CN114243063B - Solid oxide fuel cell system fault positioning method and diagnosis method - Google Patents
Solid oxide fuel cell system fault positioning method and diagnosis method Download PDFInfo
- Publication number
- CN114243063B CN114243063B CN202111555286.6A CN202111555286A CN114243063B CN 114243063 B CN114243063 B CN 114243063B CN 202111555286 A CN202111555286 A CN 202111555286A CN 114243063 B CN114243063 B CN 114243063B
- Authority
- CN
- China
- Prior art keywords
- fault
- fuel cell
- solid oxide
- oxide fuel
- cell system
- 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.)
- Active
Links
- 239000000446 fuel Substances 0.000 title claims abstract description 72
- 239000007787 solid Substances 0.000 title claims abstract description 63
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000003745 diagnosis Methods 0.000 title claims abstract description 49
- 238000012706 support-vector machine Methods 0.000 claims abstract description 32
- 238000000513 principal component analysis Methods 0.000 claims abstract description 15
- 230000015556 catabolic process Effects 0.000 claims abstract description 8
- 238000006731 degradation reaction Methods 0.000 claims abstract description 8
- 230000008859 change Effects 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 23
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 230000009467 reduction Effects 0.000 claims description 11
- 230000015654 memory Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 239000008367 deionised water Substances 0.000 claims description 4
- 229910021641 deionized water Inorganic materials 0.000 claims description 4
- 230000004807 localization Effects 0.000 claims description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Chemical compound O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 abstract description 7
- 230000006870 function Effects 0.000 description 12
- 239000011159 matrix material Substances 0.000 description 9
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 8
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 7
- 229910052799 carbon Inorganic materials 0.000 description 7
- 230000008021 deposition Effects 0.000 description 7
- 239000007789 gas Substances 0.000 description 7
- 208000028659 discharge Diseases 0.000 description 6
- 125000004122 cyclic group Chemical group 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 5
- 230000007787 long-term memory Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 description 4
- 230000036541 health Effects 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 238000002485 combustion reaction Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000005336 cracking Methods 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 239000012528 membrane Substances 0.000 description 2
- 239000003345 natural gas Substances 0.000 description 2
- 229910052757 nitrogen Inorganic materials 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 238000006057 reforming reaction Methods 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003054 catalyst Substances 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 239000000498 cooling water Substances 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000007599 discharging Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000002407 reforming Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
- H01M8/00—Fuel cells; Manufacture thereof
- H01M8/04—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids
- H01M8/04298—Processes for controlling fuel cells or fuel cell systems
- H01M8/04313—Processes for controlling fuel cells or fuel cell systems characterised by the detection or assessment of variables; characterised by the detection or assessment of failure or abnormal function
- H01M8/04664—Failure or abnormal function
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/30—Hydrogen technology
- Y02E60/50—Fuel cells
Landscapes
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- Sustainable Development (AREA)
- Sustainable Energy (AREA)
- Chemical & Material Sciences (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Electrochemistry (AREA)
- General Chemical & Material Sciences (AREA)
- Fuel Cell (AREA)
Abstract
The invention belongs to the technical field of high-temperature fuel cells, and particularly relates to a fault positioning method and a diagnosis method of a solid oxide fuel cell system. And designing a fault diagnosis algorithm combining principal component analysis and a support vector machine through the obtained fault data so as to identify and prevent faults. The invention is a complete SOFC system fault positioning, fault marking and fault diagnosis scheme, solves the problems of difficult resolution and difficult diagnosis of faults caused by large original data volume of the SOFC system and complex system mechanism, and provides a new thought for optimizing the system performance. In addition, in order to prolong the service life of the system, improve the performance of the system, meet the external load requirement, and avoid the degradation of the system performance caused by the change of the operating point by predicting the subsequent performance of the system in advance and making corresponding measures in advance.
Description
Technical Field
The invention belongs to the technical field of high-temperature fuel cells, and particularly relates to a fault positioning method and a diagnosis method for a solid oxide fuel cell system.
Background
The solid oxide fuel cell system (Solid Oxide Fuel Cell, SOFC) is an energy conversion device for directly converting chemical energy of fuel into electric energy, and has the advantages of high efficiency, no pollution, no noise and the like. The power class is from W level to MW level, and mainly relates to the fields of portable power generation, transportation, distributed power generation, fixed power generation and the like. As the tolerance of SOFC systems to reliability and safety degradation becomes lower, fault diagnosis is becoming an integral part of SOFC control systems, mainly comprising three tasks: 1. and (3) detection: detecting an unexpected state occurring in the system; 2. isolation: locating or classifying different faults; 3. analysis or identification: the type, size, or possible cause of the fault is determined.
