Fuel cell parameter identification method based on data driving

Abstract

The invention discloses a fuel cell parameter identification method based on data driving, and the method employs an improved dung beetle optimization (IDBO) algorithm to identify parameters delta 1, delta 2, delta 3, delta 4, Rc, b and theta, obtains optimal model parameters, improves the parameter identification precision of a fuel cell model, and builds a more accurate fuel cell model. The improved dung beetle optimization algorithm is improved as follows: a Cosine-Logistic-Tent composite chaotic mapping strategy is introduced in population initialization, and the uniformity and the ergodicity of the initial position of a dung beetle population are improved; a subtraction average optimizer is adopted to update the position of the dung beetle individual rolling ball, and the development and exploration ability of the algorithm is balanced; an adaptive t distribution disturbance strategy and a greedy strategy are introduced in a dung beetle group foraging stage, so that the global search capability of the algorithm is enhanced, and the calculation time is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of stability of distributed energy storage power generation in distribution networks, and in particular to a fuel cell parameter identification method based on data drive. Background Art

[0002] With a large proportion of intermittent renewable energy power connected to the power grid, ensuring the safe and stable operation of the power system requires the support of a series of technological innovations, such as large-scale energy storage technology and distributed renewable energy. Among them, fuel cell distributed power generation has the advantages of high efficiency, low noise, small size, and zero carbon emissions. It can realize modular deployment of distributed energy storage, adapt to different distribution network application scenarios, and provide flexible power supply solutions. In addition, in the distribution network, fuel cells, as energy storage systems, can balance the intermittency and volatility of renewable energy (wind power, solar energy) and improve the stability and reliability of the power grid.

[0003] The proton exchange membrane fuel cell system is composed of a stack, an air supply system, a hydrogen supply system, a water and heat management system, and a control system, of which the stack is the core of the fuel cell. A fuel cell is a nonlinear system with multiple inputs and outputs, which involves a complex mechanism of multiple couplings of multiple physical quantities of "gas-water-electricity-heat-force". It is difficult to establish an accurate fuel cell model, including the model parameters δ1, δ2, δ3, δ4 of the activation polarization loss in the stack output voltage model, and the equivalent contact resistance R of the electronic conduction of the ohmic polarization loss. c , the constant b of concentration polarization loss and the membrane water content θ of the water transfer model are usually taken according to actual engineering experience or common sense. There is a certain error between the established fuel cell model and the actual system, resulting in a large difference between the output characteristics of the fuel cell model and the actual system. Therefore, it is necessary to identify the parameters of the fuel cell model. The improved dung beetle optimization algorithm based on data-driven is used to identify the fuel cell model parameters, improve the accuracy of fuel cell model parameter identification, and establish a more accurate fuel cell model, which is conducive to PEMFC performance evaluation and control strategy development and optimization. Summary of the invention

[0004] The purpose of the present invention is to provide a fuel cell parameter identification method based on data drive in order to solve the above-mentioned problems.

[0005] To achieve the above object, the technical solution adopted by the present invention is: comprising the following steps:

[0006] S1. Collect the experimental data of fuel cell voltage and current, and use the improved wavelet threshold method to perform denoising, as the experimental data set for model parameter identification;

[0007] S2. The square error SSE between the experimental output voltage and the output voltage of the mechanism model is used as the objective function, and the improved dung beetle optimization algorithm is used to optimize the parameters δ1, δ2, δ3, δ4, R in the battery stack model. c , b and θ are identified to obtain the optimal values ​​of the model parameters;

[0008] S3. Update the stack model parameters and obtain the stack output characteristics, which are then compared with the denoised fuel cell experimental data to verify the effectiveness of the identification.

[0009] Further, the improved wavelet threshold method includes:

[0010] (1). Select the appropriate wavelet basis and decomposition layer number to perform discrete wavelet transform and obtain the wavelet coefficients of each layer;

[0011] (2). Use the threshold and threshold function to process the high-frequency coefficients in the wavelet decomposition to obtain the processed wavelet coefficients of each layer;

[0012] (3) The processed wavelet coefficients of each layer are linearly combined with the wavelet basis functions to reconstruct the signal and obtain the denoised signal.

