A method, system, device and terminal for predicting the remaining useful life of a fuel cell

Abstract

The present application belongs to the technical field of fuel cell, and discloses a kind of fuel cell remaining service life prediction method, system, equipment and terminal, and a plurality of historical detection signals of fuel cell are stored by receiving command from LabView interface and signal obtained by sensor and being received by control system, obtain fuel cell output voltage signal and the battery polarization curve of each time period;The obtained signal is preprocessed, and the voltage signal is changed into time series;Unreasonable signal in time series and polarization curve is eliminated, and final voltage sequence and multiple polarization curves are obtained;Then, a fuel cell aging model is established, and the initial parameters are identified by inputting the polarization curve into the genetic algorithm;The method combining EMD and particle filtering is constructed, and the remaining sequence is predicted by combining the aging model and the particle filtering method.The present application can accurately predict the online remaining life of the fuel cell working for a long period of time, ensuring the long-term safety and stability of the system.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of fuel cells, and particularly relates to a fuel cell residual service life prediction method, system, device and terminal. BACKGROUND

[0002] At present, fuel cells are increasingly applied in vehicle-mounted and airborne power systems due to their high energy density, cleanliness and other advantages, and the durability of fuel cells is an important factor for their large-scale commercialization. In recent years, the durability of PEMFC systems has attracted more and more attention. Hydrogen fuel proton exchange membrane fuel cells are prone to irreversible degradation in their storage and operation modes due to their own properties, resulting in accelerated performance loss and shortened life. Therefore, the prediction and health management (PHM) technology of fuel cells is crucial. By collecting system data information, the state of the fuel cell system is evaluated, faults are diagnosed, and the residual life is predicted. The output voltage of the fuel cell is usually taken as the system health index (HI) for life prediction, and the control strategy is adjusted to improve the durability of the fuel cell system.

[0003] Currently, the methods for prediction mainly include three categories: data-driven based, model-based and fusion algorithm based. The data-driven based prediction method needs to use a large amount of data to complete the life prediction of the fuel cell, and it does not depend on the internal degradation model of the fuel cell. The prediction result is relatively accurate, but the required data amount is large and the operation speed is relatively slow. The model-based algorithm generally predicts by combining the fuel cell aging model with the filtering algorithm, but this method needs the degradation model of the fuel cell. At present, the aging mechanism inside the fuel cell is not completely clear, and the accuracy of the equivalent circuit model, empirical model and semi-empirical model used is low, which brings great challenges to the model-based prediction. The fusion-based algorithm can combine the advantages of the data-driven prediction method and the model-based prediction method, and is the main research direction at present. Document CN114551945A establishes a decay model based on the decay behavior of the internal components of the proton exchange membrane fuel cell and constructs a particle filtering algorithm framework for fuel cell state evaluation according to the decay model. The battery life is predicted by iteratively predicting the decay state and decay rate, but the uncertainty of this method is large and the accuracy is low. Document CN109683093A proposes a data-driven prediction method, which trains a convolutional neural network using historical signals, and then uses the trained neural network model for prediction. However, the effect of this method on long-term RUL prediction is not ideal. Document CN114371416A proposes a prediction algorithm combining empirical mode decomposition and echo neural network, which can realize accurate short-term multi-step prediction, but this method cannot well perform long-term life prediction and the time cost will increase exponentially for each component, and cannot well perform online prediction.

[0004] Through the above analysis, the problems and defects of the prior art are:

[0005] (1) The existing data-driven based fuel cell life prediction method does not rely on the internal degradation model of the fuel cell, the prediction result is relatively accurate but the required data amount is large and the operation speed is relatively slow.

[0006] (2) The existing model-driven prediction method needs the degradation model of the fuel cell, and the internal aging mechanism of the fuel cell is not completely clear at present, so the model used has low accuracy, which brings great challenge to the model-based prediction.

[0007] (3) The existing fuel cell aging prediction method using empirical mode decomposition is mostly used for short-term prediction, and the effect on long-term RUL prediction is not ideal, the uncertainty is large and the accuracy is low; at the same time, the time cost will be doubled for predicting each component, and online prediction cannot be well performed. SUMMARY

[0008] In view of the problems existing in the prior art, the present application provides a fuel cell remaining useful life prediction method, system, device and terminal, especially a fuel cell remaining useful life prediction method, system, medium, device and terminal combining empirical mode decomposition and particle filtering.

[0009] The present application is realized in this way, a fuel cell remaining useful life prediction method, the fuel cell remaining useful life prediction method includes training phase and prediction phase, through the control system receives the command from LabView interface and the signal obtained by the sensor installed on the auxiliary system and electronic load and stores the fuel cell multiple historical detection signals, obtains the fuel cell output voltage signal and the battery polarization curve of each time period;The obtained signal is pretreated, and the voltage signal is changed into time sequence;The unreasonable signal caused by sampling error in time sequence and polarization curve is eliminated, and the final voltage sequence and multiple polarization curves are obtained;Then the fuel cell aging model is established, the processed polarization curve is input into the genetic algorithm to identify the initial parameters;The method combining EMD and particle filtering is constructed, and the remaining sequence processed by EMD is predicted by combining the aging model and the particle filtering method.