However, the current fault diagnosis schemes are not perfect enough, and have the following problems: 1. the fault diagnosis scheme based on the model is to establish a nonlinear physical model of the SOFC system to judge whether the system has faults or not. However, model-based methods have some difficulties in ensuring model accuracy: (1) Due to the presence of uncertain parameters and disturbances, a corresponding simplification of the model is necessary, which can lead to uncertainty of the model; (2) The SOFC system is a complex nonlinear model, a large number of coupling relations exist in the system, and the model is difficult to accurately restore the performance of the system; (3) Practical systems are becoming more and more complex, often requiring considerable processing time during model operation, and are not suitable for real-time diagnostics. 2. In most studies, data is generated by a model based on fault diagnosis of the data. These studies face the same problems as model-based schemes. The data generated are not accurate enough, and the diagnostic result obtained is also not accurate enough. The most accurate conclusion can be reached only by using the data obtained during the actual system operation. 3. The current diagnosis schemes only pay attention to the diagnosis link, and few publications pay attention to the problems of fault positioning and marking: most studies produce health data and fault data directly from models. However, during actual system operation, the system often must be run until shutdown to check if a fault has occurred and the type of fault (e.g., gas leakage fault and carbon deposition, which can affect the performance of the system, but not cause a shutdown of the system). And the faults in the actual system are gradual and are not easy to mark, and the correct threshold value needs to be set. In a huge data set, the positioning and marking of SOFC system faults are an important process, and if the data marking is not accurate enough, the fault diagnosis result is also affected.
Disclosure of Invention
In order to overcome the defects and improvement requirements of the prior art, the invention provides a fault positioning method and a fault diagnosis method for a solid oxide fuel cell system, and aims to accurately position and diagnose faults possibly occurring in an SOFC system.
To achieve the above object, according to one aspect of the present invention, there is provided a solid oxide fuel cell system fault locating method including:
Inputting a plurality of state parameter data of a solid oxide fuel cell system to be positioned in a certain time period into a fault positioning model based on a long-short-term memory artificial neural network to obtain a plurality of predicted state parameter data of the solid oxide fuel cell system in the subsequent time period, wherein each state parameter data comprises time information;
And collecting the predicted state parameter data of the solid oxide fuel cell systems, namely the actual state parameter data of the solid oxide fuel cell systems corresponding to the predicted state parameter data in one-to-one mode in time, comparing the predicted state parameter data corresponding to each time with the actual state parameter data, and when the difference value of the predicted state parameter data and the actual state parameter data is larger than a threshold value, enabling the solid oxide fuel cell systems to fail at the time, and completing the fault positioning of the time dimension.
Further, the training data of the fault localization model is selected from a number of state parameter data under healthy operation of the solid oxide fuel cell system.
Further, the phase of healthy operation is selected from the phase of full operating conditions of the solid oxide fuel cell system from system start-up to a relatively steady state entering a discharge peak.
Further, the state parameter data includes: the output voltage, output current, output power, fuel input flow to the combustor, air input flow to the reformer, bypass air flow, deionized water input flow, reformer temperature, heat exchanger temperature, and combustor temperature of the solid oxide fuel cell system.
Further, the method further comprises:
Acquiring a plurality of estimated fault types, acquiring other parameter information required by fault location of each estimated fault type under each fault time, and respectively performing location or type removal on each estimated fault type to obtain the fault type and the fault occurrence position under the fault time, thereby completing the fault location of the solid oxide fuel cell system with the dimension of the fault location.
The invention also provides a fault diagnosis method of the solid oxide fuel cell system, which comprises the following steps:
adopting a multi-classification support vector machine, obtaining the fault type of the solid oxide fuel cell system based on the state parameter data of the solid oxide fuel cell system, and completing fault diagnosis;
The training data for training the support vector machine is obtained by adopting the following method:
Adopting the fault positioning method of the solid oxide fuel cell system to obtain a plurality of state parameter data and fault types corresponding to each state parameter data;
and constructing and obtaining the training data by taking each state parameter data and the corresponding fault type as a training sample.
Furthermore, the state parameter data used for training the support vector machine and for fault diagnosis are subjected to outlier preprocessing and principal component analysis dimension reduction processing.
Further, the precision rate and the recall rate are adopted to evaluate the prediction quality of the support vector machine.
The invention also provides an application of the solid oxide fuel cell system fault location method, which is used for determining measures for preventing degradation according to the predicted system follow-up state parameter data extracted by the solid oxide fuel cell system fault location method so as to avoid degradation of system performance caused by change of an operating point.
The present invention also provides a computer readable storage medium comprising a stored computer program, wherein the computer program, when executed by a processor, controls a device in which the storage medium is located to perform a solid oxide fuel cell system fault localization method as described above and/or a solid oxide fuel cell system fault diagnosis method as described above.