[0013] Further, the improved dung beetle optimization algorithm includes:

[0014] a. Initialize the dung beetle population using the Cosine-Logistic-Tent composite chaotic mapping strategy;

[0015] b. Use the subtraction mean optimizer to update the position of the individual dung beetle ball;

[0016] c. Introducing the adaptive t-distribution perturbation strategy and greedy strategy during the foraging stage of dung beetle colonies to enhance the global search capability of the algorithm and reduce the calculation time.

[0017] Furthermore, the sum of squared errors (SSE) is used as the objective function to evaluate the difference between the experimental output voltage and the output voltage of the mechanism model.

[0018] Furthermore, the method is applicable to fuel cells as distributed energy storage in distribution networks to provide flexible power supply solutions.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. Improve the accuracy of fuel cell model parameter identification: By using the data-driven improved dung beetle optimization algorithm to identify the parameters in the fuel cell model, the accuracy of model parameter identification can be improved, thereby establishing a more accurate fuel cell model.

[0021] 2. Establish an accurate fuel cell model: Through improved parameter identification methods, the actual performance of fuel cells can be simulated and predicted more accurately, which is crucial for performance evaluation and development of control strategies.

[0022] 3. Improve the stability and reliability of the distribution network: Fuel cells, as distributed energy storage in the distribution network, can balance the intermittency and volatility of renewable energy and improve the stability and reliability of the power grid.

[0023] 4. Adapt to different distribution network application scenarios: Fuel cell distributed power generation has the advantages of high efficiency, low noise, small size and zero carbon emissions. It can realize modular deployment of distributed energy storage and adapt to different distribution network application scenarios.

[0024] 5. Reduce computing time: Introducing the adaptive t-distribution perturbation strategy and greedy strategy during the foraging stage of dung beetle colonies can enhance the global search capability of the algorithm and reduce computing time.

[0025] 6. Improve the global search capability of the algorithm: The improved dung beetle optimization algorithm improves the global search capability of the algorithm and reduces the risk of falling into local optimality by introducing the Cosine-Logistic-Tent composite chaotic mapping strategy, subtraction average optimizer and adaptive t distribution perturbation strategy + greedy strategy.

[0026] 7. Improve the optimization efficiency of the algorithm: Chaotic mapping strategy helps the optimization algorithm bypass the local optimal solution and improve the global search capability of the algorithm, thereby improving the optimization efficiency of the algorithm.

[0027] 8. Improve the defect that the algorithm is easily affected by local extreme values: Through the improved chaotic mapping strategy, the initial position distribution of the dung beetle population is more even, which expands the spatial search range of the dung beetle population and improves the defect that the algorithm is easily affected by local extreme values. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a parameter identification structure diagram of the improved dung beetle optimization algorithm based on data drive of the present invention;

[0029] Figure 2 A structural diagram of a proton exchange membrane fuel cell system of the present invention;

[0030] Figure 3 It is the wavelet denoising flow chart of the present invention;

[0031] Figure 4 This is a flow chart of the data-driven fuel cell parameter identification method of the present invention. DETAILED DESCRIPTION

[0032] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0033] The present invention proposes a data-driven fuel cell parameter identification method, which improves the fuel cell model parameter identification accuracy and establishes a more accurate fuel cell model when the fuel cell is used as a distributed energy storage in the distribution network. This method combines improved wavelet threshold denoising, traditional dung beetle optimization algorithm, Cosine-Logistic-Tent composite chaotic mapping, subtraction average optimizer and adaptive t-distribution perturbation strategy + greedy strategy, which can quickly and accurately identify the key parameters in the fuel cell system model, improve the fuel cell model parameter identification accuracy, and establish a more accurate fuel cell model compared to the actual fuel cell system. The overall process of fuel cell parameter identification is as follows: Figure 1 As shown:

[0034] The parameter identification process of the improved dung beetle optimization algorithm based on data-driven is as follows: integrating data-driven modeling + mechanism modeling, collecting experimental data under different operating conditions of the fuel cell system, and using improved wavelet transform to perform data denoising as the experimental data set for metaheuristic algorithm model parameter identification. The sum of squared errors (SSE) between the experimental output voltage and the output voltage of the mechanism model is used as the objective function, and the improved dung beetle optimization algorithm (IDBO) is used to optimize the parameters δ1, δ2, δ3, δ4, R c , b and θ are identified to obtain the optimal model parameters. The fuel cell model parameters are updated in real time according to the load operating conditions to obtain the corresponding battery simulation output characteristics, which are compared with the denoised experimental data to verify the effectiveness of the parameter identification of the improved dung beetle optimization algorithm.

[0035] Fuel cell stack mechanism modeling

[0036] The proton exchange membrane fuel cell system is composed of a fuel cell stack, an air supply system, a hydrogen supply system, a water and heat management system, and a control system, wherein the fuel cell stack is the core of the fuel cell. Figure 2 shown.

[0037] The present invention focuses on the identification of the parameters of the battery stack model, so a battery stack mechanism model is established. The models of other components of the system are not described here. The battery stack mechanism model is established as follows:

[0038] In the stack model, the most direct factor affecting the stack output characteristics (voltage, power) is the stack output voltage model. According to the structure and working principle of the fuel cell, ideally, at standard pressure, given gas concentration and temperature, PEMFC can output a maximum theoretical voltage of 1.229V. However, in actual operation, the activation loss, ohmic loss and concentration loss caused by changes in the temperature, pressure and gas concentration inside the stack cause the PEMFC output voltage to be lower than the maximum theoretical voltage. The actual output voltage of a single cell It can be expressed as:

[0039] (1)

[0040] Where: E n is the thermodynamic electromotive force, V act is the activation overvoltage, V ohm is the ohmic overvoltage, V con is the concentration difference overvoltage, and its unit is (V).

[0041] According to the Nernst equation, the theoretical output voltage of a single cell can be expressed by equation (2):

[0042] (2)

[0043] In formula (2), T is the operating temperature of the battery stack (K); P H2 / P O2 represents the effective partial pressure of hydrogen / oxygen at the interface between the catalyst and the gas (atm), which can be expressed as equations (3) and (4):

[0044] (3)

[0045] (4)

[0046] In formulas (3)-(4), P a and P c are the inlet pressure of the anode and cathode of the stack (atm); RH a and RH c are the relative humidity of water vapor in the anode and cathode respectively; I is the load current of the PEMFC stack (A); S represents the active area of ​​the proton exchange membrane (cm 2 );P H2O is the saturation pressure of water vapor, which can be expressed by equation (5):

[0047] (5)

[0048] Activation loss mainly occurs at the cathode of the battery, which is caused by the slow reaction rate of air and hydrogen on the membrane electrode surface and the rapid increase of load current. The calculation formula is:

[0049] (6)

[0050] Where δ1, δ2, δ3, δ4 are model coefficients; C O2 It represents the dissolved concentration of O2 at the gas-liquid interface of the proton exchange membrane (mol / cm 3 ), whose value is obtained from Henry's law:

[0051] (7)

[0052] Ohmic loss is mainly caused by the voltage loss caused by the equivalent membrane impedance of the proton exchange membrane and the resistance encountered when ions flow in the electrolyte. The calculation formula is:

[0053] (8)

[0054] In the formula, R c is the equivalent contact resistance of electronic conduction, R m is the equivalent resistance of the proton exchange membrane (Ω). R m The calculation formula is:

[0055] (9)

[0056] In the formula, ρ M and ι are the conductivity (Ω·cm) and thickness (cm) of the proton exchange membrane, respectively. The value of ι is determined by the type of proton exchange membrane, ρ M The calculation formula is:

[0057] (10)

[0058] Where θ represents the water content of the proton exchange membrane.

[0059] The concentration overvoltage is caused by the concentration gradient of reactants and products on the surface of the proton exchange membrane in the gas flow channel and the electrode. The calculation formula is:

[0060] (11)

[0061] Where b is a constant; I max Indicates the maximum current density (A / cm 2 ).

