transmitting the real-time data into a diagnosis model to obtain a fault diagnosis result. The fault diagnosis method for the centrifugal water chilling unit can improve the problem of data imbalance and can improve the recall rate of important faults.

Description

Fault diagnosis method for centrifugal water chilling unit
Technical Field
The invention belongs to the field of refrigeration, and particularly relates to a centrifugal chiller fault diagnosis method based on an MOLAD-XGboost composite model.
Technical Field
With the development of Heating, Ventilation and Air Conditioning (HVAC) technology, the system structure of HVAC equipment is more complex, which brings challenges to HVAC system operation monitoring and fault diagnosis work. The water chilling unit is the most energy-consuming device in the HVAC system, and researches show that timely elimination of water chilling unit faults can effectively reduce 20-50% of energy consumption, so that rapid and accurate judgment on the running state of the water chilling unit is an important basis for guaranteeing safe and stable running of the water chilling unit and saving energy.
Currently, a data-driven fault diagnosis method is the mainstream of research. Aiming at the complexity and uncertainty of the system, machine learning algorithms such as Artificial Neural Networks (ANNs) and Support Vector Machines (SVM) are applied to the diagnosis work of the faults of the water chilling unit, and certain results are obtained. However, in the face of large-scale data learning, the SVM has the defect of insufficient computing power, and is not suitable for training of large sample data. The ANN algorithm has strong robustness and flexibility, and can learn complex nonlinear relations among data, but the neural network searches suboptimal solutions, is easy to fall into local minimum values, can learn a proper model only by massive data, and is not an optimal choice for processing discrete table data classification problems.
The running conditions of the water chilling unit are variable, and in the early stage of a fault, partial state parameters in a normal state and a fault state are highly coupled. Due to the limitation of scale, the small sample data cannot fully contain the running state information of the unit, and the trained model cannot accurately diagnose the running state of the variable-working-condition unit. The fault state data of the water chilling unit is high in acquisition cost and limited in data volume, so that a large amount of data cannot be provided for neural network model learning.
The water chilling unit is generally in a stable normal operation state, the quantity of collected fault data is far less than that of normal data, and the data distribution is unbalanced, so that a model classification result is biased to be normal; the model assumes that the misclassification costs of various types are equal during diagnosis, and in the fault diagnosis problem of the water chilling unit, the misclassification cost of the fault state is higher than that of the normal state, and the misclassification costs of different faults are inconsistent under different requirements.
Disclosure of Invention
The invention provides a fault diagnosis method for a centrifugal water chilling unit, aiming at solving the limitation of SVM and ANN algorithms.
In order to solve the problem of unbalanced sample distribution, the invention provides an oversampling algorithm based on local sample density
(minor Oversampling under Local Area failure, MOLAD), and adopting a cost sensitive learning method and integrating XGboost algorithm to provide a MOLAD-XGboost composite fault diagnosis model, and avoiding model overfitting caused by outliers and noise on the basis of improving the number imbalance among classes; and setting the sensitive weight to improve the problem of value imbalance among classes.
In order to achieve the purpose, the invention is realized by the following technical scheme: a fault diagnosis method for a centrifugal water chilling unit specifically comprises the following steps:
s1, designing a fault simulation scheme, simulating common faults of the centrifugal water chilling unit, collecting and storing real-time state monitoring data of the unit, marking the data according to the running state of the unit, and dividing the data into a normal type and various fault types, wherein the data of the fault types are less than the normal types;
s2, screening the steady-state operation data of the centrifugal unit from the collected data, and processing the missing value and the abnormal value according to a deletion variable method;
s3, calculating the sampling multiplying power N of the data according to the quantity difference among different states, and performing training sample oversampling through an MOLAD algorithm to obtain a new data set with balanced sample quantity;
s4, setting the sensitivity weight of different faults; training the new data set by using an XGboost algorithm to obtain a fault diagnosis model of the centrifugal water chilling unit;
and S6, transmitting the real-time data of the centrifugal water chilling unit into the diagnosis model to obtain a fault diagnosis result.
In step S3, the step of oversampling by the MOLAD algorithm is as follows:
a1, calculating each sample x in the fault classiK order nearest neighbor mean of
Figure BDA0002266225980000022
And arranging all samples into an ascending Array according to the k-order neighbor mean from small to large, wherein the equation form is as follows:
Figure BDA0002266225980000021
a2, determining sampling multiplying power N, and calculating the number M of new samples needing to be generated according to the sampling multiplying power N;
a3, traversing the samples in Array sequentially, calculating the samples and k adjacent x of the sampleskGenerated new sample xnew(ii) a Meanwhile, counting the number of new samples, and stopping the algorithm when the number reaches M; new sample xnewThe generation formula is as follows:
xnew＝xi+rand(0,1)×(xk-xi)；
in the MOLAD algorithm, the nearest neighbor order k needs to be larger than the sampling multiplying factor N, and if k is equal to N, the effect is the same as that of the ordinary synthesis minority oversampling algorithm; if k is less than N, the number of generated samples is less than M, and the number of samples is still unbalanced. If k is far larger than N, the newly generated samples are generated from a few samples with the maximum local density and k-order neighbor samples thereof, so that the model repeatedly learns the samples with high local density, overfitting is easy to occur, and the generalization performance is reduced. In summary, the value of k should be1 to 2 times of the sampling magnification N.
The method for training the fault diagnosis model through the XGboost algorithm in the step S5 comprises the following steps:
XGboost is a lifting tree model, i.e. a tree is used to predict a value initially, then the deviation of the value from the actual value is obtained, and then a tree is added to learn the deviation. After t trees are added:
Figure BDA0002266225980000031
wherein f ist(xi) The discriminant function for the t-th tree for the ith data,
Figure BDA0002266225980000032
for the discrimination result of the strong model integrated by the t decision tree models, η is the learning rate, and the XGboost objective function is defined as follows:
Figure BDA0002266225980000033
the first part is to calculate the difference between the predicted value and the true value, and n is the number of training samples; another part is the regularization term:
Figure BDA0002266225980000034
t is the number of leaf nodes, and omega is the weight of the leaf nodes. Gamma represents a node segmentation threshold, lambda is an L2 regularization coefficient, and the two regularization coefficients are combined to control the complexity of the model to prevent the model from being over-fitted. According to the objective function, the optimal output is solved as follows:
Figure BDA0002266225980000035
Figure BDA0002266225980000036
Figure BDA0002266225980000037
wherein Obj*The method is a scoring function and is used for measuring the quality of a tree structure, and the smaller the value is, the better the tree structure is represented. XGboost adopts a greedy strategy, each feature is traversed, the split Gains of all the features are calculated according to the following formula, and the optimal split Gain is selected to represent Obj before and after splitting*And (3) constructing a basic model by using the maximum Gain characteristic as a splitting point:
Figure BDA0002266225980000038
wherein G isL,GRIs the sum of the first derivatives of the error functions of the split front left child node and right child node, HL,HRThe sum of the second derivatives of the error functions of the split front left and right child nodes, respectively.
Wherein, the setting of the sensitivity weights of different faults in step S5 is:
during training of the XGboost, the weight coefficients of all samples are defaulted to be1, a data set consisting of c-type fault data and normal data is set, the number of each type is i, and the sensitivity weight of the normal data is set to be bnorAnd the sensitivity weights of the c fault data are be1 and be2 … bec respectively, so that the sample weight coefficient matrix w after the sensitivity weights are adjustedbComprises the following steps:
Figure BDA0002266225980000041
the improved objective function obtained by combining the XGboost objective function with the original XGboost objective function is as follows:
Figure BDA0002266225980000042
Figure BDA0002266225980000043
where L represents the prediction loss of n samples constituting the transpose of the vector,
Figure BDA0002266225980000044
represents the prediction loss of the nth sample, wbL represents the loss of the sample prediction multiplied by the sensitivity weight of the sample, allowing the model to learn the sensitivity weight information of the sample.
Has the advantages that:
the XGboost integrates a basic classification model in a lifting mode, so that the state monitoring data of the water chilling unit can be better classified; the MOLAD-XGboost composite model can effectively solve the problem of data imbalance; the cost sensitive weight can effectively improve the recall rate of important faults.
Drawings
FIG. 1 is a flow chart of the present invention
FIG. 2 is a diagram illustrating the lifting principle of the XGboost algorithm in the present invention
FIG. 3 is a graph of the effect of sensitivity weight on classifier performance under evaporator surface fouling failure
The specific implementation mode is as follows:
the technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
The embodiment of the invention provides a technical scheme, and as shown in figure 1, a fault diagnosis method for a centrifugal water chilling unit specifically comprises the following steps:
s1, failure simulation experiment. The centrifugal chiller fault simulation experiment adopts a 90-ton centrifugal chiller, seven faults of condenser scaling (cf), lubricating oil excess (eo), condenser water flow reduction (fwc), evaporator water flow reduction (fwe), non-condensable gas (nc), refrigerant leakage (rl) and refrigerant excess (ro) are simulated, and the specific simulation degree is shown in table 1. 14 real-time state monitoring data shown in a table 2 are acquired, calculated and stored through temperature and pressure sensors, and the data are marked according to eight running states of the unit (a normal state and seven fault states);
s2, in order to select data under a stable working condition to establish a high-efficiency steady-state fault diagnosis model, using three parameters of evaporator side inlet Temperature (TEI), evaporator side outlet Temperature (TEO) and condenser side inlet Temperature (TCI) as judgment parameters for judging whether the operation state of the centrifugal unit reaches a steady state; the method comprises the steps that partial transient data and abnormal value data exist in the original operation monitoring data of the water chilling unit acquired through a sensor and measurement software, and interference is generated on establishment of a correct fault diagnosis model, steady-state analysis is conducted on the stored data in a geometric weighted average mode, abnormal values and null values in the data are deleted, 40000 pieces of normal-state steady-state monitoring data are finally obtained, and 4000 pieces of seven pieces of fault-state steady-state monitoring data are finally obtained.
S3, in order to quantitatively balance the various fault status data and normal status data, a sample oversampling is performed according to the MOLAD algorithm of the present invention:
a1, calculating each sample x in each fault sampleiK order nearest neighbor mean of
Figure BDA0002266225980000051
Arranging all samples into an ascending Array according to the ascending order of k-order neighbor mean values from small to big,
Figure BDA0002266225980000052
the calculation equation is of the form:
Figure BDA0002266225980000053
a2, the number of normal data is 40000, the number of each fault data is 4000, and in order to balance the number of fault data with the number of normal data, the sampling magnification N is 10, that is, each fault data sample is extended by 10 times on the basis of the original number, and the number M of new samples that need to be generated for each sample data in a fault state is 36000.
A3, traversing the samples in Array sequentially, calculating new samples x generated by the samples and k neighbors of the samplesnewWhen the number of generated samples of each fault state reaches M, the generation of new samples is stopped, and finally a new data set with balanced sample number is obtained, wherein the total number of the new data set is 320000.