In general, through the above technical solutions conceived by the present invention, the following beneficial effects can be obtained:
(1) The method of the invention firstly collects experimental data of the SOFC system, adopts a long-short-term memory artificial neural network (LSTM) method to evaluate the system performance, judges whether faults occur in the experimental process, and locates and marks the faults. Then, a fault diagnosis algorithm combining Principal Component Analysis (PCA) and a Support Vector Machine (SVM) is designed through the obtained fault data to identify and prevent faults.
(2) The invention provides a complete SOFC system fault positioning, fault marking and fault diagnosis scheme. The invention solves the problems of difficult resolution and diagnosis of faults and the like caused by large raw data volume and complex system mechanism of the SOFC system, and provides a new thought for optimizing the system performance. The invention does not depend on a complex system mechanism model, can locate, mark and diagnose faults only through experimental data, not only can be used for SOFC systems, but also is suitable for a plurality of systems with faults difficult to diagnose directly, such as proton exchange membrane fuel cell systems, rolling bearing systems, gear box systems and the like. The invention provides guidance for the improvement of the SOFC system and the design of the controller.
(3) In order to prolong the service life of the system, improve the performance of the system and meet the requirement of external load, if the subsequent performance of the system can be predicted in advance and corresponding measures can be taken in advance, the degradation of the system performance caused by the change of the operating point can be well avoided.
Drawings
FIG. 1 is a block diagram of a method for locating a fault in a SOFC system according to an embodiment of the present invention;
FIG. 2 is a block diagram of a specific flow of fault locating and marking according to an embodiment of the present invention;
FIG. 3 is a block diagram of a cyclic unit of an LSTM provided by an embodiment of the present invention;
FIG. 4 is a block diagram of a diagnosis flow of a fault diagnosis scheme based on a support vector machine according to an embodiment of the present invention;
Fig. 5 is a schematic diagram of an implementation of an SVM according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
In the prior art, data-driven solid oxide fuel cell system leakage fault diagnosis method is used for collecting data of the whole operation stage of an actual SOFC system, extracting key features representing the operation state of the system after data preprocessing, reducing the variable dimension of the system by combining principal component analysis, inputting the principal component comprehensive factors after dimension reduction into a support vector machine model for training, and establishing a fault diagnosis model, thereby realizing effective detection of two types of leakage faults of the SOFC system. But it has the following disadvantages: the scheme assumes that fault data has been extracted from all experimental data, but the process is actually quite complex. During actual system operation, the system often must be run until shutdown to check if a fault has occurred and the type of fault (e.g., gas leakage fault and carbon deposition, which can affect system performance, but not cause system shutdown). And the faults in the actual system are gradual and are not easy to mark, and the correct threshold value needs to be set. The positioning and marking of SOFC system faults in huge data sets is an important process, and if the data marking is not accurate enough, fault diagnosis results are also affected. Therefore, finding a complete method for positioning, marking and diagnosing faults of a data-driven SOFC system becomes a technical problem to be solved urgently, and the method is specifically as follows.
Example 1
A solid oxide fuel cell system fault location method, as shown in fig. 1, comprising:
Inputting a plurality of state parameter data of a solid oxide fuel cell system to be positioned in a certain time period into a fault positioning model based on a long-short-term memory artificial neural network to obtain a plurality of predicted state parameter data of the solid oxide fuel cell system in the subsequent time period, wherein each state parameter data comprises time information;
And collecting a plurality of solid oxide fuel cell system prediction state parameter data corresponding to the solid oxide fuel cell system actual state parameter data in one-to-one correspondence in time, comparing the prediction state parameter data corresponding to each time with the actual state parameter data, and when the difference value of the prediction state parameter data and the actual state parameter data is larger than a threshold value, enabling the solid oxide fuel cell system to fail at the time, and completing the fault positioning of a time dimension.
Preferably, the state parameter data includes: the output voltage, output current, output power, fuel input flow to the combustor, air input flow to the reformer, bypass air flow, deionized water input flow, reformer temperature, heat exchanger temperature, combustor temperature of the solid oxide fuel cell system.
Specifically, as shown in fig. 2, the full-condition experimental steps of the SOFC system can be divided into: a preparation stage, a discharge stage, a cold standby, a hot standby and an end stage. Wherein,
A. the preparation stage: sufficient amounts of natural gas, nitrogen, deionized water, and cooling water were first prepared and then checked for proper readings from the various sensors in the SOFC system. And then, an auxiliary power supply is connected to the SOFC system, a small amount of natural gas and air are respectively introduced into the anode and the cathode, and whether air leakage exists at the joint of each pipeline is checked. And checking whether the control system and the communication equipment of the SOFC system are normal or not and whether the readings of all the gas meters are normal or not.