[0062] The stack output voltage and power are:

[0063] (12)

[0064] (13)

[0065] Where V stack and P stack are the stack output voltage and power respectively, I is the stack output current, and N is the number of battery cells.

[0066] For fuel cells of different power levels or under different operating conditions, the model parameters δ1, δ2, δ3, δ4 of the activation polarization loss in the stack mechanism model, the equivalent contact resistance R of the electron conduction of the ohmic polarization loss c , the constant b of concentration polarization loss and the membrane water content θ of the water transport model are usually fixed according to actual engineering experience or common sense, so their parameters need to be identified.

[0067] Data collection and preprocessing

[0068] The proton exchange membrane fuel cell power generation system transmits electrical energy to the outside through the stack, so a voltage sensor and a Hall current sensor are designed in the controller to detect the output voltage and current of the stack accurately and quickly in real time and display them. In the actual fuel cell system, due to the vibration of mechanical components (air compressor, circulating water pump, cooling fan, hydrogen circulating pump, etc.), magnetic field interference between onboard electrical appliances and the constant error of the sensor itself, the collected voltage and current data will inevitably contain noise signals, making the data contain some uncertain error terms. Therefore, before parameter identification, the collected data needs to be denoised to improve the parameter identification accuracy.

[0069] Among the many signal denoising methods, the most commonly used one is wavelet threshold denoising, which is a method of signal denoising using wavelet transform. It has good local change characteristics and excellent adaptability in both time domain and frequency domain. These characteristics are consistent with the characteristics of slow changes in low-frequency signals and rapid changes in high-frequency signals. When removing noise, the spikes and mutations in useful information can be almost completely retained. Noise signals mostly exist in the form of white noise with Gaussian distribution. After wavelet transform decomposition, the wavelet coefficients of useful signals are relatively large, and the wavelet coefficients of noise signals are relatively small. The signal denoising process is achieved by quantizing the wavelet coefficients. The collected fuel cell voltage and current data are discretely recorded at a certain frequency. Using discrete wavelet transform, the wavelet denoising process is as follows: Figure 3 shown.

[0070] Assume that the fuel cell system outputs a noisy signal in the following form:

[0071] (14)

[0072] In the formula, f(t) is the original signal, x(t) is the noisy signal, ε(t) is the noise signal, and the discrete wavelet of x(t) is:

[0073] (15)

[0074] In the formula, φ j,k (t) is the discrete wavelet basis function, then equation (15) can be expressed as:

[0075] (16)

[0076] Where D j,k , U j,k 、E j,k The wavelet transform coefficients belong to the noisy signal, the original signal without noise and the noise signal respectively.

[0077] The wavelet coefficients show different behaviors on useful signals and noise signals. The threshold can be selected as an appropriate number. When the wavelet coefficient is smaller than the threshold, it is considered that the coefficient at this moment is mainly caused by noise and needs to be set to zero; when the wavelet coefficient is larger than the threshold, it can be regarded as caused by the effect of useful signals. These coefficients can be directly retained or shrunk to zero by a specified amount.

[0078] The steps of wavelet threshold denoising are as follows:

[0079] (1) Discrete wavelet decomposition of noisy signals. According to the main characteristics of the noisy signal, the appropriate wavelet basis and decomposition layer number are selected, and discrete wavelet transform is performed to obtain the wavelet coefficients D of each layer. j,k .

[0080] (2) Threshold quantization processing: use the threshold and threshold function to process the high-frequency coefficients in the wavelet decomposition to obtain the processed wavelet coefficients D' of each layer j,k .

[0081] (3) Wavelet coefficient reconstruction: the processed wavelet coefficients D' of each layer j,k The signal is reconstructed by linear combination with the wavelet basis function to obtain the denoised signal x'(t). Since the D4 wavelet has good regularity and good time-frequency decomposition performance, the D4 wavelet is used as the wavelet basis function during reconstruction, and the obtained data signal is more continuous and smooth. The D4 wavelet formula is:

[0082] (17)

[0083] Where t is a discrete time point and u(t) is the unit function.

