Fault diagnosis and processing method for fuel cell power generation device

Abstract

The invention discloses a fuel cell power generation device fault diagnosis and processing method, and the method comprises the following steps: training a built neural network algorithm model based on the historical operation data of a fuel cell power generation device in a health state, and obtaining a trained neural network algorithm model; inputting real-time operation parameters of a power generation device control system into the trained neural network algorithm model to obtain predicted single-section voltage standard deviation, total pile voltage, lowest-section voltage change rate and lowest-section voltage change amplitude parameter values; calculating to obtain the standard deviation of the total voltage and the single-section voltage, the total voltage of the galvanic pile, the change rate of the lowest-section voltage and the deviation value of the change amplitude of the lowest-section voltage; carrying out grading and fault judgment on the standard deviation of the total voltage and the single-section voltage, the total voltage of the electric pile, the change rate of the lowest-section voltage and the deviation value of the change amplitude of the lowest-section voltage; and different treatment measures are respectively adopted based on the grading condition to adjust the operation condition of the battery.

Description

Technical Field

[0001] The invention belongs to the technical field of fuel cells and relates to a fault diagnosis and processing method for a fuel cell power generation device. Background Art

[0002] Fuel cells can effectively utilize hydrogen energy and convert chemical energy into electrical energy in an efficient and environmentally friendly manner through electrochemical reactions. Among various types of fuel cells, proton exchange membrane fuel cells (PEMFC) have the characteristics of low noise, zero pollution, no corrosion, long life, etc. They also have the advantages of large output current, low operating temperature, high energy efficiency, fast startup, compact structure, etc., and are widely used. Proton exchange membrane fuel cells are multivariable, strongly coupled nonlinear systems, and many factors have a huge impact on their performance. With the increase in the number of cells and power, more prone to failures and inconsistencies between cells will have a serious impact on their stability and durability, limiting their application. Fault diagnosis of fuel cell systems can provide current health status information for system control, and then perform feedback control to reduce the degree of failure to increase the service life of fuel cells.

[0003] Patent CN201910092227.6 is a battery fault diagnosis method based on vertical and horizontal cross optimization fuzzy BP neural network. This method artificially analyzes common faults of sample data to obtain fault symptoms, and then uses the fault symptoms as input to train the neural network. The cause of the battery failure is used as the output to perform fault diagnosis. This method has done a lot of work in algorithm design, mainly to solve the problem of the algorithm falling into local optimality and improve the accuracy of the algorithm. This method directly establishes a mapping relationship between system state parameters and fault causes, and predefines the faults, so it works well for known common faults. The effect is poor for unknown or unexpected faults.

[0004] Patent CN116184207A defines the fault types as severe membrane dryness, mild membrane dryness, normal state, mild flooding, and severe flooding. The cell consistency, cell voltage deviation value, number of deviations, and anode pressure drop are used as the model input layer to construct an LSTM neural network. This method is suitable for diagnosing fuel cell membrane electrode water management related faults, but cannot diagnose the cause of system faults, and cannot diagnose the causes of flooding and dryness. Summary of the invention

[0005] Existing fuel cell system fault diagnosis technologies mainly include those based on alternating current impedance (EIS) detection, electromagnetic field fault detection, and data-driven model fault diagnosis. The first two require reliance on additional equipment, which increases the difficulty of system integration and reduces the reliability of the entire machine. The general data-driven model adopts an online model, which has high requirements on hardware computing power and is not convenient for application in the system. In order to solve the above problems, the technical solution adopted by the present invention is: a method for fault diagnosis and processing of fuel cell power generation devices, comprising the following steps:

[0006] Obtaining real-time operating parameters of the fuel cell power generation device control system;

[0007] Build a neural network algorithm model;

[0008] Based on the historical operation data of the fuel cell power generation device in a healthy state, the constructed neural network algorithm model is trained to obtain a trained neural network algorithm model;

[0009] The real-time operating parameters of the power generation device control system are input into the trained neural network algorithm model to obtain the predicted single-cell voltage standard deviation, total stack voltage, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values;

[0010] Then, the predicted single-cell voltage standard deviation, total voltage of the stack, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values ​​are compared with the single-cell voltage standard deviation, total voltage of the stack, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values ​​collected in real time, and the deviation values ​​of the total voltage and the single-cell voltage standard deviation, total voltage of the stack, minimum cell voltage change rate, and minimum cell voltage change amplitude are calculated;

[0011] Classify and make fault determination on the deviation values ​​of total voltage and single-cell voltage standard deviation, total voltage of the battery stack, rate of change of the lowest cell voltage, and amplitude of change of the lowest cell voltage;

[0012] According to the influencing factors of the deviation values ​​of the total voltage, standard deviation, minimum voltage change rate and minimum voltage change amplitude, different treatment measures are taken based on the classification situation to adjust the battery operating conditions.