B. Discharge phase: firstly, the system is started and slowly heated to 550-600 ℃ to reach the discharge temperature. Then starting to start the electronic load, and gradually pulling up the current. As the current increases, so does the stack temperature and power. After the power is increased to the maximum operating power, the power is reduced to rated power to stably operate, and the electrical characteristics and the thermal characteristics of the system are observed. During the experiment, the supply of raw materials is checked regularly, and whether fuel leakage phenomenon exists or not is checked, so that the operation of the system is ensured to be in an absolute safe environment.
C. cold standby: when the external load does not work for a long time, the temperature of the system is reduced from the high temperature of discharge to normal temperature, and the whole system is powered off to save energy consumption, and the temperature is increased from the normal temperature to the discharge stage when the external load needs to be started.
D. And (3) hot standby: when the external load is deactivated, the SOFC system stops discharging, the current drops to 0, but the system is still in a high-temperature state so as to meet the requirements of the external load at any time.
E. Ending: after the experiment is finished, the electronic load is closed, and then nitrogen is introduced into the system as a shielding gas to protect the galvanic pile. Simultaneously, the air quantity introduced is reduced, so that the SOFC system is gradually cooled. When the temperature drops to room temperature, all power sources of the SOFC system are turned off.
In order to prolong the service life of the system, improve the performance of the system and meet the requirement of external load, if the subsequent performance of the system can be predicted in advance and corresponding measures can be taken in advance, the degradation of the system performance caused by the change of the operating point can be well avoided. And the system can be judged to have failed if the difference is too large by comparing the predicted data with the actual data.
In fault location and marking, the system is started to a relatively steady state of entering a discharge peak at the beginning of the system experiment. The current and voltage data are relatively smooth at this time, and the probability of system failure is low during this time. The initial data of the experiment is therefore labeled as health data (including state parameters and their corresponding time information), and then a long-short term memory artificial neural network (LSTM) predictive model is trained from these health data. Then, predicting a plurality of state parameter data to be fault located through the obtained LSTM model, judging and locating the fault by using the prediction error, and judging that the fault is about to occur once the prediction error exceeds a certain threshold (for example, 1%).
Wherein the LSTM introduces a new internal state based on the conventional cyclic neural networkLinear cyclic information transfer is exclusively performed while external states/>, through hidden layersDelivering nonlinear output information. The internal state c t is calculated by the following formula:
ht＝ot⊙tanh(ct) (2)
Wherein f t∈[0,1]D、it∈[0,1]D and o t∈[0,1]D are three gates controlling information transfer; the "; c t-1 is the memory unit at the previous time; is a candidate state obtained by a nonlinear function, and is expressed as follows:
At each time t, the internal state c t records history information to the current time. The three gates in LSTM are input gate i t, forget gate f t, and output gate o t, respectively. The three doors function as:
(1) The forget gate f t controls how much information the internal state c t-1 of the last time needs to be forgotten.
(2) Input gate i t controls the candidate state at the current timeHow much information needs to be saved.
(3) The output gate o t controls how much information of the internal state c t at the current time needs to be output to the external state h t.
The gating value in LSTM is between (0, 1), indicating that information is allowed to pass in a certain proportion. For example, when f t＝0,it =1, this means that the history information is emptied and the candidate state vector is setAll writes. When f t＝1,it =0, the information at the previous time is completely copied, and no new information is written.
The three gates are calculated in the following ways:
it＝σ(Wixt+Uiht-1+bi) (4)
ft＝σ(Wfxt+Ufht-1+bf) (5)
ot＝σ(Woxt+Uoht-1+bo) (6)
Wherein σ (·) is a Logistic function, its output interval is (0, 1), x t is the input at the current time, and h t-1 is the external state at the previous time.
Fig. 3 shows the structure of the cyclic unit of LSTM, which is calculated as: (1) First, three gates and candidate states are calculated by using the external state h t-1 at the previous time and the input x t at the current time(2) Updating the memory cell c t in combination with the forget gate f t and the input gate i t; (3) The information of the internal state is transferred to the external state h t through the output gate o t.
Through the LSTM circulation unit, the whole network can establish a time sequence dependency relationship with a longer distance. Formulas (1) - (6) may be briefly described as:
ht＝ot⊙tanh(ct) (9)
Wherein the method comprises the steps of For input at the present moment,/>And/>Is a network parameter.
In a conventional recurrent neural network, the hidden state h can be regarded as a short-term memory cell. In LSTM networks, the memory unit c may capture certain key information at a certain moment and may be able to save this key information for a certain time interval. The information stored in the memory unit c is longer than the information stored in the short-term memory unit h, but shorter than the long-term memory, and is therefore called long-term memory.