[0084] Wavelet basis function, decomposition level, threshold and threshold function are the most important factors affecting the wavelet denoising effect. Wavelet basis function and decomposition level are selected conventionally, and the focus is on improving the threshold and threshold function.

[0085] Wavelet basis function. In order to ensure that the signal can be accurately reconstructed, the same wavelet basis function is used in the decomposition and reconstruction processes. The corresponding discrete wavelet basis function is D4 wavelet, as shown in Equation (17).

[0086] Decomposition level. As the number of decomposition levels increases, different characteristics between real signals and noise signals will emerge. However, as the number of decomposition levels increases, less information is obtained, which will affect subsequent signal reconstruction and cause signal distortion. Usually, the support length, regularity and orthogonality of the wavelet basis should be considered. Therefore, the number of decomposition levels is selected to be 4.

[0087] Threshold. A reasonable threshold will separate the noise signal in the high-frequency coefficients and set it to zero, then re-estimate the high-frequency coefficients to obtain the estimated coefficients, and finally reconstruct the denoised image. The threshold is mainly determined by the noise variance and the energy of the sub-band coefficients. The traditional threshold selection method does not change with the change of the decomposition scale. The new threshold selection is designed as follows:

[0088] (18)

[0089] In the formula, λ is the threshold, α is the median absolute deviation of the noise, and j is the decomposition scale. The threshold changes with the decomposition scale. The larger the decomposition scale, the smaller the threshold will be. The median absolute deviation expression is as follows:

[0090] (19)

[0091] In the formula, x i is the i-th data point in the data set, and median(x) represents the median of the data set x.

[0092] Threshold function. The threshold function will affect the smoothness of the reconstructed signal and its high-frequency information. The two traditional threshold functions are hard threshold and soft threshold function. The disadvantage of the hard threshold function is that it is discontinuous at the threshold point; the disadvantage of the soft threshold function is that there is a constant deviation between the original coefficient and the wavelet decomposition coefficient. Construct a new threshold function so that the threshold function is continuous at its threshold point and has the property of high-order differentiability. The new threshold function is as follows:

[0093] (20)

[0094] Wherein, α and n are both positive numbers, the new threshold function is continuous and high-order differentiable. When α=0, n→0, the characteristics of the new threshold function are close to the soft threshold function. When α=0, n→∞, the characteristics of the new threshold function are close to the hard threshold function.

[0095] Improved dung beetle optimization algorithm parameter identification

[0096] For most constrained optimization problems in engineering and industry, the use of metaheuristic algorithms can effectively avoid falling into local optimality when searching for the optimal parameters of the model. Therefore, the improved dung beetle optimization algorithm is used to estimate the optimal value of the unknown parameters within the constraints. The Dung Beetle Optimizer (DBO) is a new metaheuristic algorithm proposed by Professor Shen Bo's team in 2022. Inspired by the natural behaviors of dung beetle populations such as rolling, dancing, foraging, stealing and reproduction, the algorithm takes into account global search and local development, and has the characteristics of fast search speed and strong optimization ability.

[0097] Traditional dung beetle optimization algorithm

[0098] The whole process of the Dung Beetle Optimization Algorithm (DBO) is divided into four stages, namely the dung beetle ball rolling stage, the egg-laying stage, the small dung beetle feeding stage and the dung beetle stealing stage.

[0099] (1) Dung beetle rolling stage

[0100] The dung beetle will navigate based on environmental factors such as light and wind direction, and randomly select a direction in the global search space to push the ball in a straight line. When no obstacles are encountered, the dung beetle's rolling ball position update rule can be expressed as:

[0101] (twenty one)

[0102] Where t is the current iteration number; x i (t) represents the position of the i-th dung beetle at the t-th iteration; β represents the natural coefficient of deviation from the original direction, which is assigned a value of -1 or 1; m represents the defect coefficient, m∈ ; n is a constant, ∈ ; Δx represents the light intensity; X ω represents the worst position in the population. Where β is:

[0103] (twenty two)

[0104] Where ρ is a random number, ρ∈[0,1]; η is a probability value, η=0.9. When ρ<η, it is an obstacle-free state, otherwise it is an obstacle-existing state.

[0105] When the dung beetle encounters an obstacle and cannot move forward, it re-determines the moving direction through dancing behavior and obtains a new forward route. The tangent function is used to simulate the dancing behavior to obtain a new rolling direction. At this time, the position update can be expressed as:

[0106] (twenty three)

[0107] Where φ represents the deflection angle, φ[0,π], when it is equal to 0, π / 2 or π, the position of the dung beetle will not be updated.

[0108] (2) Egg-laying stage

[0109] Dung beetles roll dung balls to a safe area and hide them to provide a safe environment for female dung beetles to lay eggs. Based on this, a boundary selection strategy is proposed to simulate the egg-laying area of ​​female dung beetles, which can be expressed as:

[0110] (twenty four)

[0111] Where Ub* and Lb* represent the upper and lower bounds of the egg-laying area, respectively; X* represents the current optimal position of all dung beetles in the population; T max represents the maximum number of iterations; Ub and Lb represent the upper and lower bounds of the optimization problem, respectively.

[0112] After the egg-laying area is determined, each female dung beetle produces a chick each time. The egg-laying area changes dynamically with the number of iterations, and the position of the chick also changes dynamically, which can be expressed as:

[0113] (25)

[0114] In the formula, y i (t) represents the position of the i-th ball at the t-th iteration; v1 and v2 are both independent random vectors of 1×D dimensions, and D is the dimension of the optimization problem.

[0115] (3) Small dung beetle foraging stage

[0116] When the young dung beetles in the brood ball grow up, they need to crawl out to find food. Therefore, it is necessary to establish an optimal foraging area and guide the young dung beetles to forage in order to achieve the purpose of exploring the space. The boundary of the optimal foraging area can be defined as:

[0117] (26)

[0118] Where Ubˊ and Lbˊ represent the upper and lower limits of the optimal foraging area, respectively; Xˊ represents the global optimal position; other parameter definitions are the same as those in formula (24).

[0119] After determining the best foraging area, the position of the dung beetle during foraging is updated as follows:

[0120] (27)

[0121] Where u1 is a random number that obeys the normal distribution; u2 is a random vector, u2∈(0,1).

[0122] (4) Dung beetle stealing stage

[0123] In DBO, some dung beetles will steal the dung balls of other dung beetles. These dung beetles are called thieving dung beetles. They do not steal at random positions, but steal near the global optimal position. The position update method of thieving dung beetles is as follows:

[0124] (28)

[0125] Where R is a constant and w is a 1×D dimensional random vector that obeys a normal distribution.

[0126] The dung beetle optimization algorithm has the disadvantages of imbalanced global exploration and local development capabilities in the parameter identification process, easy to fall into local optimality, and weak global exploration capabilities. Therefore, in order to improve the search performance of DBO, the improved dung beetle optimization algorithm (IDBO) is adopted to improve the shortcomings of the DBO algorithm by integrating the Cosine-Logistic-Tent composite chaotic mapping strategy, subtractive mean optimizer and adaptive t distribution perturbation strategy + greedy strategy.

[0127] Improved dung beetle optimization algorithm

[0128] (1) Cosine-Logistic-Tent composite chaotic mapping strategy

[0129] The population initialization of the DBO algorithm adopts a random generation method, which leads to an uneven distribution of the initial population, thereby reducing the diversity of the population. Chaotic mapping has strong regularity and ergodicity, which can increase the diversity of the initial population, thereby improving the optimization efficiency of the algorithm. Based on this, the present invention uses chaotic mapping to initialize the individual positions of dung beetles, uses the chaotic mapping relationship to project the obtained values ​​into the chaotic variable space, and then maps the generated chaotic values ​​into the initial space of the DBO algorithm through linear transformation. Since the chaotic variables replace the random variables in the operation process of the DBO algorithm during the initialization position update process, the chaotic mapping strategy has a wider search range for the solution space than the random search strategy.

[0130] Chaotic mapping helps the optimization algorithm bypass the local optimal solution and improve the global search ability of the algorithm, but it also has limitations such as limited parameter range and relatively low chaos level. In view of the above limitations, the Cosine-Logistic-Tent composite (CLT) chaotic mapping strategy is adopted, and the specific formula can be defined as:

[0131] (29)

[0132] In the formula, x i is the i-th chaotic number; r is the control parameter, r∈[0,1].

[0133] The Cosine-Logistic-Tent composite (CLT) chaotic mapping strategy combines the conventional Cosine mapping, Logistic mapping and Tent mapping. The fused chaotic value has better ergodicity and more uniform distribution. The improved initial position distribution of the dung beetle population based on the CLT chaotic mapping strategy is more uniform, which expands the spatial search range of the dung beetle population and improves the defect of the algorithm being easily affected by local extreme values, thereby improving the optimization efficiency of the DBO algorithm.

[0134] (2) Subtraction mean optimizer updates the position of the individual dung beetle rolling ball

[0135] The DBO algorithm only relies on the worst value in the first stage of the rolling process, lacks communication with other dung beetles, and has many parameters. In order to enhance the diversity of the population and improve the global search capability, the Subtraction-Average-Based Optimizer (SABO) is selected to update the position of the individual dung beetle rolling ball in the first stage. SABO updates the position of the population members in the search space by calculating the arithmetic average position of all individuals in the optimization group, replacing the better position to prevent the algorithm from relying on specific population members, and the optimization effect and convergence speed are better. SABO is based on the special operation "— v ” is called the subtraction of individual B to individual A, and is defined as follows:

[0136] (30)

[0137] Where F(X) is the objective function value of individual X, is a random number vector of dimension D generated from the set {1, 2}.

[0138] The DBO algorithm only relies on the worst value in the first stage of rolling the ball, lacks communication with other dung beetles, and has many parameters. Therefore, the global exploration strategy of the SABO algorithm is used to improve the above limitations of the DBO algorithm, randomly determine the position of a dung ball and roll it to improve the global search ability of the DBO algorithm. The formula for updating the individual position in the search space is as follows:

[0139] (31)

[0140] Where X´ new represents the new position of the i-th dung beetle individual; rand is a random number between [0,1].

[0141] Add a greedy strategy, compare the fitness values ​​before and after the update, and select the better value. The greedy strategy is as follows:

[0142] (32)

[0143] (3) Adaptive t-distribution perturbation strategy + greedy strategy to update the dung beetle foraging stage

[0144] During the foraging stage of small dung beetles, the DBO algorithm tends to converge, and the differences between individual dung beetles are small. Introducing small disturbances at this time can prevent the DBO algorithm from falling into the local optimum. The adaptive t-distribution perturbation strategy is used to perturb the foraging behavior of small dung beetles, update the non-optimal positions of dung beetles, and reduce the calculation time to improve the convergence speed of the DBO algorithm. At the same time, in order to make the DBO algorithm have better global development capabilities in the early stage of iteration and good local exploration capabilities in the later stage of iteration, the number of iterations of the DBO algorithm is used as the degree of freedom parameter of the adaptive t-distribution, which can greatly improve the convergence speed and solution efficiency of the DBO algorithm. The position update of the small dung beetle after disturbance can be expressed as:

[0145] (33)

[0146] Where, X new (t) represents the position of the individual dung beetle after the adaptive t distribution change disturbance; X best (t) represents the global optimal solution in the current iteration stage; t(C iter ) represents a random number generated by adaptive t distribution.

[0147] Introduce a greedy strategy to update the optimal individual of dung beetles before and after disturbance, which is defined as follows:

[0148] (34)

[0149] The execution steps of the improved dung beetle optimization algorithm are as follows:

[0150] 1) Randomly initialize the dung beetle population, set the population size and number of iterations;

[0151] 2) Use Cosine-Logistic-Tent composite chaotic mapping to initialize the dung beetle population and calculate the initial fitness value of each dung beetle;

[0152] 3) Update the position of the dung beetle according to different update rules. When ρ<0.9, the fusion subtraction average optimizer updates the position of the rolling dung beetle. When ρ≥0.9 and θ≠0, π / 2, π, the moving direction is updated through dancing behavior, otherwise the position is not updated.