In the process of executing the MOLAD algorithm, the k value is 1-2 times of N, 10-20 is taken as the k value, the MOLAD algorithm is combined with the XGboost and the comparison group respectively, the neighbor order k is verified, and the average F1 value of a plurality of states is used as an evaluation index to determine the optimal value of k, which is detailed in Table 3. The MOLAD algorithm can effectively consider the spatial distribution information of the samples, so that some samples with smaller local density do not participate in the generation of new samples, the generation range of the samples is biased to a cluster with larger density, the quality of a newly generated sample set is improved, and the abnormal data and high-noise data are prevented from being excessively learned by a model.
S4, the model assumes equal misclassification costs during diagnosis, and in the fault diagnosis problem of the water chilling unit, the misclassification cost in the fault state is higher than that in the normal state, and the misclassification costs of different faults are inconsistent under different requirements. Setting a sensitivity weight b for a fault type sample needing to be adjusted, setting sensitivity weights of other types as 1 by default, increasing the sensitivity weight of the unbalance weight of a certain fault type, improving the fault classification cost of the fault type and increasing the detection probability of the fault. In the following, the sensitivity weights of the common fault of the condenser fouling (cf) are set to be 0, 5, 10, 15, 20 and 25 to train different XGBoost fault diagnosis models so as to verify the effect of the sensitivity weights on improving the condenser fouling (cf) fault recall rate.
S5, setting corresponding super parameters of the XGboost algorithm as shown in FIG. 2, wherein the XGboost algorithm is used for training a new data set to obtain a fault diagnosis model of the centrifugal water chilling unit; the basic learner selected by the XGboost is a classification regression tree, and the loss function adopts the square sum loss:
Figure BDA0002266225980000061
s6, for the trained fault diagnosis model, transmitting real-time data of the centrifugal chiller into the diagnosis model to obtain a fault diagnosis result (fault type), and comparing the diagnosis result with an actual fault type to obtain a confusion matrix of the diagnosis result, thereby defining accuracy, recall rate and F1 as measurement of classification results. In order to uniformly measure the performance of the established different parameter models, a group of data is separated from the original sample and is used for simulating real-time fault diagnosis.
Precision (Precision, P): the index for measuring the accuracy of the model for diagnosing a certain type of fault is defined as follows:
Figure BDA0002266225980000062
recall (Recall, R): the index for measuring the capability of the model to detect a certain type of fault is defined as follows:
Figure BDA0002266225980000063
for a certain class, TP is the number of correct identifications, FP is the number of samples that other classes are misclassified into the class, and FN is the number of samples that the class is misclassified into other classes.
P and R are used for evaluating the recognition capability of the multi-classification model on a certain class, the recognition capability and the recognition capability are mutually restricted, F1 is introduced, the accuracy and the recall rate are compromised, and the higher the F1 is, the better the classification performance of the classifier is.
Figure BDA0002266225980000064
To verify the effectiveness of the proposed method, three control groups were set that could be non-linearly classified: the kernel function is an SVM with a radial basis, a Multi-layer Perceptron (MLP) and a standard XGboost, and training samples of the SVM and MLP models are standardized to improve the performance of the models. The comparison results are shown in table 4.
Table 3 shows the average F1 value of the combination of the MOLAD algorithm and each classification model under different k values, and as can be seen from table 3, the optimal neighbor order k in the MOLAD algorithm is 13, and the diagnosis performance of the MOLAD-XGBoost is better than that of the other composite algorithms.
Table 4 shows the accuracy and the recall rate of the unbalanced samples in different classification methods, and it can be known through comparison that the recall rate of the first three models in the normal state all reaches more than 97%, the accuracy is obviously less than the recall rate, and it indicates that a large amount of fault type data are mistakenly classified into normal types; the other 7 fault types are opposite to normal types in performance on accuracy and recall due to the small number of training samples, and the fact that the quantity imbalance of the samples among the classes has a great influence on the classification performance of the models is proved.
The XGboost model is obviously superior to the first three models in the accuracy and recall rate of 8 state classification, and the XGboost algorithm is more suitable for fault data classification tasks of a water chilling unit.
Compared with the XGboost model, most of the state types of the MOLAD-XGboost model have improved F1, especially the cf state and the rl state; from the change of the whole F1, the model improves the classification capability of the unbalanced data after increasing the unbalanced weight.
As can be seen from fig. 3, the recall rate R of cf faults increases with the sensitivity weight when the sensitivity weight is less than 15, and the recall rate no longer increases when the sensitivity weight is greater than 15; the accuracy rate continues to decrease with increasing sensitivity weight; f1 increases somewhat when the sensitivity weight is small, and F1 starts to decrease when the sensitivity weight is greater than 5. Meanwhile, considering that increasing the sensitivity weight of cf fault can reduce the classification capability of the classifier on other states, a smaller value should be taken when setting the sensitivity weight, and a higher recall rate for fault types with high misjudgment cost is obtained at the cost of small reduction of the performance of the classifier.
TABLE 1 Fault simulation types
Figure BDA0002266225980000071
TABLE 2 characteristic information
Figure BDA0002266225980000072
Figure BDA0002266225980000081
TABLE 3 mean F1 values for MOLAD Algorithm at different k values in combination with various classification models
Figure BDA0002266225980000082
TABLE 4 accuracy and recall for different classification methods for unbalanced samples
Figure BDA0002266225980000083
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and is not intended to limit the practice of the invention to these embodiments. For those skilled in the art to which the invention pertains, several simple deductions or substitutions may be made without departing from the inventive concept, which should be construed as falling within the scope of the present invention.

Claims (2)

1. A fault diagnosis method for a centrifugal water chilling unit is characterized by comprising the following steps:
s1, designing a fault simulation scheme, simulating common faults of the centrifugal water chilling unit, collecting and storing real-time state monitoring data of the unit, marking the data according to the running state of the unit, and dividing the data into a normal type and various fault types, wherein the data of the fault types are less than the normal types;
s2, screening the steady-state operation data of the centrifugal unit from the collected data, and processing the missing value and the abnormal value according to a deletion variable method;
s3, calculating the sampling multiplying power N of the data according to the quantity difference among different states, and performing training sample oversampling through an MOLAD algorithm to obtain a new data set with balanced sample quantity;
the MOLAD algorithm comprises the following steps:
a1, calculating each sample x in the fault classiK order nearest neighbor mean of
Figure FDA0002266225970000011
And arranging all samples into an ascending Array according to the k-order neighbor mean from small to large, wherein the equation form is as follows:
Figure FDA0002266225970000012
a2, determining sampling multiplying power N, and calculating the number M of new samples needing to be generated according to the sampling multiplying power N;
a3, traversing the samples in Array sequentially, calculating the samples and k adjacent x of the sampleskGeneratingNew sample x ofnew(ii) a Meanwhile, counting the number of new samples, and stopping the algorithm when the number reaches M; new sample xnewThe generation formula is as follows:
xnew＝xi+rand(0,1)×(xk-xi)；
s4, setting the sensitivity weight of different faults; training the new data set by using an XGboost algorithm to obtain a fault diagnosis model of the centrifugal water chilling unit;
defaulting that the weight coefficient of all samples is 1, setting a data set consisting of c-type fault data and normal data, wherein the number of each type is i, and setting the sensitivity weight of the normal data as bnorAnd the sensitivity weights of the c-type fault data are be1 and be2 … bec respectively, so that the sample weight coefficient matrix w after the sensitivity weights are adjustedbComprises the following steps:
Figure FDA0002266225970000013
in combination with the XGBoost objective function, the improved objective function is:
Figure FDA0002266225970000021
Figure FDA0002266225970000022
Figure FDA0002266225970000023
where L represents the prediction loss of n samples constituting the transpose of the vector,
Figure FDA0002266225970000024
representing the prediction loss of the nth sample, wherein T is the number of leaf nodes, and omega is the weight of the leaf nodes; gamma denotes the node segmentation threshold, lambda is the L2 regularization coefficient, wbL represents the loss of the sample prediction multiplied by the sensitivity weight of the sample, so thatLetting the model learn the sensitivity weight information of the sample;
and S5, transmitting the real-time data of the centrifugal water chilling unit into the diagnosis model to obtain a fault diagnosis result.
2. The fault diagnosis method for the centrifugal chiller according to claim 1, wherein: the value of k is 1-2 times of the sampling multiplying power N.
CN201911088733.4A 2019-11-08 2019-11-08 Fault diagnosis method for centrifugal water chilling unit Pending CN110986407A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911088733.4A CN110986407A (en) 2019-11-08 2019-11-08 Fault diagnosis method for centrifugal water chilling unit

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911088733.4A CN110986407A (en) 2019-11-08 2019-11-08 Fault diagnosis method for centrifugal water chilling unit

Publications (1)

Publication Number Publication Date
CN110986407A true CN110986407A (en) 2020-04-10

Family

ID=70083556

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911088733.4A Pending CN110986407A (en) 2019-11-08 2019-11-08 Fault diagnosis method for centrifugal water chilling unit

Country Status (1)

Country Link
CN (1) CN110986407A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112782589A (en) * 2021-01-26 2021-05-11 武汉理工大学 Vehicle-mounted fuel cell remote fault classification diagnosis method and device and storage medium
CN113361359A (en) * 2021-05-31 2021-09-07 电子科技大学 XGboost algorithm-based nuclear power pipeline fault identification method
CN114909819A (en) * 2021-02-08 2022-08-16 青岛海尔电冰箱有限公司 System debugging method of semiconductor refrigeration equipment
CN114970710A (en) * 2022-05-20 2022-08-30 武汉理工大学 Mail steamer HVAC fault diagnosis method based on confidence GA-SVM
CN116593198A (en) * 2023-07-17 2023-08-15 蘑菇物联技术（深圳）有限公司 Method, apparatus and medium for diagnosing non-condensable gas faults

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2010025474A (en) * 2008-07-22 2010-02-04 Samsung Electronics Co Ltd Failure diagnostic device used in refrigerating cycle equipment
US20170074566A1 (en) * 2012-06-25 2017-03-16 Whirlpool Corporation Fault detection and diagnosis for refrigerator from compressor sensor
CN107545275A (en) * 2017-07-27 2018-01-05 华南理工大学 The unbalanced data Ensemble classifier method that resampling is merged with cost sensitive learning
CN108921604A (en) * 2018-06-22 2018-11-30 华南理工大学 A kind of ad click rate prediction technique integrated based on Cost-Sensitive Classifiers
CN109540562A (en) * 2018-12-12 2019-03-29 上海理工大学 A kind of water cooler method for diagnosing faults
CN109829236A (en) * 2019-01-31 2019-05-31 浙江工业大学 A kind of Compressor Fault Diagnosis method based on XGBoost feature extraction

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2010025474A (en) * 2008-07-22 2010-02-04 Samsung Electronics Co Ltd Failure diagnostic device used in refrigerating cycle equipment
US20170074566A1 (en) * 2012-06-25 2017-03-16 Whirlpool Corporation Fault detection and diagnosis for refrigerator from compressor sensor
CN107545275A (en) * 2017-07-27 2018-01-05 华南理工大学 The unbalanced data Ensemble classifier method that resampling is merged with cost sensitive learning
CN108921604A (en) * 2018-06-22 2018-11-30 华南理工大学 A kind of ad click rate prediction technique integrated based on Cost-Sensitive Classifiers
CN109540562A (en) * 2018-12-12 2019-03-29 上海理工大学 A kind of water cooler method for diagnosing faults
CN109829236A (en) * 2019-01-31 2019-05-31 浙江工业大学 A kind of Compressor Fault Diagnosis method based on XGBoost feature extraction

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
杨浩、陈红梅: "结合样本局部密度的非平衡数据集成分类算法", 《计算机科学与探索》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112782589A (en) * 2021-01-26 2021-05-11 武汉理工大学 Vehicle-mounted fuel cell remote fault classification diagnosis method and device and storage medium
CN114909819A (en) * 2021-02-08 2022-08-16 青岛海尔电冰箱有限公司 System debugging method of semiconductor refrigeration equipment
CN114909819B (en) * 2021-02-08 2023-06-16 青岛海尔电冰箱有限公司 System debugging method of semiconductor refrigeration equipment
CN113361359A (en) * 2021-05-31 2021-09-07 电子科技大学 XGboost algorithm-based nuclear power pipeline fault identification method
CN114970710A (en) * 2022-05-20 2022-08-30 武汉理工大学 Mail steamer HVAC fault diagnosis method based on confidence GA-SVM
CN116593198A (en) * 2023-07-17 2023-08-15 蘑菇物联技术（深圳）有限公司 Method, apparatus and medium for diagnosing non-condensable gas faults
CN116593198B (en) * 2023-07-17 2023-09-22 蘑菇物联技术（深圳）有限公司 Method, apparatus and medium for diagnosing non-condensable gas faults

Similar Documents

Publication Publication Date Title
CN110986407A (en) Fault diagnosis method for centrifugal water chilling unit
CN110361176B (en) Intelligent fault diagnosis method based on multitask feature sharing neural network
CN113256066B (en) PCA-XGboost-IRF-based job shop real-time scheduling method
CN111444940A (en) Fault diagnosis method for critical parts of fan
JP2021064370A (en) Method and system for semi-supervised deep abnormality detection for large-scale industrial monitoring system based on time-series data utilizing digital twin simulation data
CN107844799B (en) Water chilling unit fault diagnosis method of integrated SVM (support vector machine) mechanism
CN113255848B (en) Water turbine cavitation sound signal identification method based on big data learning
CN110410282A (en) Wind turbines health status on-line monitoring and method for diagnosing faults based on SOM-MQE and SFCM
CN109213127A (en) A kind of HVAC system gradual failure diagnostic method based on deep learning
CN106649919A (en) Method and system for predicting carbon content of fly ash in coal-fired power plant boiler
CN111190349A (en) Method, system and medium for monitoring state and diagnosing fault of ship engine room equipment
CN111160457B (en) Scroll engine fault detection method based on soft-class extreme learning machine
CN114429152A (en) Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption
KR20170127430A (en) Method and system for detecting, classifying and / or mitigating sensor error
CN113157732A (en) Underground scraper fault diagnosis method based on PSO-BP neural network
CN111178594A (en) Thermal power generating unit peak regulation capability prediction method, device and system
Li et al. Across working conditions fault diagnosis for chillers based on IoT intelligent agent with deep learning model
CN114358116A (en) Oil-immersed transformer fault diagnosis method and system and readable storage medium
CN110033181B (en) Power generation equipment state evaluation method based on self-encoder
CN114330486A (en) Power system bad data identification method based on improved Wasserstein GAN
CN110826587B (en) Vortex shaft engine fault detection method based on improved weighted support vector machine
CN117574780A (en) Intelligent thermal power plant load online modeling method and system
CN116204825A (en) Production line equipment fault detection method based on data driving
Lutska et al. Forecasting the efficiency of the control system of the technological object on the basis of neural networks
CN105203327B (en) A kind of gas circuit measurement parameter selection method applied to engine air passage analysis

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20200410

RJ01 Rejection of invention patent application after publication