Preferably, the method further comprises:
Acquiring a plurality of estimated fault types, acquiring other parameter information required by fault location of each estimated fault type under each fault time, and respectively performing location or type removal on each estimated fault type to obtain the fault type and the fault occurrence position under the fault time, thereby completing the fault location of the solid oxide fuel cell system with the dimension of the fault location.
After fault time positioning through the LSTM, the system state is further analyzed through stack parameters, BOP key parameters, combustor temperature parameters and system gas supply flow rate data, and the type of the fault is marked by combining expert knowledge, and the specific flow is shown in figure 2.
For example, cracking of the heat exchanger can result in communication with the outside environment and a drop in temperature. And the subsequent air supply is insufficient, so that excessive heat cannot be taken away, and the temperature of the combustion chamber is increased. The heat exchanger is directly connected with the cathode inlet of the electric pile, and the rupture of the heat exchanger can lead to the pressure reduction or even disappearance of the cathode inlet of the electric pile. And the controller increases the air demand in order to regulate system performance. From this information, it can be determined that a heat exchanger cracking failure has occurred.
The carbon deposition fault is a gradual accumulation process, does not cause severe change of performance, and only gradually reduces the performance of the system. After carbon deposition in the reformer is formed, it is randomly distributed in the reformer. If carbon deposition occurs at the gas path outlet or covers the reforming catalyst, incomplete reforming reactions may result, failing to produce the desired gas required by the stack, and further resulting in reduced SOFC system performance. And due to randomness of carbon deposition, uncertainty of methane reforming reaction can be caused, and frequent jitter of methane flow rate and combustion chamber temperature can be caused. The occurrence of the carbon deposition fault can be judged by the information.
Example two
A solid oxide fuel cell system fault diagnosis method, comprising:
adopting a multi-classification support vector machine, obtaining the fault type of the solid oxide fuel cell system based on the state parameter data of the solid oxide fuel cell system, and completing fault diagnosis;
the training data for training the support vector machine is obtained by adopting the following method:
Adopting the fault positioning method of the solid oxide fuel cell system to obtain a plurality of state parameter data and fault types corresponding to each state parameter data;
and constructing and obtaining the training data by taking each state parameter data and the corresponding fault type as a training sample.
Preferably, the state parameter data for training the support vector machine and for fault diagnosis are subjected to outlier preprocessing and principal component analysis dimension reduction processing.
Preferably, the precision and recall are used to evaluate the predicted quality of the support vector machine.
Specifically, the fault diagnosis model of the Support Vector Machine (SVM) is trained and validated using fault data located by the LSTM method. In the SOFC system experiment, the data volume acquired by the sensor in the cold area and the hot area is large, and the fault diagnosis algorithm is difficult to quickly detect and position faults from multidimensional information. Therefore, the data needs to be subjected to dimension reduction through Principal Component Analysis (PCA), then the dimension-reduced data is divided into two parts, one part is used for training an SVM fault diagnosis model, and the other part is used for verifying the accuracy of the model.
The detailed diagnostic steps can be divided into four steps, data preprocessing, data dimension reduction, model training, and model performance assessment, as shown in fig. 4.
(A) Data preprocessing
In the acquired real data, the data is generally affected by noise. And when the data set is relatively large, it is highly likely to contain data caused by an abnormality. Low quality data can affect the results of the fault diagnosis algorithm and therefore the data is preprocessed before training the model. Firstly, the data are cleaned, missing values are filled in, noise data are smoothed, and outliers are identified and deleted. And then carrying out normalization processing on the data to eliminate the influence of the dimension and numerical difference among different variables.
And (3) data normalization processing:
let m variables in the data, X1, X2,..xm represent each variable. There are N data in each variable, they can be represented by a matrix of N x m:
Then, the matrix X N×m is normalized to generate a standard matrix The normalized formula is:
Wherein i=1, 2 …, N; j=1, 2 …, m; u j and s j are the mean and standard deviation, respectively, of the variable x ij.
(B) Data dimension reduction:
in order to quickly diagnose faults, the data is subjected to dimension reduction. Principal Component Analysis (PCA) is a data dimension reduction method of unsupervised learning, which compresses and denoises data through eigenvalue decomposition. The PCA implementation flow is as follows:
Let the matrix after normalization by (11) be D N×m＝(x(1),x(2),…,x(m)).
1) Calculating a covariance matrix of the matrix D: DD T;
2) Performing eigenvalue decomposition on the matrix DD T;
Analyzing the eigenvalues of the matrix DD T, taking out the eigenvectors (omega 1,ω2,…,ωn′) with the biggest n' eigenvalue relativity, and normalizing the eigenvectors to form an eigenvector matrix W;
3) Converting each sample x (i) to a new sample z (i)＝WTx(i);
4) Finally, the dimension-reduced data set D' n′×m＝(z(1),z(2),…,z(m) is obtained.