[0153] 4) Update the positions of the chicks, small dung beetles and thieving dung beetles according to their respective boundary conditions, and calculate the fitness value of each dung beetle to obtain the optimal solution;

[0154] 5) Introduce an adaptive t-distribution dynamic selection strategy to perturb the current optimal solution and obtain a new solution. The greedy strategy evaluates whether the dung beetle position needs to be updated;

[0155] 6) Determine whether the algorithm has reached the maximum number of iterations. If so, terminate the operation and output the global optimal solution. Otherwise, return to step 3) to continue execution.

[0156] The process of fuel cell parameter identification method based on data drive is as follows: Figure 4 As shown in the figure, a data-driven fuel cell parameter identification method is invented. As a distributed energy storage of the distribution network, fuel cells can provide a flexible power supply solution for the distribution network. Due to the complex mechanism of fuel cells, the model is far from the actual system, and it is necessary to accurately model the fuel cell. The specific steps are as follows:

[0157] S1. Collect the experimental data of fuel cell voltage and current, and use the improved wavelet threshold method (improved threshold and threshold function) to denoise them as the experimental data set for model parameter identification.

[0158] S2. Taking the sum of square errors (SSE) between the experimental output voltage and the output voltage of the mechanism model as the objective function, the improved dung beetle optimization algorithm is used to optimize the δ1, δ2, δ3, δ4, and R c , b and θ are identified to obtain the optimal values ​​of the model parameters.

[0159] S3. Update the stack model parameters to obtain the stack output characteristics, compare them with the denoised fuel cell experimental data, and verify the effectiveness of the identification.

[0160] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0161] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A fuel cell parameter identification method based on data drive, characterized in that: The following steps are involved: S1. Collect the experimental data of fuel cell voltage and current, and use the improved wavelet threshold method to perform denoising, as the experimental data set for model parameter identification; S2. The square error SSE between the experimental output voltage and the output voltage of the mechanism model is used as the objective function, and the improved dung beetle optimization algorithm is used to optimize the parameters in the battery stack model. δ 1. δ 2. δ 3. δ 4. R c , b and θ Perform identification to obtain the optimal values ​​of model parameters; S3. Update the stack model parameters and obtain the stack output characteristics, which are then compared with the denoised fuel cell experimental data to verify the effectiveness of the identification.

2. The method for identifying fuel cell parameters based on data drive according to claim 1, characterized in that: Improved wavelet thresholding methods include: (1) Select the appropriate wavelet basis and decomposition layer number to perform discrete wavelet transform and obtain the wavelet coefficients of each layer; (2) Use the threshold and threshold function to process the high-frequency coefficients in the wavelet decomposition to obtain the processed wavelet coefficients of each layer; (3) The processed wavelet coefficients of each layer are linearly combined with the wavelet basis functions to reconstruct the signal and obtain the denoised signal.

3. The method according to claim 1, characterized in that Improvements to the dung beetle optimization algorithm include: a. Initialize the dung beetle population using the Cosine-Logistic-Tent composite chaotic mapping strategy; b. Use the subtraction mean optimizer to update the position of the individual dung beetle ball; c. Introducing the adaptive t-distribution perturbation strategy and greedy strategy during the foraging stage of dung beetle colonies to enhance the global search capability of the algorithm and reduce the calculation time.

4. The method for identifying fuel cell parameters based on data drive according to claim 1, characterized in that: The sum of squared errors (SSE) is used as the objective function to evaluate the difference between the experimental output voltage and the output voltage of the mechanism model.

5. The method for identifying fuel cell parameters based on data drive according to claim 1, characterized in that: This approach is applicable to fuel cells as distributed energy storage in distribution networks to provide flexible electricity supply solutions.

Images