[0013] Furthermore: the operating parameters include stack current, hydrogen reflux flow, circulating water flow, stack inlet temperature, stack outlet temperature, hydrogen side pressure, oxygen side pressure and static drainage pressure difference.

[0014] Furthermore, the neural network algorithm model adopts BP neural network, which adopts a three-layer network structure, including an input layer, an intermediate layer and an output layer. The number of neurons in the intermediate layer is 30-60, the learning rate is 0.2-0.4, and the training iteration termination condition is deviation <1%*data volume*number of neurons. The activation function adopts the commonly used Sigmoid nonlinear function, and the formula is:

[0015]

[0016] Further: the fuel cell includes N fuel cell stacks;

[0017] The calculation formula of the total voltage V of the battery stack is as follows:

[0018] V=V1+V2+……+V N

[0019] Where: V1, V2...V N Respectively represent the voltages of the first battery stack, the second battery stack, ... the Nth battery stack;

[0020] The calculation formula of the minimum voltage change rate CR is as follows:

[0021] CR=(V min (T)-V min (T-1)) / V min (T-1)

[0022] Where: Vmin(T) is the lowest voltage at the current moment, V min (T-1) is the lowest voltage at the previous moment;

[0023] Minimum voltage change amplitude VA

[0024] VA=V average -V min

[0025] Where: V average is the average voltage of the fuel cell, V min is the minimum voltage of the fuel cell.

[0026] Further: The influencing factors of total voltage, standard deviation, minimum voltage change rate and minimum voltage change amplitude are as follows:

[0027] The influencing factors of the total voltage deviation are hydrogen reflux, operating temperature, and operating pressure; the influencing factors of the standard deviation are hydrogen reflux flow, operating temperature, tail discharge volume, and operating pressure; the influencing factors of the minimum voltage change rate are operating temperature and tail discharge volume, and the influencing factors of the minimum voltage change amplitude are operating temperature and tail discharge volume.

[0028] Further: the specific process of grading and fault judgment of the deviation values ​​of the total voltage and single-cell voltage standard deviation, the total voltage of the battery stack, the lowest cell voltage change rate, and the lowest cell voltage change amplitude is as follows:

[0029] Total voltage deviation of the stack E v There are 3 levels:

[0030] First level: 0.015V*number of cells <E v ≤0.02V*number of cells, pressure fault;

[0031] Second level: 0.02V*number of cells <E v ≤0.025V*number of cells, temperature fault;

[0032] Level 3: E v >0.025V*number of cells, hydrogen reflux fault;

[0033] Single cell voltage standard deviation E SD There are 3 levels:

[0034] Level 1: 10 <E SD ≤15, cathode tail row fault;

[0035] Level 2: 15 <E SD ≤20, cathode gas supply failure;

[0036] Level 3: E SD >25, anode reflow failure;

[0037] Minimum voltage change rate deviation E CR There are 3 levels:

[0038] Level 1: 20 <E CR ≤40, tail displacement failure;

[0039] Level 2: 40 <E CR ≤60, temperature fault

[0040] Level 3: E CR >100, hydrogen reflux failure;

[0041] Minimum voltage change amplitude E VA There are 3 levels:

[0042] First level: 0.1V <E VA ≤0.15V, tail displacement fault;

[0043] Second level: 0.15V <E VA ≤0.2V, temperature fault;

[0044] Level 3: E VA >0.2V, hydrogen reflux failure.

[0045] Further: the process of taking different treatment measures based on the classification situation is as follows:

[0046] Total voltage deviation E v The corresponding measures for the first, second and third stages are as follows: the speed of the hydrogen circulation pump is increased by 5%, 10% and 15% respectively, the operating pressure is increased by 3%, 6% and 9% respectively, and the operating temperature is increased by 3%, 6% and 9% respectively;

[0047] Standard Deviation E SD The corresponding measures for the first, second and third stages are as follows: the speed of the hydrogen circulation pump is increased by 5%, 10% and 15% respectively, the cathode tail discharge volume is increased by 20%, 30% and 50% respectively, and the cathode gas supply volume is increased by 5%, 10% and 15% respectively;

[0048] Minimum voltage change rate deviation E CR The corresponding measures for the first, second and third stages are as follows: tail discharge volume is increased by 20%, 30% and 50% respectively, and the operating temperature is increased by 3%, 6% and 9% respectively;

[0049] Minimum voltage change amplitude E VA Corresponding measures are taken for the first, second and third stages as follows: tail discharge volume is increased by 20%, 30% and 50% respectively, and operating temperature is increased by 3%, 6% and 9% respectively.

[0050] The present invention provides a method for diagnosing and processing a fuel cell power generation device fault, which has the following advantages:

[0051] This method is based on the working principle of fuel cells. Based on the historical data of various operating parameters when the device is working, the artificial intelligence algorithm of neural network is used to build a model. This model describes the nonlinear relationship between several operating parameters of the fuel cell device when it is working healthily and several key state parameters of the fuel cell itself. The device uses the operating data collected online for real-time prediction and real-time comparison, reports and classifies the generated deviations, and makes appropriate adjustments to the device operating parameters based on the known working principle of the deviations and the severity of the above-mentioned deviations, so as to try to improve the battery operating conditions. Through this method, the fuel cell power generation device can monitor, identify, process and troubleshoot faults at an early stage, making the device operation more stable, more automated and more intelligent. It cooperates with the three-level fault protection processing and hardware emergency stop protection measures to improve the safety, robustness and reliability of the device operation.

[0052] (1) In the model design, this application uses eight parameters as the model input layer parameters, namely, fuel cell current, hydrogen reflux, circulating water flow, stack inlet temperature, stack outlet temperature, hydrogen side pressure, oxygen side pressure, and static drainage pressure difference; four parameters, namely, the standard deviation of a single cell voltage of the fuel cell, the total cell voltage, the minimum cell voltage change rate, and the minimum cell voltage change amplitude, are selected as the model output layer parameters. These four parameters respectively describe the uniformity, overall performance, stability, and deviation of the battery. The use of these parameters can better describe the state of the fuel cell system and the performance of the fuel cell. This application uses the processed system state parameters as the input layer, and uses the battery state parameters as the output layer, and then uses real-time data to compare with the model output. While diagnosing known faults based on the comparison data, it can also judge unknown faults based on the deviation. It is not necessary to analyze the cause of the fault first, and even when the fault cannot be diagnosed, it can make corresponding processing.

[0053] (2) The present application expands the input layer to system operating parameters such as the pressure on both sides of the fuel cell membrane electrode (hydrogen side and oxygen side), hydrogen reflux flow, stack inlet and outlet temperature, and water permeable plate pressure difference. The output layer introduces the single-cell voltage deviation amplitude and the minimum single-cell voltage change rate. Fault diagnosis is calculated by calculating the deviation between the calculation model and the real-time data, and system-level faults can be diagnosed.

[0054] (3) In fault diagnosis, the present application uses the predicted value calculated online by the offline model to compare with the online monitoring value to obtain the deviation, and judges and processes it according to the deviation classification. If the fault severity is relatively mild, it can be corrected by this method. If the fault is relatively serious, it can also provide an early warning. Fault diagnosis based on battery status can improve the accuracy of fault diagnosis, prevent false alarms caused by factors such as sensors, and improve the reliability of the fuel cell system. This method has fast calculation speed and processing speed. For deviations, using deviation cause analysis fault, graded alarm and processing of faults can better correct system deviations, more stably automatically correct early fuel cell faults, and improve system stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0056] Figure 1 is a flow chart of the method;

[0057] Figure 2 is a flow chart of an embodiment;

[0058] Figure 3 This is a diagram of the neural network model. DETAILED DESCRIPTION

[0059] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0060] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] In terms of system safety protection, the electronic control system is designed with three lines of defense. The first line of defense is fault diagnosis based on data modeling. The second line of defense is software three-level fault alarm and hardware interlocking processing. The last line of defense is hardware manual emergency stop. This method is the first line of defense for system safety. At the same time, in order to solve the above problems and increase the reliability of fault diagnosis, the design scheme is as follows: Process the typical historical operation data of the power generation device control system when it is running healthily in an offline state. The system operation parameters are used as the model input layer, and the battery single-cell performance parameters are used as the output layer. The model is established using the BP neural network algorithm. The model is transplanted into the fuel cell controller (FCU) for online operation, and the data collected in real time by the FCU is input. After the output data is predicted, it is compared with the real-time data collected and processed by the fuel cell CVM to calculate the total deviation and the deviation value of each output parameter. According to the deviation value and the principle of deviation generation combined with expert experience, the cause of the fault is manually analyzed, and corresponding treatment measures are taken according to the severity of the calculated deviation.