(C) Model training
A Support Vector Machine (SVM) is a supervised learning model that is commonly used to address the two classification problem. The multi-classification problem can also be implemented by training multiple classification classifiers. The method has good robustness, is widely used in many tasks, and shows strong advantages.
As shown in FIG. 5, a two-class dataset is assumedWhere y n ε { +1, -1}, if the two classes of samples are linearly separable, then there is one hyperplane:
ωTx+b＝0 (12)；
The two types of samples can be separated, then each sample has y (n)(ωTx(n) +b) >0.
The distance from each sample x (n) in dataset R to the segmentation hyperplane is:
let γ be the shortest distance of all samples in the whole dataset R to the segmentation hyperplane:
The larger γ represents the more stable the division of the two data sets by the hyperplane, and is less susceptible to factors such as noise. The goal of the SVM is to find a hyperplane (ω *,b*) such that γ is the largest, i.e.:
Since scaling ω and b at the same time does not change the distance of the sample to the segmentation hyperplane, where the constraint ω γ=1, then equation (15) is equivalent to:
for a linearly separable dataset, there are multiple segmented hyperplanes, but the most spaced hyperplanes are unique, as shown in FIG. 5, where ω is the weight and b is the bias.
The SVM can solve the problem of nonlinearity by mapping the raw data into a higher dimensional space using a kernel function. For example, in a transformed feature space phi, the decision function of the SVM is:
Where k (x, z) =phi (x) T phi (z) is a kernel function. Common kernel functions are linear kernel functions, radial Basis Functions (RBFs), and polynomial kernel functions. Where RBF kernel functions are typically used to address the non-linear separability problem. The RBF kernel function can be expressed by the following formula:
where σ is the width of the RBF.
M classification problems are realized through SVM, and m class-II classifiers need to be trained. The i-th class data is set as class 1 (positive class), and the labels of all other classes except m-1 i are set as class 2 (negative class), so that one class classifier needs to be trained for each class, and finally, m classifiers are used. For a data x to be classified, the classification label with the highest confidence is generally selected as the classification result.
(D) Model performance assessment
In order to verify the performance of the fault diagnosis algorithm, some index of the quality of the prediction needs to be predicted using an evaluation model. The test data set is brought into the SVM model, and the predicted data is compared with the real label to obtain the precision, recall ratio and comprehensive evaluation index, as shown in Table 2.
Table 2 diagnostic results of fault diagnosis algorithm
Precision of category 1: p=a/(a+d+g), recall for class 1: r=a/(a+b+c), the overall evaluation index F1: 2x p x r/(p+r).
The precision represents the proportion of the correct number of predicted categories to the predicted value for that category, and the recall represents the proportion of positive samples (i.e., recall/true) that the model correctly predicts. The precision and recall are a pair of contradictory indicators, and generally, when the precision is high, the recall is generally very low; at high recall rates, the precision is generally low unless predicted to be very accurate for all categories. In actual model evaluation, the model is not complete by using the precision or recall alone, and two values of precision/recall are needed for evaluating the model. Thus, the comprehensive evaluation index F1 is introduced at the same time, and the F1 metric is defined based on the harmonic mean (halmic mean) of the precision and recall.
Based on the complete SOFC system fault positioning, fault marking and fault diagnosis scheme, the problems of difficult resolution and difficult diagnosis of faults and the like caused by large original data size of the SOFC system and complex system mechanism are solved, and a new thought is provided for optimizing the system performance. The invention does not depend on a complex system mechanism model, can locate, mark and diagnose faults only through experimental data, can be used for an SOFC system, is also suitable for a plurality of systems with faults difficult to diagnose directly, such as a proton exchange membrane fuel cell system, a rolling bearing system, a gear box system and the like, and provides guidance for the improvement of the SOFC system and the design of a controller.
In summary, the invention realizes the positioning of the faults of the solid oxide fuel cell system on one hand, and the key technical means is to design a SOFC system performance prediction model by combining long-short-term memory artificial neural network (LSTM) with experimental data to position and mark the faults. Long-term memory artificial neural networks (LSTM) are a class of neural networks with long-term memory capabilities. It is capable of processing successive data of one sample at a time. In this way, the LSTM may adaptively simulate dynamic information of continuous data across multiple scales. It introduces state variables to store past information, while introducing gating mechanisms to selectively join new information and to selectively forget previous information, together with current inputs to determine current outputs. And LSTM can well solve gradient explosion or disappearance problem in cyclic neural network. Long term memory artificial neural network (LSTM) predictive models are thus trained from health data. And then, predicting the subsequent data through the obtained LSTM model, judging and positioning the fault by using the prediction error, and judging that the fault is about to occur once the prediction error exceeds a certain threshold value.