[0010] Further, the fuel cell remaining useful life prediction method includes the following steps:

[0011] Step one, start the aging test, receive the command from LabView interface and the signal obtained by the sensor installed on the auxiliary system and electronic load through the control system; the control system sends control signal, displays the signal obtained by the sensor on LabView interface and stores it in the host computer, and records the fuel cell parameters;

[0012] Step two, extract the historical detection signal obtained in step one, obtain the fuel cell output voltage signal and the cell polarization curve in each time period;

[0013] Step three, pretreat the output voltage signal obtained in step two, change the voltage signal into hourly time sequence; eliminate the unreasonable signal in time sequence and polarization curve caused by sampling error, and obtain the final voltage sequence and multiple polarization curves;

[0014] Step four, establish a semi-empirical aging model based on the aging behavior inside the proton exchange membrane fuel cell;

[0015] Step five, identify the aging model parameters at the initial time by fitting the polarization curve obtained in step three through genetic algorithm;

[0016] Step six, use empirical mode decomposition method to perform intrinsic mode decomposition on the voltage sequence in step three, and obtain a series of intrinsic mode components and residual sequence;

[0017] Step seven, establish a particle filtering algorithm framework according to the aging model;

[0018] Step eight, input the residual sequence obtained in step six into the particle filtering algorithm framework constructed to update the model state parameters, including the internal resistance increase rate α, the limiting current density decay rate β, and the exchange current density decay rate γ;

[0019] Step nine, no longer update the weight and resample the particle filtering, and input the final model state parameters into the aging model to obtain the predicted voltage of the fuel cell;

[0020] Step ten, set the fuel cell voltage threshold, record the time when the predicted voltage reaches the threshold, and calculate the remaining service life of the fuel cell.

[0021] Further, the signal obtained by the sensor in step one includes fuel cell voltage, load current, working temperature, inlet and outlet water temperature, air flow and hydrogen flow, and the fuel cell parameters include output voltage, current and aging time.

[0022] In step three, the collected voltage is changed into time sequence according to the following formula:

[0023] V time (n)=V measure (n+1);

[0024] wherein V time is the output voltage sequence, V measure is the measured voltage signal, n is the current time point, and n+1 is the next time point.

[0025] The voltage screening is performed according to the following formula:

[0026]

[0027] wherein V final is the final voltage, V usizer is the voltage to be screened, n is the current time point, n-1 is the previous time point, and β is the tolerance coefficient.

[0028] The semi-empirical aging model of the proton exchange membrane fuel cell established in step four is:

[0029]

[0030] wherein V s is the output voltage of the stack, i s is the current of the stack, N is the number of single cells, E ocv is the open circuit voltage, R0 is the initial value of the ohmic resistance, α is the rate of increase of the internal resistance, a and b are constants, T is the operating temperature, t is the time, i0 is the initial value of the exchange current density, β is the decay rate of the exchange current density, i L0 is the initial value of the limiting current density, and γ is the decay rate of the limiting current density.

[0031] Further, the initial polarization curve is fitted according to the following genetic algorithm in step five, and the initial parameters of the aging model are identified:

[0032] The model is initialized, the population number of the genetic algorithm is selected to be 50, and then an initial population is created at random, the crossover rate is set to be 0.85, and the mutation rate is set to be 0.12; the voltage V s at the kth moment is calculated by using the created aging model and the parameters at the kth moment; the fitness of the population is calculated by using the model, and the selection, crossover, and mutation operations are performed on the variables at the kth moment; the operations are repeatedly performed, and it is judged whether the termination condition is met; the optimal solution is selected according to the fitness function, and the fitting result is obtained.

[0033] The fitness of the genetic function is defined as:

[0034]

[0035] wherein f fit is the fitness, k is the kth sampling point of the polarization curve, V sk is the output voltage of the stack at the kth moment, V is the voltage at the kth moment fitted, and Eocv is the open circuit voltage, R0 is the initial ohmic resistance, i0 is the initial exchange current density, and i L0 is the initial limiting current density.

[0036] Objective function:

[0037] min f fit (E ocv , R0, i0, i L0 );

[0038] where E ocv , R0, i0, i L0 are the optimized variables, and the parameter identification problem is converted into an optimization problem of minimizing the objective function.

[0039] Further, the voltage sequence in step six is empirically mode decomposed by the following process:

[0040] (1) Identify all local extrema of the time series V(t), including local minima and local maxima;

[0041] (2) Fit all local extrema to form the envelope line of the original data using numerical analysis method, and the upper envelope line is vu(t) and the lower envelope line is vl(t);

[0042] (3) Calculate the average envelope line of the upper and lower envelope lines:

[0043] n(t) = [vu(t) + vl(t)] / 2;

[0044] (4) Subtract the average envelope n(t) from the original data sequence V(t) to obtain a new data sequence h(t);

[0045] h(t) = v(t) - n(t);

[0046] (5) Determine whether the termination criterion is met, if met, output v(t) as the remainder and stop the decomposition calculation; if not met, execute step (6);

[0047] (6) Let h(t) be one of the IMFs, replace v(t) with the residual r(t), and execute step (1);

[0048] r(t) = v(t) - h i (t);

[0049] The final result of the empirical mode decomposition is:

[0050]

[0051] Where the criterion is selected as:

[0052] When the standard deviation is less than a given threshold value, the algorithm stops;

[0053]

[0054] where V is the voltage sequence to be decomposed, vu is the upper envelope voltage sequence, vl is the lower envelope voltage sequence, n is the average envelope voltage sequence, h is the extracted voltage sequence, r is the residual voltage sequence, h i is the voltage component of modal decomposition, t is time.

[0055] Further, the PF framework established in step seven is:

[0056] Construct internal aging parameters:

[0057]

[0058] Construct state equation and observation equation:

[0059]

[0060] where the first equation is the state equation, R is the process noise covariance matrix, the second equation is the observation equation, Q is the observation noise covariance, V s is the stack output voltage, i s is the stack current, N is the number of monolithic cells, E ocv is the open circuit voltage, R0 is the initial ohmic resistance, a is the resistance increase rate, a, b are constants, T is the operating temperature, i0 is the initial exchange current density, β is the exchange current density decay rate, i L0 is the initial limiting current density, γ is the limiting current density decay rate, k is the kth time point, t is time.

[0061] The particle filter algorithm framework is established according to the aging model, which includes:

[0062] (1) Initialize to generate sampling particles and weights according to the following formula:

[0063]

[0064]

[0065] (2) Update the particles according to the state equation:

[0066]

[0067] (3) Update the weights according to the observation equation and voltage value:

[0068]

[0069] (4) Weight normalization and resampling are performed according to the following formula:

[0070]

[0071]

[0072] (5) Output filter value:

[0073]

[0074] where p is a probability density function, f is a state equation, V is an observation equation, is the i th particle at time k, is the weight of the i th particle at time k, and n is the total number of particles.

[0075] The voltage threshold set in the tenth step is 96% of the initial voltage, and the calculation formula of the RUL is:

[0076] RUL = |T pre -T s |;

[0077] where T pre is the time when the voltage prediction value reaches the set threshold, and T s is the time when the prediction starts.

[0078] Another object of the present application is to provide a fuel cell remaining useful life prediction system applying the fuel cell remaining useful life prediction method, which comprises:

[0079] A signal receiving module is configured to receive commands from a LabView interface and signals obtained by sensors installed on an auxiliary system and an electronic load through a control system and store various historical detection signals of the fuel cell.

[0080] A signal preprocessing module is configured to obtain fuel cell output voltage signals and battery polarization curves at each time period, and preprocess the obtained signals to change the voltage signals into time series.

[0081] An unreasonable signal elimination module is configured to eliminate unreasonable signals in the time series and the polarization curves caused by sampling errors to obtain final voltage sequences and multiple polarization curves.

[0082] An initial parameter identification module is configured to establish a fuel cell aging model and input the processed polarization curves into a genetic algorithm to identify initial parameters.

[0083] An aging life prediction module is configured to build a method combining EMD and particle filtering, and to predict the aging life of the residual sequence processed by EMD by combining an aging model and a particle filtering method.

[0084] Another object of the present application is to provide a computer device comprising a memory and a processor, the memory storing a computer program, the computer program being executed by the processor to cause the processor to perform the steps of the fuel cell residual service life prediction method.

[0085] Another object of the present application is to provide a computer readable storage medium storing a computer program, the computer program being executed by a processor to cause the processor to perform the steps of the fuel cell residual service life prediction method.

[0086] Another object of the present application is to provide an information data processing terminal for implementing the fuel cell residual service life prediction system.

[0087] In combination with the above technical solutions and the technical problems solved, the technical solutions to be protected by the present application have the following advantages and positive effects:

[0088] First, in view of the technical problems existing in the prior art and the difficulty in solving the problems, the technical solutions to be protected by the present application and the results and data obtained during the research and development process are combined to analyze in detail and profoundly how the technical solutions solve the technical problems and bring some creative technical effects after solving the problems. The specific description is as follows:

[0089] The present application first receives commands from a LabView interface and signals obtained by sensors installed on an auxiliary system and an electronic load through a control system, stores various historical detection signals of the fuel cell, obtains fuel cell output voltage signals and battery polarization curves in each time period, then pre-processes the obtained signals, converts the voltage signals into time series, removes unreasonable signals caused by sampling errors in the time series and the polarization curves, obtains final voltage sequences and multiple polarization curves, establishes a fuel cell aging model, inputs the processed polarization curves into a genetic algorithm to identify initial parameters, and then builds a method combining EMD and particle filtering, and predicts the aging life of the residual sequence processed by EMD by combining an aging model and a particle filtering method. This method can accurately predict the online residual life of the fuel cell working for a long period of time, and ensure the long-term safety and stability of the system.

[0090] The application can effectively reduce the interference of sampling noise signals, can reasonably reflect the influence of historical voltage on fuel cell attenuation, and improve the precision of fuel cell voltage prediction. The particle filtering framework proposed in the application involves three fuel cell internal aging parameters, which is conducive to improving the accuracy of life prediction. In addition, the application also constructs a method combining empirical mode decomposition and particle filtering, which processes the residual sequence after EMD decomposition through particle filtering, improves the adaptability of the overall model under complex coupling conditions, and improves the accuracy and stability of life prediction. At the same time, this method saves time cost and can realize online prediction well.

[0091] Secondly, from the perspective of the product or as a whole, the technical effect and advantages of the technical solution to be protected by the application are described as follows:

[0092] The application provides a fuel cell life prediction method based on model and data driving fusion based on empirical mode decomposition and particle filtering, which has simple algorithm, high prediction precision and fast running speed. The scheme can realize more accurate prediction of the remaining service life of the fuel cell, and can update the fuel cell health state in real time according to the measured data after online operation, obtain the remaining life information, and provide key information for subsequent energy management strategy and top allocation strategy; at the same time, more accurate model establishment and real-time aging parameter updating can provide more decision information for fuel cell fault diagnosis control, and ensure that the overall system runs more stably and safely.

[0093] Thirdly, the creativity of the claims of the application is also reflected in the following important aspects:

[0094] The expected income and commercial value of the technical solution of the application after transformation are:

[0095] The online fuel cell life prediction scheme designed by the application improves the accuracy and running speed of life prediction, and has excellent flexibility and adaptability, and can accurately predict the life of the fuel cell system under various operating conditions, and provide decision information for fuel cell energy distribution, fault diagnosis and multi-stack energy management. The empirical mode decomposition and particle filtering scheme proposed also provides a method and idea for the research of fuel cell aging prediction. BRIEF DESCRIPTION OF DRAWINGS

[0096] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiments of the application will be briefly introduced as follows. Obviously, the drawings described below are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0097] Figure 1 is a flow chart of a fuel cell remaining service life prediction method provided by an embodiment of the present application;

[0098] Figure 2 is a principle diagram of a fuel cell remaining service life prediction method provided by an embodiment of the present application;

[0099] Figure 3 is a flow chart of empirical mode decomposition provided by an embodiment of the present application;

[0100] Figure 4 is a flow chart of particle filtering provided by an embodiment of the present application;

[0101] Figure 5 is a fuel cell original life degradation data schematic diagram provided by an embodiment of the present application;

[0102] Figure 6 is a fuel cell pretreatment data diagram provided by an embodiment of the present application;

[0103] Figure 7 is a fuel cell polarization curve diagram provided by an embodiment of the present application;

[0104] Figure 8 is an initial time polarization curve fitting diagram provided by an embodiment of the present application;

[0105] Figure 9 is a fuel cell voltage data EMD decomposition diagram provided by an embodiment of the present application;

[0106] Figure 10 is a fuel cell stack voltage prediction effect diagram provided by an embodiment of the present application;

[0107] Figure 11 is a fuel cell stack remaining service life prediction effect diagram provided by an embodiment of the present application. DETAILED DESCRIPTION

[0108] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0109] In view of the problems in the prior art, the present application provides a fuel cell remaining service life prediction method, system, device and terminal, which will be described in detail below with reference to the drawings.

[0110] I. Explanation of embodiments. In order to enable those skilled in the art to fully understand how the present application is specifically implemented, this part is an explanation of the embodiments of the technical solutions of the claims.

[0111] As Figure 1 shown, the fuel cell remaining service life prediction method provided by the embodiment of the application comprises the following steps:

[0112] S101, receiving commands from a LabView interface and signals obtained by sensors installed on auxiliary systems and electronic loads through a control system and storing various historical detection signals of the fuel cell;

[0113] S102, obtaining fuel cell output voltage signals and cell polarization curves in each time period; pre-processing the obtained signals to change the voltage signals into time series;

[0114] S103, removing unreasonable signals in the time series and the polarization curves caused by sampling errors to obtain final voltage series and multiple polarization curves;

[0115] S104, establishing a fuel cell aging model, inputting the processed polarization curves into a genetic algorithm to identify initial parameters;

[0116] S105, constructing a method combining EMD and particle filtering, and performing aging life prediction on the remaining sequence processed by EMD through the aging model and the particle filtering method.

[0117] As a preferred embodiment, as Figure 2 shown, the fuel cell remaining service life prediction method provided by the embodiment of the application comprises a training phase and a prediction phase;

[0118] The training phase comprises the following steps:

[0119] Step 1: starting aging test, receiving commands from a LabView interface and signals obtained by sensors installed on auxiliary systems and electronic loads through a control system, and then sending control signals to the systems by the control system. The signals obtained by the sensors (fuel cell voltage, load current, working temperature, inlet and outlet water temperature, air flow and hydrogen flow) are displayed on the LabView interface and stored in the host computer, and the fuel cell parameters (output voltage, current, aging time, etc.) are recorded;

[0120] Step 2: extracting the historical detection signals obtained in Step 1 to obtain fuel cell output voltage signals and cell polarization curves in each time period;

[0121] Step 3: pre-processing the output voltage signals obtained in Step 2 to change the voltage signals into hourly time series, and then removing unreasonable signals in the time series and the polarization curves caused by sampling errors to obtain final voltage series and multiple polarization curves;

[0122] The step 3 provided by the embodiment of the application is to convert the collected voltage into a time sequence according to the following formula:

[0123] V time (n)=V measure (n+1)

[0124] V time is an output voltage sequence, and V measur is a measured voltage signal, n is a current time point, and n+1 is a next time point.

[0125] The voltage screening is performed according to the following formula:

[0126]

[0127] V final is a final voltage, V usizer is a voltage to be screened, n is a current time point, n-1 is a previous time point, and β is a tolerance coefficient.

[0128] Step 4: a semi-empirical aging model is established based on the aging behavior inside a proton exchange membrane fuel cell;

[0129] The semi-empirical aging model of the proton exchange membrane fuel cell established in step 4 provided by the embodiment of the application is as follows:

[0130]

[0131] V s is a stack output voltage, i s is a stack current, N is a number of single cells, E ocv is an open circuit voltage, R0 is an initial value of ohmic resistance, α is an increase rate of the ohmic resistance, a and b are constants, T is an operating temperature, t is time, i0 is an initial value of exchange current density, β is an attenuation rate of the exchange current density, i L0 is an initial value of limiting current density, and γ is an attenuation rate of the limiting current density.

[0132] Step 5: an aging model parameter at an initial time is identified by fitting a polarization curve obtained in step 3 through a genetic algorithm;

[0133] The initial polarization curve is fitted in step 5 provided by the embodiment of the application according to the following genetic algorithm, and initial parameters of the aging model are identified:

[0134] The model is initialized, the population number of the genetic algorithm is selected as 50, an initial population is then created randomly, a crossover rate is set as 0.85, and a mutation rate is set as 0.12. The created aging model and the parameter at the k time point are used to calculate the voltage V sThe fitness of the population is calculated by using the model, and selection, crossover and mutation operations are performed on the variables at time k. The above operations are repeatedly performed, and it is judged whether the termination condition is met. The optimal solution is selected according to the fitness function, and the fitting result is obtained.

[0135] The fitness of the genetic function can be defined as:

[0136]

[0137] Wherein, f fit is the fitness, k is the kth sampling point of the polarization curve, V sk is the output voltage of the stack at time k, is the fitted voltage at time k, E ocv is the open circuit voltage, R0 is the initial value of ohmic resistance, i0 is the initial value of exchange current density, i L0 is the initial value of limiting current density.

[0138] Objective function:

[0139] min f fit (E ocv , R0, i0, i L0 )

[0140] Wherein, E ocv , R0, i0, i L0 are the optimized variables, and the parameter identification problem is converted into an optimization problem of minimizing the objective function.

[0141] Step 6: The voltage sequence in step 3 is decomposed into a series of intrinsic mode components and a residual sequence by using the empirical mode decomposition (EMD) method;

[0142] As shown in FIG. 1, the voltage sequence in step 6 is decomposed into a series of intrinsic mode components and a residual sequence by using the following process: Figure 3

[0143] (1) All local extreme values of the time sequence V(t) are identified, including local minimum values and local maximum values;

[0144] (2) All these local extreme values are fitted to form the envelope line of the original data by using numerical analysis method. The upper envelope line is vu(t), and the lower envelope line is vl(t);

[0145] (3) The average envelope line of the upper and lower envelope lines is calculated:

[0146] n(t)=[vu(t)+vl(t)] / 2

[0147] (4) The original data sequence V(t) is subtracted from the average envelope n(t) to obtain a new data sequence h(t); ​

[0148] h(t) = v(t) - n(t)

[0149] (5) judge whether the termination criterion is met, if yes, output v(t) as the residual, stop decomposition calculation; if not, go to step 6;

[0150] (6) set h(t) as one of the IMFs, replace v(t) with the residual r(t), and then perform step 1;

[0151] r(t) = v(t) - h i (t)

[0152] Finally, the result of the empirical mode decomposition is obtained:

[0153]

[0154] The judgment criterion is selected as:

[0155] When the standard deviation (SD) is less than a given threshold, the algorithm stops.

[0156]

[0157] wherein V is a voltage sequence to be decomposed, vu is an upper envelope voltage sequence, vl is a lower envelope voltage sequence, n is an average envelope voltage sequence, h is a voltage sequence extracted, r is a residual voltage sequence, h i is a voltage component of the mode decomposition, and t is time.

[0158] Step 7: establishing a particle filter (PF) algorithm framework according to the aging model;

[0159] As Figure 4 shown in the figure, the PF framework established in step 7 provided by the embodiment of the application is:

[0160] Constructing internal aging parameters:

[0161]

[0162] Constructing state equation and observation equation:

[0163]

[0164] wherein the first equation is a state equation, R is a process noise covariance matrix, the second equation is an observation equation, Q is an observation noise covariance, V s is a stack output voltage, i s is a stack current, and N is a number of monolithic cells. ocvis the open circuit voltage, R0 is the initial ohmic resistance, a is the resistance increase rate, a, b are constants, T is the operating temperature, t is the time, i0 is the initial exchange current density, β is the exchange current density decay rate, i L0 is the initial limiting current density, γ is the limiting current density decay rate, k is the kth time point.

[0165] The specific process is as follows:

[0166] (1) Initialization to generate sampling particles and weights according to the following formula:

[0167]

[0168]

[0169] (2) Update the particles according to the state equation:

[0170]

[0171] (3) Update the weights according to the observation equation and the voltage value:

[0172]

[0173] (4) Weight normalization and resampling according to the following formula:

[0174]

[0175]

[0176] (5) Output the filtering value:

[0177]

[0178] Where, p is the probability density function, f is the state equation, V is the observation equation, is the ith particle at k time, is the weight of the ith particle at k time, and n is the total number of particles.

[0179] Step 8: Input the remaining sequence obtained in step 6 into the particle filtering algorithm framework constructed to update the model state parameters including the resistance increase rate a, the limiting current density decay rate β, and the exchange current density decay rate γ.

[0180] Online prediction phase:

[0181] Step 9: No longer update the weights and resample the particle filtering, and bring the final model state parameters into the aging model to obtain the predicted voltage of the fuel cell;

[0182] Step 10: setting a fuel cell voltage threshold, recording the time when the predicted voltage reaches the threshold, and calculating the remaining useful life (RUL) of the fuel cell.

[0183] The voltage threshold set in step 10 provided by the embodiment of the application is 96% of the initial voltage, and the formula for calculating the RUL is:

[0184] RUL = |T pre -T s |

[0185] Wherein, T pre is the time when the voltage prediction value reaches the set threshold, T s is the time when the prediction starts.

[0186] The fuel cell remaining useful life prediction system provided by the embodiment of the application comprises:

[0187] A signal receiving module for receiving commands from a LabView interface and signals obtained by sensors installed on auxiliary systems and electronic loads through a control system and storing various historical detection signals of the fuel cell;

[0188] A signal preprocessing module for obtaining fuel cell output voltage signals and battery polarization curves in each time period, and preprocessing the obtained signals to change the voltage signals into time series;

[0189] An unreasonable signal elimination module for eliminating unreasonable signals in the time series and the polarization curves caused by sampling errors to obtain final voltage sequences and multiple polarization curves;

[0190] An initial parameter identification module for establishing a fuel cell aging model and inputting the processed polarization curves into a genetic algorithm to identify initial parameters;

[0191] An aging life prediction module for constructing a method combining EMD and particle filtering, and predicting the aging life of the remaining sequence processed by EMD by combining the aging model and the particle filtering method.

[0192] II. Application Examples. In order to prove the creativity and technical value of the technical solutions of the application, this part is an application example of the technical solutions of the claims on specific products or related technologies.

[0193] The technical scheme of the present application can realize the remaining service life prediction under different load conditions such as fuel cell steady state condition, dynamic condition and cyclic dynamic condition, and can provide decision information for the equalization management, energy distribution management and fault diagnosis of the fuel cell stack of the fuel cell locomotive, ship and aircraft, thereby ensuring the durability and reliability of the system.

[0194] III. Evidence of the effects of the embodiments. The embodiments of the present application have achieved some positive effects in the development or use process, and indeed have great advantages compared with the prior art. The following content is described in combination with the data and graphs of the test process.

[0195] As shown in Figures 5 to 11 , the present application embodiment uses the data provided by FCLAB Federation to verify the advantages of the fuel cell remaining service life prediction method provided by the present application embodiment, determines the construction of the prediction model and the specific implementation effect, Figure 10 The first 600 hours of data in the above are used for training, and the last 400 hours of data are used for verification. The present application embodiment selects 96% of the initial voltage as the failure threshold, and considers that the prediction life deviation within +10% to -20% is tolerable, and will not jump out of the online prediction state.

[0196] It should be noted that the embodiments of the present application can be realized by hardware, software or a combination of software and hardware. The hardware part can be realized by using special logic; the software part can be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or a specially designed hardware. Those skilled in the art can understand that the above devices and methods can be realized by computer executable instructions and / or included in processor control code, such as provided on a carrier medium such as a magnetic disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices of the present application and their modules can be realized by hardware circuits such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc. They can also be realized by software executed by various types of processors, or by a combination of the above hardware circuits and software, such as firmware.

[0197] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any modification, equivalent replacement and improvement within the technical range disclosed by the present application and within the spirit and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method of predicting the remaining useful life of a fuel cell, characterized by, The fuel cell remaining useful life prediction method comprises a training stage and a prediction stage, commands from a LabView interface and signals obtained by sensors installed on an auxiliary system and an electronic load are received by a control system and historical detection signals of the fuel cell are stored, output voltage signals of the fuel cell and polarization curves of the fuel cell in each time period are obtained; the obtained signals are preprocessed, the voltage signals are changed into time series; unreasonable signals caused by sampling errors in the time series and the polarization curves are removed, final voltage series and multiple polarization curves are obtained; an aging model of the fuel cell is established, the processed polarization curves are input into a genetic algorithm to identify initial parameters; an EMD and particle filtering combined method is constructed, the remaining series processed by the EMD are predicted for aging life by combining the aging model and the particle filtering method; The fuel cell remaining useful life prediction method comprises the following steps: Step one, starting the aging test, commands from a LabView interface and signals obtained by sensors installed on an auxiliary system and an electronic load are received by a control system; the control system sends control signals, the signals obtained by the sensors are displayed on a LabView interface and stored in an upper computer, and fuel cell parameters are recorded; Step two, extracting the historical detection signals obtained in step one, obtaining output voltage signals of the fuel cell and polarization curves of the fuel cell in each time period; Step three, preprocessing the output voltage signals obtained in step two, changing the voltage signals into hourly time series; removing unreasonable signals caused by sampling errors in the time series and the polarization curves, obtaining final voltage series and multiple polarization curves; Step four, establishing a semi-empirical aging model based on the aging behavior inside a proton exchange membrane fuel cell; Step five, identifying aging model parameters at an initial time by fitting the polarization curves obtained in step three through a genetic algorithm; Step six, performing intrinsic mode decomposition on the voltage series in step three by using an empirical mode decomposition method, obtaining a series of intrinsic mode components and a remaining series; Step seven, establishing a particle filtering algorithm framework according to the aging model; Step eight, input the remaining sequence obtained in step six into the constructed particle filter algorithm framework to update the model state parameters including the internal resistance increase rate , the limiting current density decay rate , the exchange current density decay rate ; Step nine, no longer updating weights and resampling the particle filtering, bringing final model state parameters into the aging model to obtain predicted voltage of the fuel cell; Step ten, setting a fuel cell voltage threshold, recording a time when the predicted voltage reaches the threshold, and calculating a remaining useful life of the fuel cell; The PF framework established in step seven is: Constructing internal aging parameters: ； Constructing a state equation and an observation equation: ； where the first equation is the state equation, R is the process noise covariance matrix, the second equation is the observation equation, Q is the observation noise covariance, is the stack output voltage, is the stack current, is the number of monolithic cells, is the open circuit voltage, is the ohmic internal resistance initial value, is the internal resistance increase rate, , is the constant, is the operating temperature, is the time, is the exchange current density initial value, is the exchange current density decay rate, is the limiting current density initial value, is the limiting current density decay rate, is the first time point; The particle filtering algorithm framework established according to the aging model comprises: (1) generating sampling particles and weights according to the following formula: ； ； (2) updating the particles according to a state equation: ； (3) updating the weights according to an observation equation and a voltage value: ； (4) performing weight normalization and resampling according to the following formula: ； ； (5) outputting a filtering value: ； wherein, is a probability density function, is a state equation, is an observation equation, is is the th particle at time t, is is the weight of the th particle at time t, is the total number of particles; The voltage threshold set in step ten is 96% of an initial voltage, and the calculation formula of the RUL is: ； wherein, is the time at which the voltage prediction reaches a set threshold, is the time at which the prediction starts.

2. The method of claim 1, wherein the remaining useful life of the fuel cell is predicted based on the change in the performance of the fuel cell. The signals obtained by the sensor in the step one include fuel cell voltage, load current, working temperature, inlet and outlet water temperature, air flow and hydrogen flow, and the fuel cell parameters include output voltage, current and aging time; In the step three, the collected voltage is changed into time series according to the following formula: ； wherein is an output voltage sequence, is a measurement voltage signal, is a current time point, is a next time point; The voltage screening is carried out according to the following formula: ； wherein, is the final voltage, is the voltage to be screened, is the current time point, is the previous time point, is the tolerance factor; The semi-empirical aging model of the proton exchange membrane fuel cell established in the step four is: ； wherein, is the stack output voltage, is the stack current, is the number of monolithic cells, is the open circuit voltage, is the initial ohmic internal resistance, is the internal resistance increase rate, , is a constant, is the operating temperature, is time, is the initial exchange current density, is the exchange current density decay rate, is the initial limiting current density, is the limiting current density decay rate.

3. The method of claim 1, wherein the remaining useful life of the fuel cell is predicted based on the change in the performance of the fuel cell. In the step five, the initial polarization curve is fitted according to the following genetic algorithm, and the initial parameters of the aging model are identified: For model initialization, the population number of genetic algorithm is selected as 50, then initial population is created randomly, the crossover rate is set as 0.85, and the mutation rate is set as 0.12; the created aging model and the parameters at time k are used to calculate the voltage at time k ; the fitness of the population is calculated by using the model, and selection, crossover and mutation operations are performed on the variables at time k; the operations are repeatedly performed, and whether the termination condition is met is judged; the optimal solution is selected according to the fitness function, and the fitting result is obtained; The fitness of the genetic function is defined as: ； wherein, is the applicability, k is the kth sampling point of the polarization curve, is the stack output voltage at time k, is the fitted voltage at time k, is the open circuit voltage, is the ohmic internal resistance initial value, is the exchange current density initial value, is the limiting current density initial value; The objective function is: min ; wherein For optimization of variables, the parameter identification problem is converted into an optimization problem with a minimum objective function.

4. The method of claim 1, wherein the remaining useful life of the fuel cell is predicted based on the change in the performance of the fuel cell. In the step six, the voltage series is decomposed according to the following process: (1) identifying all local extrema, including local minima and local maxima, of a time series ; (2) all local extreme values are fitted to form the envelope line of the original data by numerical analysis method, the upper envelope line is , and the lower envelope line is ; (3) The average envelope line of the upper and lower envelope lines is calculated: ； (4) Convert the original data sequence Subtract the average envelope A new data sequence is obtained. ; ； (5) determine if the termination criteria are met, if so, output As a remainder, stop the decomposition calculation; if not, perform step (6); (6) Set For one of the IMFs, use the residual Instead of And perform step (1); ； The result of the empirical mode decomposition is finally obtained as: ； The judgment criteria are selected as: When the standard deviation is less than a given threshold, the algorithm stops; ； wherein, is the sequence of voltages to be decomposed, is the sequence of upper envelope voltages, is the sequence of lower envelope voltages, is the sequence of average envelope voltages, is the sequence of extracted voltages, is the sequence of residual voltages, is the modal decomposed voltage component, is the time.

5. A remaining useful life prediction system for a fuel cell, which applies the method for predicting the remaining useful life of a fuel cell according to any one of claims 1 to 4, characterized by The fuel cell remaining service life prediction system comprises: A signal receiving module is configured to receive commands from a LabView interface and signals obtained by sensors installed on auxiliary systems and electronic loads through a control system and store various historical detection signals of the fuel cell; A signal preprocessing module is configured to obtain fuel cell output voltage signals and battery polarization curves in each time period, and preprocess the obtained signals to change the voltage signals into time series; An unreasonable signal elimination module is configured to eliminate unreasonable signals in the time series and the polarization curves caused by sampling errors to obtain final voltage series and multiple polarization curves; An initial parameter identification module is configured to establish a fuel cell aging model, and input the processed polarization curves into a genetic algorithm to identify initial parameters; An aging life prediction module is configured to construct a method combining EMD and particle filtering, and predict the aging life of the remaining sequence processed by EMD by combining the aging model and the particle filtering method.

6. A computer device, comprising: The computer device comprises a memory and a processor, and the memory stores a computer program, and the computer program is executed by the processor to enable the processor to execute the steps of the fuel cell remaining service life prediction method according to any one of claims 1-4.

7. A computer readable storage medium storing a computer program, the computer program being executed by a processor to enable the processor to execute the steps of the fuel cell remaining service life prediction method according to any one of claims 1-4.

8. An information data processing terminal, characterized by The information data processing terminal is configured to implement the fuel cell remaining service life prediction system according to claim 5.

Images