[0062] Figure 1 is a flow chart of the method;

[0063] A method for fault diagnosis and treatment of fuel cell power generation device is designed, which includes the following steps:

[0064] S1: Obtaining real-time operating parameters of the fuel cell power generation device control system;

[0065] S2: Build a neural network algorithm model;

[0066] S3: Based on the historical operation data of the fuel cell power generation device in a healthy state, the constructed neural network algorithm model is trained to obtain a trained neural network algorithm model; the model is transplanted into the fuel cell controller (FCU) for online operation, and the data collected in real time by the FCU is input;

[0067] S4: The fuel cell CVM collects and processes the real-time operating parameters of the power generation device control system and inputs them into the trained neural network algorithm model to obtain the predicted single-cell voltage standard deviation, stack total voltage, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values;

[0068] S5: Then compare the predicted single-cell voltage standard deviation, total voltage of the stack, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values ​​with the single-cell voltage standard deviation, total voltage of the stack, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values ​​collected in real time by the fuel cell controller (FCU), and calculate the deviation values ​​of the total voltage and the single-cell voltage standard deviation, total voltage of the stack, minimum cell voltage change rate, and minimum cell voltage change amplitude;

[0069] S6: Calculate the deviation and take corresponding treatment measures according to the deviation value and the cause of the deviation. Classify and determine the fault of the deviation values ​​of the total voltage and single-cell voltage standard deviation, the total voltage of the battery stack, the minimum cell voltage change rate, and the minimum cell voltage change amplitude;

[0070] S7: According to the influencing factors of the total voltage, standard deviation, minimum voltage change rate and minimum voltage change amplitude deviation value, different treatment measures are taken based on the classification situation to adjust the battery operating conditions.

[0071] Execute in order of steps S1 / S2 / S3 / S4 / S5 / S6 / S7; or execute in order of steps S2 / S3 / S1 / S4 / S5 / S6 / S7; steps S1 / S2 / S3 can be completed offline;

[0072] The neural network algorithm model adopts BP neural network, which adopts a three-layer network structure, including an input layer, an intermediate layer and an output layer. The number of neurons in the intermediate layer is 30-60, the learning rate is 0.2-0.4, and the training iteration termination condition is deviation <1%*(data volume*number of neurons). The activation function adopts the commonly used Sigmoid nonlinear function, and the formula is:

[0073]

[0074] Follow the steps below to train the model:

[0075] Step 1: Data processing: firstly, the maximum and minimum values ​​of the selected input and output data are calculated, and then normalized;

[0076] Step 2: Model initialization: initializing each neuron of the BP neural network with random numbers;

[0077] Step 3: Calculate according to the set network parameters and network structure, perform forward propagation, and then use;

[0078] Step 4: Use the gradient descent method to perform back propagation calculations on the model data. Through continuous iterative calculations, the model calculations select weight matrices that fit the training data more and more accurately. Finally, the calculation is exited according to the above iterative termination conditions.

[0079] The embedded software of the fuel cell module controller writes the neural network model matrix obtained by the above calculation into the embedded software of the controller as a model algorithm. The embedded software collects data in real time and then calculates it in a cycle of 5 seconds. The calculation results are compared and calculated with the corresponding battery parameters collected in real time, and the calculation results are processed, and the deviation values ​​of the four parameters are uploaded. Then the deviation values ​​are graded and the corresponding system control parameters are adjusted according to the level for correction.

[0080] In the offline state, historical data under various working conditions when the fuel cell system is running healthily are selected, and the data are as diverse and representative as possible. The operating parameters include stack current, hydrogen reflux, circulating water flow, stack inlet temperature, stack outlet temperature, hydrogen side pressure, oxygen side pressure and static drainage pressure difference. The typical operating parameters of the control system of the above-mentioned power generation device are stack current, hydrogen reflux, circulating water flow, stack inlet temperature, stack outlet temperature, hydrogen side pressure, oxygen side pressure, and static drainage pressure difference. Eight parameters are used as model input parameters, and four parameters related to the voltage of the single cell of the stack, the standard deviation of the voltage of the single cell, the total voltage of the stack, the change rate of the minimum cell voltage, and the change amplitude of the minimum cell voltage are selected as output parameters. This method selects data based on the current density of the fuel cell. Manually operate the system parameters, select the relevant data of the stack at different temperatures, pressures, reflux, and water flow parameters under each electric density, use the host computer for processing and record in real time with a period of 500ms. 200 groups of differential data are selected from the collected data for standby use, the input data uses the historical data collected by the control system, and the output data is calculated based on the single cell voltage collected by CVM.

[0081] The fuel cell comprises N fuel cell stacks;

[0082] The calculation formula of the total voltage V of the battery stack is as follows:

[0083] V=V1+V2+……+V N

[0084] Where: V1, V2...V N Respectively represent the voltages of the first battery stack, the second battery stack, ... the Nth battery stack;

[0085] The standard deviation SD is a conventional calculation method.

[0086] The calculation formula of the minimum voltage change rate CR is as follows:

[0087] CR=(V min (T)-V min (T-1)) / V min (T-1)

[0088] Where: Vmin(T) is the lowest voltage at the current moment, V min (T-1) is the lowest voltage at the previous moment;

[0089] Minimum voltage change amplitude VA

[0090] VA=V average -V min

[0091] Where: V average is the average voltage of the fuel cell, V min is the minimum voltage of the fuel cell.

[0092] Furthermore, the influencing factors of the deviation values ​​of the total voltage, standard deviation, minimum voltage change rate and minimum voltage change amplitude are as follows:

[0093] The influencing factors of the total voltage deviation are hydrogen reflux, operating temperature, and operating pressure; the influencing factors of the standard deviation are hydrogen reflux flow, operating temperature, tail discharge volume, and operating pressure; the influencing factors of the minimum voltage change rate are operating temperature and tail discharge volume, and the influencing factors of the minimum voltage change amplitude are operating temperature and tail discharge volume.

[0094] Fault diagnosis method, the output layer parameters inferred by the model obtained by the above calculation, namely, the standard deviation of the single-cell voltage of the fuel cell, the total voltage of the stack, the minimum cell voltage change rate, and the minimum cell voltage change amplitude. Use this data to compare and calculate the corresponding parameters calculated by real-time acquisition. The deviations of each value are graded, and different levels represent different faults.

[0095] Furthermore, the fault output layer deviation defines the fault type, and the specific process of grading the deviation values ​​of the total voltage and single-cell voltage standard deviation, the total voltage of the battery stack, the lowest cell voltage change rate, and the lowest cell voltage change amplitude is as follows:

[0096] Total voltage deviation of the stack E v There are 3 levels:

[0097] First level: 0.015V*number of cells <Ev ≤0.02V*number of cells, pressure fault;

[0098] Second level: 0.02V*number of cells <E v ≤0.025V*number of cells, temperature fault;

[0099] Level 3: E v >0.025V*number of cells, hydrogen reflux fault;

[0100] Single cell voltage standard deviation E SD There are 3 levels:

[0101] Level 1: 10 <E SD ≤15, cathode tail row fault;

[0102] Level 2: 15 <E SD ≤20, cathode gas supply failure;

[0103] Level 3: E SD >25, anode reflow failure;

[0104] Minimum voltage change rate deviation E CR There are 3 levels:

[0105] Level 1: 20 <E CR ≤40, tail displacement failure;

[0106] Level 2: 40 <E CR ≤60, temperature fault

[0107] Level 3: E CR >100, hydrogen reflux failure;

[0108] Minimum voltage change amplitude E VA There are 3 levels:

[0109] First level: 0.1V <E VA ≤0.15V, tail displacement fault;

[0110] Second level: 0.15V <E VA ≤0.2V, temperature fault;

[0111] Level 3: E VA >0.2V, hydrogen reflux failure.

[0112] Furthermore, the process of taking different treatment measures based on the classification situation is as follows:

[0113] Total voltage deviation Total voltage E vThe corresponding measures for the first, second and third stages are as follows: the speed of the hydrogen circulation pump is increased by 5%, 10% and 15% respectively, the operating pressure is increased by 3%, 6% and 9% respectively, and the operating temperature is increased by 3%, 6% and 9% respectively;

[0114] Standard Deviation E SD The corresponding measures for the first, second and third stages are as follows: the speed of the hydrogen circulation pump is increased by 5%, 10% and 15% respectively, the cathode tail discharge volume is increased by 20%, 30% and 50% respectively, and the cathode gas supply volume is increased by 5%, 10% and 15% respectively;

[0115] Minimum voltage change rate deviation E CR The corresponding measures for the first, second and third stages are as follows: tail discharge volume is increased by 20%, 30% and 50% respectively, and the operating temperature is increased by 3%, 6% and 9% respectively;

[0116] Minimum voltage change amplitude E VA Corresponding measures are taken for the first, second and third stages as follows: tail discharge volume is increased by 20%, 30% and 50% respectively, and operating temperature is increased by 3%, 6% and 9% respectively.

[0117] This method mainly takes the single-cell voltage of the fuel cell stack and the parameters calculated based on this as the basis for judgment. The three-level fault protection is directly judged according to the real-time collected data. The hardware emergency stop protection does not rely on the controller. Switch components are arranged in key gas and circuit parts. Emergency stop directly cuts off the gas supply and power output to ensure basic device safety.

[0118] Embodiment 1:

[0119] Figure 2 is a flow chart of an embodiment;

[0120] A method for diagnosing and processing a fault of a fuel cell power generation device comprises the following steps:

[0121] Offline:

[0122] S11: Build a neural network algorithm model;

[0123] S12: Select data under the healthy operation status of the fuel cell stack and the system, and various input and output parameters should be scattered as much as possible. Based on the historical operation data of the fuel cell power generation device under the healthy state, train the built neural network algorithm model to obtain a trained neural network algorithm model; transplant the model into the fuel cell controller (FCU) for online operation, and input the data collected in real time by the FCU;

[0124] When online:

[0125] S21: Online real-time monitoring of input layer parameters: obtaining real-time operating parameters of the fuel cell power generation device control system;

[0126] S23: Model prediction value: input the real-time operating parameters of the fuel cell CVM collection and processing power generation device control system into the trained neural network algorithm model to obtain the predicted single-cell voltage standard deviation, stack total voltage, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values;

[0127] S33: The system monitors the output layer parameters online in real time: obtaining the single-cell voltage standard deviation, the total voltage of the stack, the minimum cell voltage change rate, and the minimum cell voltage change amplitude parameter values ​​collected in real time by the fuel cell controller (FCU);

[0128] S34: Data comparison: The predicted single-cell voltage standard deviation, total stack voltage, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values ​​are compared with the single-cell voltage standard deviation, total stack voltage, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values ​​collected in real time by the fuel cell controller (FCU), and the deviation values ​​of the total voltage and the single-cell voltage standard deviation, total stack voltage, minimum cell voltage change rate, and minimum cell voltage change amplitude are calculated;

[0129] S35: Fault determination: According to the deviation value and the cause of the deviation, calculate the deviation and take corresponding treatment measures. Classify and determine the fault of the deviation values ​​of the total voltage and single-cell voltage standard deviation, the total voltage of the battery stack, the minimum cell voltage change rate, and the minimum cell voltage change amplitude;

[0130] S36: Fault handling; according to the influencing factors of the total voltage, standard deviation, minimum voltage change rate and minimum voltage change amplitude deviation value, different treatment measures are taken based on the classification situation to adjust the battery operating conditions.

[0131] Figure 3 It is a modeling structure diagram of the neural network algorithm model.

[0132] The input layer parameters of the neural network algorithm model include: stack current, hydrogen reflux, circulating water flow, stack inlet temperature, stack outlet temperature, hydrogen side pressure, oxygen side pressure and static drainage pressure difference;

[0133] The output layer parameters include: the total voltage collected by the fuel cell inspection, the standard deviation of the single-cell voltage collected by the fuel cell inspection, the rate of change of the single-cell minimum voltage collected by the fuel cell inspection, and the deviation of the single-cell minimum voltage collected by the fuel cell inspection.

[0134] The total voltage collected by the fuel cell inspection explains the overall performance of the fuel cell;

[0135] The standard deviation of the single-cell voltage collected during the fuel cell inspection is an explanation of the fuel cell homogeneity;

[0136] The lowest node voltage change rate is an explanation for the fuel cell stability;

[0137] The deviation value of the minimum cell voltage change amplitude is an explanation of the deviation of the single cell voltage of the battery;

[0138] In the safety protection design of the 16-20kW hydrogen-oxygen fuel cell module system, the stack has 120 sections, and the electronic control system is designed with this method based on data modeling fault diagnosis, software three-level fault alarm and hardware interlocking processing and hardware manual emergency stop. As can be seen from the figure, in an offline state, the above-mentioned fuel cell power generation device is operated to screen and process data from long-term normal operation data. In the selection of data items, the following 8 columns are extracted as input layer data, namely fuel cell current, hydrogen reflux, circulating water flow, stack inlet temperature, stack outlet temperature, hydrogen side pressure, oxygen side pressure, and static drainage pressure difference. The single-cell voltage data of the battery recorded by CVM is selected, and the four columns of data of the single-cell voltage standard deviation of the fuel cell, the total battery voltage, the minimum cell voltage change rate, and the minimum cell voltage change amplitude are obtained after processing and calculation as output layer data. In data screening, the changes of each parameter are evenly distributed as much as possible to ensure the representativeness of the sample data. The model is established using the BP neural network algorithm. The BP neural network adopts a basic three-layer network structure, with 30-60 neurons in the middle layer, a learning rate of 0.2-0.4, and a training iteration termination condition of deviation <1%*(data volume*number of neurons). The activation function adopts the commonly used Sigmoid nonlinear function, and the corresponding weight matrix is ​​trained through iterative operations.

[0139] The weight matrix is ​​transplanted into the fuel cell controller (FCU) for online operation, and the data collected in real time by the FCU is input. After the output data is predicted, it is compared with the real-time data collected and processed by the fuel cell CVM to calculate the total deviation and the deviation value of each output parameter. The total voltage of the device stack is predicted to be 100, the standard deviation is 20, the minimum voltage change rate is 14, and the minimum voltage change amplitude is 16. After comparing with the online monitoring data, the operation deviation at a certain moment is 0.97 for the total voltage deviation, 2.91 for the standard deviation, 2.58 for the minimum voltage change rate deviation, 27.47 for the minimum voltage amplitude deviation, and 33.95 for the total deviation.

[0140] The deviation values ​​are graded and reported to the host computer. The fault items are analyzed according to the factors affecting the deviation values, and the fault analysis results are reported to the host computer. The system software takes corresponding treatment measures for the relevant influencing factors according to the severity of the calculated deviation, and adjusts the control parameters through corresponding targeted means. According to the method in this article, the total voltage deviation belongs to the first level, the hydrogen pump speed increases by 5%, the operating pressure increases by 3%, and the operating temperature increases by 3%. The standard deviation belongs to the second level, the hydrogen pump speed increases by 10%, the oxygen operating pressure increases by 10%, and the exhaust volume increases by 30%. The minimum voltage change rate does not need to be adjusted if it does not reach the lower limit of adjustment, and the minimum voltage change amplitude belongs to the first level. The tail exhaust volume increases by 20% and the operating temperature increases by 3%. Following the above calculation results, the upper limit adjustment scheme is selected for each operating parameter adjustment, the hydrogen pump speed is increased from 1500rpm to 1650rpm, the hydrogen side pressure is increased from 1.5bar to 1.575bar, the oxygen side pressure is increased from 1.5bar to 1.65bar, the temperature setting is increased from 55℃ to 57.7℃, and the exhaust volume is increased from 200ml / min to 260ml / min. In this way, the battery fault is automatically corrected. If the fault is not serious, it can be corrected by this method. If the fault is serious, it can also provide an early warning.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing and handling faults of a fuel cell power generation device, characterized in that: The following steps are involved: Obtaining real-time operating parameters of the fuel cell power generation device control system; Build a neural network algorithm model; Based on the historical operation data of the fuel cell power generation device in a healthy state, the constructed neural network algorithm model is trained to obtain a trained neural network algorithm model; The real-time operating parameters of the power generation device control system are input into the trained neural network algorithm model to obtain the predicted single-cell voltage standard deviation, total stack voltage, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values; Then, the predicted single-cell voltage standard deviation, total voltage of the stack, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values ​​are compared with the single-cell voltage standard deviation, total voltage of the stack, minimum cell voltage change rate, and minimum cell voltage change amplitude parameter values ​​collected in real time, and the deviation values ​​of the total voltage and the single-cell voltage standard deviation, total voltage of the stack, minimum cell voltage change rate, and minimum cell voltage change amplitude are calculated; Classify and make fault determination on the deviation values ​​of total voltage and single-cell voltage standard deviation, total voltage of the battery stack, rate of change of the lowest cell voltage, and amplitude of change of the lowest cell voltage; According to the influencing factors of the deviation values ​​of the total voltage, standard deviation, minimum voltage change rate and minimum voltage change amplitude, different treatment measures are taken based on the classification situation to adjust the battery operating conditions.

2. A method for diagnosing and processing a fuel cell power generation device fault according to claim 1, characterized in that: The operating parameters include stack current, hydrogen reflux flow, circulating water flow, stack inlet temperature, stack outlet temperature, hydrogen side pressure, oxygen side pressure and static drainage pressure difference.

3. A method for diagnosing and processing a fuel cell power generation device fault according to claim 1, characterized in that: The neural network algorithm model adopts BP neural network, which adopts a three-layer network structure, including an input layer, an intermediate layer and an output layer. The number of neurons in the intermediate layer is 30-60, the learning rate is 0.2-0.4, and the training iteration termination condition is deviation <1%*data volume*number of neurons. The activation function adopts the commonly used Sigmoid nonlinear function, and the formula is:

4. A method for diagnosing and processing a fuel cell power generation device fault according to claim 1, characterized in that: The fuel cell comprises N fuel cell stacks; The calculation formula of the total voltage V of the battery stack is as follows: V=V1+V2+……+V N Where: V1, V2, ... V N Respectively represent the voltages of the first battery stack, the second battery stack, ... the Nth battery stack; The calculation formula of the minimum voltage change rate CR is as follows: CR=(V min (T)-V min (T-1)) / V min (T-1) Where: Vmin(T) is the lowest voltage at the current moment, V min (T-1) is the lowest voltage at the previous moment; Minimum voltage change amplitude VA AND=V average -V min Where: V average is the average voltage of the fuel cell, V min is the minimum voltage of the fuel cell.

5. A method for diagnosing and processing a fuel cell power generation device fault according to claim 1, characterized in that: The influencing factors of total voltage, standard deviation, minimum voltage change rate and minimum voltage change amplitude are as follows: The influencing factors of the total voltage deviation are hydrogen reflux, operating temperature, and operating pressure; the influencing factors of the standard deviation are hydrogen reflux flow, operating temperature, tail discharge volume, and operating pressure; the influencing factors of the minimum voltage change rate are operating temperature and tail discharge volume, and the influencing factors of the minimum voltage change amplitude are operating temperature and tail discharge volume.

6. A method for diagnosing and processing a fuel cell power generation device fault according to claim 1, characterized in that: The specific process of grading and fault judgment of the deviation values ​​of the total voltage and single-cell voltage standard deviation, the total voltage of the battery stack, the lowest cell voltage change rate, and the lowest cell voltage change amplitude is as follows: Total voltage deviation of the stack E v There are 3 levels: First level: 0.015V*number of cells <E v ≤0.02V*number of cells, pressure fault; Second level: 0.02V*number of cells <E v ≤0.025V*number of cells, temperature fault; Level 3: E v >0.025V*number of cells, hydrogen reflux fault; Single cell voltage standard deviation E SD There are 3 levels: Level 1: 10 <E SD ≤15, cathode tail row fault; Level 2: 15 <E SD ≤20, cathode gas supply failure; Level 3: E SD >25, anode reflow failure; Minimum voltage change rate deviation E CR There are 3 levels: Level 1: 20 <E CR ≤40, tail displacement failure; Level 2: 40 <E CR ≤60, temperature fault Level 3: E CR >100, hydrogen reflux failure; Minimum voltage change amplitude E VA There are 3 levels: First level: 0.1V <E VA ≤0.15V, tail displacement fault; Second level: 0.15V <E VA ≤0.2V, temperature fault; Level 3: E VA >0.2V, hydrogen reflux failure.

7. A method for diagnosing and processing a fuel cell power generation device fault according to claim 1, characterized in that: The process of taking different treatment measures based on the classification situation is as follows: Total voltage deviation E v The corresponding measures for the first, second and third stages are as follows: the speed of the hydrogen circulation pump is increased by 5%, 10% and 15% respectively, the operating pressure is increased by 3%, 6% and 9% respectively, and the operating temperature is increased by 3%, 6% and 9% respectively; Standard Deviation E SD The corresponding measures for the first, second and third stages are as follows: the speed of the hydrogen circulation pump is increased by 5%, 10% and 15% respectively, the cathode tail discharge volume is increased by 20%, 30% and 50% respectively, and the cathode gas supply volume is increased by 5%, 10% and 15% respectively; Minimum voltage change rate deviation E CR The corresponding measures for the first, second and third stages are as follows: tail discharge volume is increased by 20%, 30% and 50% respectively, and the operating temperature is increased by 3%, 6% and 9% respectively; Minimum voltage change amplitude E VA Corresponding measures are taken for the first, second and third stages as follows: tail discharge volume is increased by 20%, 30% and 50% respectively, and operating temperature is increased by 3%, 6% and 9% respectively.

Images