On the other hand, the invention also realizes the diagnosis of the faults of the solid oxide fuel cell system, and the key technical means is that the main component analysis (PCA) is used for reducing the dimension of the marked experimental data, then the Support Vector Machine (SVM) classifier is trained to obtain a fault diagnosis model, and the faults are diagnosed. The technical means can solve the problems: 1. principal Component Analysis (PCA) is an analysis technique for simplifying data, converting a problem from high dimension to low dimension through linear transformation, retaining low-order principal components, and deleting high-order components to achieve the purpose of reducing the dimension of a data set. The original complex multidimensional data is converted into simple, visual and irrelevant low-dimensional data through dimension reduction, so that the difficulty and complexity of data analysis are effectively reduced. The SVM is a data mining method based on statistical learning theory, can successfully process regression problems, pattern recognition and other problems, and can be popularized and applied to the fields of prediction, comprehensive evaluation and the like. The mechanism of the SVM is to find an optimal classification hyperplane meeting the classification requirement, so that the hyperplane can maximize the blank areas at two sides of the hyperplane while ensuring the classification precision, and therefore, the fault diagnosis in the SOFC system can be realized.
Example III
A computer readable storage medium comprising a stored computer program, wherein the computer program, when run by a processor, controls a device in which the storage medium is located to perform a solid oxide fuel cell system fault localization method as described above and/or a solid oxide fuel cell system fault diagnosis method as described above.
The related technical solutions are the same as the first embodiment and the second embodiment, and are not described herein again.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (6)
1. A solid oxide fuel cell system fault diagnosis method, characterized by comprising:
adopting a multi-classification support vector machine, obtaining the fault type of the solid oxide fuel cell system based on the state parameter data of the solid oxide fuel cell system, and completing fault diagnosis;
The training data for training the support vector machine is obtained by adopting the following method:
adopting a fault positioning method of the solid oxide fuel cell system to obtain a plurality of state parameter data and fault types corresponding to each state parameter data;
Each state parameter data and the corresponding fault type are used as a training sample, and the training data are constructed and obtained;
the fault positioning method of the solid oxide fuel cell system specifically comprises the following steps:
Inputting a plurality of state parameter data of a solid oxide fuel cell system to be positioned in a certain time period into a fault positioning model based on a long-short-term memory artificial neural network to obtain a plurality of predicted state parameter data of the solid oxide fuel cell system in the subsequent time period, wherein each state parameter data comprises time information; the training data of the fault location model is selected from a plurality of state parameter data of the solid oxide fuel cell system under the healthy operation;
Collecting the predicted state parameter data of the solid oxide fuel cell systems, namely the actual state parameter data of the solid oxide fuel cell systems corresponding to the predicted state parameter data of the solid oxide fuel cell systems one by one in time, comparing the predicted state parameter data corresponding to each time with the actual state parameter data, and when the difference value of the predicted state parameter data and the actual state parameter data is larger than a threshold value, enabling the solid oxide fuel cell systems to fail at the time, and completing the fault positioning of a time dimension;
Acquiring a plurality of estimated fault types, acquiring other parameter information required by fault location of each estimated fault type under each fault time, and respectively performing location or type removal on each estimated fault type to obtain the fault type and the fault occurrence position under the fault time, thereby completing the fault location of the solid oxide fuel cell system with the dimension of the fault location;
The state parameter data includes: the output voltage, output current, output power, fuel input flow to the combustor, air input flow to the reformer, bypass air flow, deionized water input flow, reformer temperature, heat exchanger temperature, and combustor temperature of the solid oxide fuel cell system.
2. A solid oxide fuel cell system fault diagnosis method according to claim 1, wherein the phase of healthy operation is selected from the phase of full operation of the solid oxide fuel cell system from system start-up to a relatively steady state entering a discharge peak.
3. The method for diagnosing a fault in a solid oxide fuel cell system according to claim 1, wherein the state parameter data for training the support vector machine and for fault diagnosis are subjected to outlier preprocessing and principal component analysis dimension reduction processing.
4. The method for diagnosing a fault in a solid oxide fuel cell system as claimed in claim 1, wherein the prediction quality of the support vector machine is evaluated using a precision rate and a recall rate.
5. Use of a solid oxide fuel cell system failure diagnosis method according to any one of claims 1 to 4, characterized in that measures for preventing degradation are determined to avoid degradation of system performance due to a change in operating point, based on predicted system follow-up state parameter data extracted by a failure localization method in a solid oxide fuel cell system failure diagnosis method according to any one of claims 1 to 4.
6. A computer readable storage medium, characterized in that the computer readable storage medium comprises a stored computer program, wherein the computer program, when being executed by a processor, controls a device in which the storage medium is located to perform a solid oxide fuel cell system failure diagnosis method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111555286.6A CN114243063B (en) | 2021-12-17 | 2021-12-17 | Solid oxide fuel cell system fault positioning method and diagnosis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111555286.6A CN114243063B (en) | 2021-12-17 | 2021-12-17 | Solid oxide fuel cell system fault positioning method and diagnosis method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114243063A CN114243063A (en) | 2022-03-25 |
| CN114243063B true CN114243063B (en) | 2024-05-14 |
Family
ID=80758594
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111555286.6A Active CN114243063B (en) | 2021-12-17 | 2021-12-17 | Solid oxide fuel cell system fault positioning method and diagnosis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114243063B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117491793B (en) * | 2023-12-29 | 2024-05-10 | 国网浙江省电力有限公司电力科学研究院 | Hydrogen electric coupling system comprehensive performance test method, device and medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107180983A (en) * | 2017-05-16 | 2017-09-19 | 华中科技大学 | A kind of SOFC pile method for diagnosing faults and system |
| CN111157898A (en) * | 2020-01-07 | 2020-05-15 | 清华大学深圳国际研究生院 | New energy vehicle online battery fault detection and analysis method and device |
-
2021
- 2021-12-17 CN CN202111555286.6A patent/CN114243063B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107180983A (en) * | 2017-05-16 | 2017-09-19 | 华中科技大学 | A kind of SOFC pile method for diagnosing faults and system |
| CN111157898A (en) * | 2020-01-07 | 2020-05-15 | 清华大学深圳国际研究生院 | New energy vehicle online battery fault detection and analysis method and device |
Non-Patent Citations (1)
| Title |
|---|
| Data-driven fault diagnosis method for the safe and stable operation of solid oxide fuel cells system;Yi Zheng等;Journal of Power Sources;第490卷;第1-13页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114243063A (en) | 2022-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Data-driven fault diagnosis for PEMFC systems of hybrid tram based on deep learning | |
| Li et al. | Data-driven simultaneous fault diagnosis for solid oxide fuel cell system using multi-label pattern identification | |
| Li et al. | Diagnosis for PEMFC based on magnetic measurements and data-driven approach | |
| CN114899457A (en) | Fault detection method for proton exchange membrane fuel cell system | |
| Ma et al. | Online fault diagnosis for open-cathode PEMFC systems based on output voltage measurements and data-driven method | |
| Peng et al. | Generalized spatial–temporal fault location method for solid oxide fuel cells using LSTM and causal inference | |
| CN114243063B (en) | Solid oxide fuel cell system fault positioning method and diagnosis method | |
| Guo et al. | Deep learning technique for process fault detection and diagnosis in the presence of incomplete data | |
| Pang et al. | Intelligent health states recognition of fuel cell by cell voltage consistency under typical operating parameters | |
| Benouioua et al. | Diagnostic of fuel cell air supply subsystems based on pressure signal records and statistical pattern recognition approach | |
| Guo et al. | Early diagnosis of battery faults through an unsupervised health scoring method for real-world applications | |
| Lu et al. | PEMFC water management fault diagnosis method based on principal component analysis and support vector data description | |
| Klass et al. | Lifelong performance monitoring of PEM fuel cells using machine learning models | |
| Arunthavanathan et al. | Remaining useful life estimation using fault to failure transformation in process systems | |
| Moser et al. | Joint feature and model selection for SVM fault diagnosis in solid oxide fuel cell systems | |
| Dang et al. | A novel diagnosis method of proton exchange membrane fuel cells based on the PCA and XGBoost algorithm | |
| CN112598057B (en) | Coal-fired power plant boiler fault diagnosis method based on FEKNN strategy | |
| Wang et al. | Data-driven fault diagnosis-tolerant control integrated technique for solid oxide fuel cell | |
| Peng et al. | Control-oriented extraction and prediction of key performance features affecting performance variability of solid oxide fuel cell system | |
| Sepulvene et al. | Analysis of machine learning techniques in fault diagnosis of vehicle fleet tracking modules | |
| Yuan et al. | Fault classification of proton exchange membrane fuel cell based on convolutional neural network coupled with support vector machine | |
| Peng et al. | Multistate Reliability Analysis of Solid Oxide Fuel Cells Using Automatic Spectral Clustering and Neighborhood Rough Sets | |
| Petrone et al. | Diagnostics and prognostics for the wind-hydrogen plant | |
| CN118114101B (en) | Fuel cell fault diagnosis method and system | |
| Li et al. | Spatial–temporal synchronous fault feature extraction and diagnosis for proton exchange membrane fuel cell systems |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |