Methods, devices, and electronic equipment for predicting fuel cell operating parameters

Abstract

This invention relates to the field of fuel cell testing technology, and specifically provides a method, apparatus, and electronic device for predicting fuel cell operating parameters. The prediction method of this invention is used to predict the current and temperature of individual cells within a fuel cell under different operating conditions, and includes: determining a combination of test conditions, and testing the current and temperature of individual cells under each test condition using testing equipment; establishing a prediction model based on the test data of the current and temperature of individual cells corresponding to each test condition; and predicting the current and temperature of individual cells using the prediction model during fuel cell operation, based on the actual operating conditions. The prediction model is an artificial neural network model trained using a K-fold cross-validation algorithm, and the weights and bias coefficients of the artificial neural network model are determined using a bat algorithm. This prediction method can save on equipment investment costs when testing the current and temperature of individual cells within a fuel cell.

Description

Technical Field

[0001] This invention relates to the field of fuel cell testing technology, and in particular to a method, apparatus, and electronic device for predicting fuel cell operating parameters. Background Technology

[0002] The fuel cell stack is the core power generation unit within a fuel cell and is composed of multiple individual cells stacked together. During the operation of the fuel cell stack, the current density and temperature distribution of the individual cells within the stack affect the overall performance of the stack, such as heat dissipation and output performance, and consequently, the stack's lifespan. Therefore, it is necessary to monitor the current and temperature of the individual cells during the operation of the fuel cell stack to obtain the current density and temperature distribution of the individual cells within the stack.

[0003] In existing technologies, current-temperature zoning testing equipment (hereinafter referred to as testing equipment) is mainly used to test the current and temperature of individual cells in fuel cell stacks. Each fuel cell stack is equipped with a corresponding current-temperature zoning testing equipment. This testing equipment typically collects the current and temperature data of each individual cell in real time through contact sensors or non-contact probes embedded between the individual cells, and then integrates them into the current density distribution and temperature distribution of the individual cells in the stack.

[0004] However, in practical applications, if a current and temperature zone testing device is installed for each fuel cell stack to conduct tests using existing technologies, the manufacturing cost of fuel cell stacks will be high. Summary of the Invention

[0005] In view of this, the present invention aims to propose a method for predicting the operating parameters of a fuel cell, so as to save equipment investment costs when testing the current and temperature of individual cells in a fuel cell.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0007] A method for predicting fuel cell operating parameters, used to predict the current and temperature of a single cell within a fuel cell under different operating conditions, the prediction method comprising:

[0008] Based on the testing range of the testing equipment, a combination of testing conditions is determined, and under each testing condition, the current and temperature of the single battery are tested using the testing equipment.

[0009] Based on the operating data of each test condition and the test data of the current and temperature of the single battery corresponding to each test condition, a prediction model for predicting the current and temperature of the single battery is established.

[0010] During the operation of the fuel cell, the current and temperature of the individual cells within the fuel cell are predicted using the prediction model based on the actual operating data of the fuel cell.

[0011] The testing equipment is used to measure the current and temperature of the single cell in the fuel cell, and the prediction model is an artificial neural network model trained using the K-fold cross-validation algorithm, with the weights and bias coefficients of the artificial neural network model determined by the bat algorithm.

[0012] Furthermore, the determination of the test condition combination based on the test range of the test equipment includes:

[0013] Based on the preset orthogonal array and the test condition range, the test condition combination is determined, and the test condition combination covers the test condition range of the test equipment.

[0014] Furthermore, the step of establishing a predictive model for predicting the current and temperature of the single battery based on the operating condition data of each of the test conditions and the test data of the current and temperature of the single battery corresponding to each of the test conditions includes:

[0015] The working condition data and the corresponding test data are combined respectively, and the combined data are divided into training data and validation data. The training data is then preprocessed.

[0016] Based on the preprocessed training data and the validation data, the weights and bias coefficients of the artificial neural network model are determined by the bat algorithm, and then the artificial neural network model is trained by the K-fold cross-validation algorithm until the optimal artificial neural network model is output.

[0017] The output of the optimal artificial neural network model is used as the prediction model for predicting the current and temperature of the single cell.

[0018] Furthermore, after determining the weights and bias coefficients of the artificial neural network model using the bat algorithm, the artificial neural network model is trained using the K-fold cross-validation algorithm until the optimal artificial neural network model is output, including:

[0019] Initialize the parameters of the bat algorithm, establish a bat population, and determine the initial position of the bat population as the weights and bias coefficients of the artificial neural network model;

[0020] Using the preprocessed training data, the artificial neural network model is trained using the K-fold cross-validation algorithm to obtain the target training model;

[0021] The accuracy of the target training model is used as the fitness value of the bat algorithm, and the position of the bats in the bat population is updated iteratively. In each iteration, the weights and bias coefficients of the artificial neural network model are updated according to the updated bat position, and the fitness value of the bat algorithm is calculated in each iteration.

[0022] When the iteration of the bat population reaches the preset requirements, the optimal artificial neural network model is determined and output based on the target training model after the current iteration.

[0023] Furthermore, the step of using the preprocessed training data to train the artificial neural network model through a K-fold cross-validation algorithm to obtain the target training model includes:

[0024] The preprocessed training data is divided into K folds of data on average.

[0025] Select one fold of data from the K-fold data as the validation set, and use the remaining folds of data as the training set;

[0026] The artificial neural network model is trained using the training set;

[0027] The trained artificial neural network model is validated using the validation set, and the performance of the trained artificial neural network model is evaluated.

[0028] Repeat the selection of the validation set and the training set, as well as the validation and evaluation process of the artificial neural network model, until the validation set has traversed every fold of data.

[0029] The optimal model is selected from the trained artificial neural network models and used as the target training model.

[0030] Furthermore, evaluating the performance of the trained artificial neural network model includes:

[0031] Based on the validation set, the accuracy of the trained artificial neural network model is calculated;

[0032] The accuracy of the artificial neural network model includes the fitting coefficient and / or prediction accuracy of the artificial neural network model.

[0033] Furthermore, the preset requirements include the bat population reaching the maximum number of iterations, or the target training model reaching a preset accuracy standard.

[0034] Furthermore, the speed update formula for the bats in the bat population is as follows:

[0035]

[0036] The formula for updating the position of the bats in the bat population is:

[0037]

[0038] Where t is the current time, v i Let x be the speed of the i-th bat. * f is the current globally optimal position. i Let be the frequency emitted by the i-th bat.

[0039] Compared with the prior art, the present invention has the following advantages:

[0040] The fuel cell operating parameter prediction method of this invention trains an artificial neural network model using test data under various test conditions to obtain a prediction model. This prediction model, along with the actual operating conditions, can then predict and determine the current and temperature of individual cells within the fuel cell. In practical applications, by inputting the operating conditions of each fuel cell into the prediction model, the current and temperature of each individual cell within the fuel cell can be obtained, enabling the testing of the temperature and current of each individual cell. This allows for the determination of the current density distribution and temperature distribution of the individual cells within the fuel cell. It eliminates the need to configure and install a separate testing device for each fuel cell to measure the current density and temperature distribution; at a minimum, only one testing device is required to test the current and temperature of individual cells within the fuel cell under various operating conditions, thus reducing equipment investment costs when testing fuel cell operating parameters.

[0041] Another objective of this invention is to provide a device for predicting fuel cell operating parameters, used to predict the current and temperature of a single cell within a fuel cell under different operating conditions. The prediction device includes:

[0042] The testing module is used to determine the combination of test conditions according to the test range of the testing equipment, and to test the current and temperature of the single battery under each test condition by means of the testing equipment.

[0043] The model building module is used to establish a prediction model for predicting the current and temperature of the single battery based on the operating data of each test condition and the test data of the current and temperature of the single battery corresponding to each test condition.

[0044] The prediction module, during the operation of the fuel cell, predicts the current and temperature of each cell within the fuel cell based on the actual operating data of the fuel cell using the prediction model.

[0045] The testing equipment is used to measure the current and temperature of the single cell in the fuel cell, and the prediction model is an artificial neural network model trained using the K-fold cross-validation algorithm, with the weights and bias coefficients of the artificial neural network model determined by the bat algorithm.

[0046] The present invention also proposes an electronic device comprising:

[0047] Memory, which stores computer programs;

[0048] processor;

[0049] When the computer program is executed by the processor, it implements the above-described method for predicting fuel cell operating parameters.

[0050] The fuel cell operating parameter prediction device and electronic equipment of the present invention can establish and train a prediction model, and use the prediction model and actual operating conditions to predict and determine the current and temperature of individual cells within the fuel cell. Furthermore, in practical applications, by inputting the operating conditions of each fuel cell into the prediction model, the current and temperature of individual cells within each fuel cell can be obtained. Thus, it is unnecessary to configure and install a separate testing device for each fuel cell to measure current and temperature; only one testing device is needed to obtain the test data, which helps reduce equipment investment costs. Attached Figure Description

[0051] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0052] Figure 1 This is a flowchart illustrating the prediction method described in an embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of the process of training the prediction model in the prediction method described in the embodiment of the present invention;

[0054] Figure 3 This is an example diagram showing the division of the test range in the prediction method described in this embodiment of the invention;

[0055] Figure 4 This is a schematic diagram of the process of training an artificial neural network model using the bat algorithm and K-fold cross-validation algorithm in the prediction method described in the embodiments of the present invention;

[0056] Figure 5 This is a schematic diagram of the process of training the target training model using the K-fold cross-validation algorithm in the prediction method described in the embodiments of the present invention;

[0057] Figure 6 This is a schematic diagram of the architecture of the prediction model described in an embodiment of the present invention;

[0058] Figure 7 This is a schematic diagram of the configuration of the prediction device according to an embodiment of the present invention;

[0059] Figure 8 This is a schematic diagram illustrating the structure of the electronic device described in an embodiment of the present invention;

[0060] Explanation of reference numerals in the attached figures:

[0061] 510. Testing module; 520. Model building module; 530. Prediction module;

[0062] 610. Memory; 620. Processor. Detailed Implementation

[0063] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0064] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0065] In the description of this invention, it should be noted that the use of terms such as "upper," "lower," "inner," and "outer," indicating orientation or positional relationship, is based on the orientation or positional relationship shown in the accompanying drawings and is only for the convenience of describing the invention and simplifying the description. It does not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the use of terms such as "first" and "second" is also for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0066] Furthermore, in the description of this invention, unless otherwise explicitly defined, the terms "installation," "connection," "joining," and "connector" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention in light of the specific circumstances.

[0067] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0068] Example 1

[0069] This embodiment relates to a method for predicting fuel cell operating parameters, which is used to predict the current and temperature of a single cell in a fuel cell under different operating conditions, thereby predicting the current density distribution and temperature distribution of a single cell in a fuel cell.

[0070] Current density (or simply current density) generally refers to the current passing through a unit area, and its unit is A / cm². 2 For example, a single cell can be divided into multiple independent regions (such as a 5×5 grid). The current in each region is measured, and the current density in that region is obtained by dividing the current by the area of ​​the corresponding region. Then, the current densities of each region can be integrated to obtain the overall current density distribution of the single cell.

[0071] The prediction method in this embodiment establishes a prediction model for the current and temperature of a single cell, and uses this prediction model to predict the current and temperature distribution of a single cell in a fuel cell, thereby reducing the investment cost of testing equipment.

[0072] In related technologies, the uniformity of current density and temperature distribution within a fuel cell affects its performance. For example, excessive differences in current density distribution can lead to localized voltage differences and voltage drops within the fuel cell, thus degrading its performance. Furthermore, localized high temperatures can threaten the durability of the fuel cell stack, thereby impacting its lifespan.

[0073] Currently, related technologies typically utilize S++ equipment (a current distribution and temperature zone testing device) to measure the current density and temperature distribution of fuel cells. S++ equipment is a fuel cell current density and temperature testing system applicable to various types of fuel cells, such as proton exchange membrane fuel cells and direct methanol fuel cells. S++ equipment divides a single cell into multiple independent small regions, each of which is measured by independent current and temperature sensors. This yields the current and temperature of each region within the single cell, thus providing the overall current and temperature of the single cell, as well as the current density and temperature distribution within the cell.

[0074] However, in related technologies, to measure the current density distribution and temperature distribution of individual cells in multiple fuel cell stacks, it is necessary to install an S++ device in each of the multiple stacks for measurement, which results in high costs.

[0075] Furthermore, in fuel cells, the current density distribution and temperature distribution of individual cells within the stack are affected by the cell operating conditions, such as anode and cathode temperatures, anode and cathode stoichiometry, and anode and cathode relative humidity. These operating conditions will affect the current density and temperature distribution of the fuel cell, and different operating conditions will also interact with each other, thereby increasing the impact on the current density and temperature distribution.

[0076] In related technologies, when the uniformity of current density and temperature distribution is poor, adjustments are typically made to the operating conditions of the fuel cell stack to regulate the current density and temperature distribution of individual cells within the stack. However, these technologies only allow measurement of the current density and temperature distribution under the current operating conditions using S++ equipment. When the uniformity of current density and temperature distribution is poor, no adjustment suggestions can be provided; adjustments must be made manually based on experience, resulting in poor convenience.

[0077] Therefore, in order to overcome the shortcomings of existing methods for measuring the current density distribution and temperature distribution of individual cells within a fuel cell stack, this embodiment provides a method for predicting fuel cell operating parameters, and combines... Figure 1 The prediction method mainly includes the following steps.

[0078] Step S110: Determine the test condition combination according to the test range of the test equipment, and test the current and temperature of the single battery under each test condition using the test equipment.

[0079] In step S110, for example, the operating conditions of the fuel cell may specifically include the following operating parameters: anode metering ratio, cathode metering ratio, anode pressure, cathode pressure, anode humidity, cathode humidity, cooling water inlet temperature, cooling water inlet and outlet temperature difference, anode nitrogen content, etc.

[0080] The operating condition test range is the range formed by the upper and lower limits of each operating condition parameter. The upper and lower limits represent the upper and lower limits of the values ​​that the parameter may reach during fuel cell operation. For example, the operating condition test range includes: 1.2 < anode metering ratio < 1.8, 1.8 < anode metering ratio < 3.0, 1.0 < anode pressure < 2.5 bar, 1.0 < cathode pressure < 2.0 bar, 80% anode humidity < 100% RH (relative humidity), 80% < cathode humidity < 100% RH, 50℃ < cooling water inlet temperature < 70℃, 5℃ < cooling water inlet / outlet temperature difference < 10℃, and anode nitrogen content < 2%.

[0081] In step S110, the test condition combination is a combination of test conditions composed of the values ​​of various operating parameters. For example, an anode metering ratio of 1.3, a cathode metering ratio of 1.9, anode pressure of 1.5, cathode pressure of 1.2, anode humidity of 90%, cathode humidity of 85%, cooling water inlet temperature of 60°C, cooling water inlet and outlet temperature difference of 8°C, and anode nitrogen content of 1.2% constitute a test condition combination. All the determined test conditions constitute this test condition combination.

[0082] Furthermore, in step S110, the aforementioned testing equipment is a device used to measure the current and temperature of a single cell within a fuel cell, and this testing equipment can generally be the aforementioned S++ device.

[0083] In practice, the parameter values ​​of the actual operating conditions of a fuel cell stack are set as the parameter values ​​of a set of test conditions. Under these test conditions, the S++ device can be used to measure the current and temperature of a single cell under these test conditions, thereby measuring the current density distribution and temperature distribution of the single cell under these test conditions.

[0084] Step S120: Based on the operating data of each test condition and the test data of the current and temperature of the single cell corresponding to each test condition, establish a prediction model for predicting the current and temperature of the single cell.

[0085] In step S120, the prediction model can specifically be an artificial neural network (ANN) model. More specifically, the prediction model is an artificial neural network model trained using the K-fold cross-validation (KFold) algorithm, and the weights and bias coefficients of the artificial neural network model are determined by the bat algorithm.

[0086] K-fold cross-validation is an algorithm that divides a dataset into K equal subsets (where K is a positive integer) and performs multiple iterations of training and validation. K-fold cross-validation is an important technique in machine learning for evaluating model performance; by iterating through training and validation multiple times, it can avoid underfitting, a problem that easily occurs with limited training data.

[0087] The Bat Algorithm is an optimization algorithm based on the behavior of bat colonies. It simulates the behavior of bats searching for food and avoiding obstacles, and progressively searches for the optimal solution by updating rules. In this embodiment, the Bat Algorithm is used to determine the weights and bias coefficients of the artificial neural network, which facilitates finding the optimal solution for the artificial neural network model. This minimizes the difference between the prediction results and the actual results, thereby improving the accuracy of the prediction model.

[0088] Specifically, in step S120, each test condition and the test data measured under each test condition can be used as sample data to train the artificial neural network model. During the training process, the K-fold cross-validation algorithm is used for training, and the weights and bias coefficients of the artificial neural network model are determined by the bat algorithm. After training is completed, the prediction model can be obtained.

[0089] Step S130: During fuel cell operation, based on the actual operating conditions data of the fuel cell, the current and temperature of the individual cells in the fuel cell are predicted by a prediction model.

[0090] In step S130, specifically, the operating condition data includes the aforementioned operating condition parameters and their corresponding parameter values. For example, the actual input operating condition data of the fuel cell includes the actual value of the anode metering ratio, the actual value of the cathode metering ratio, the actual value of the anode pressure, the actual value of the cathode pressure, the actual value of the anode humidity, the actual value of the cathode humidity, the actual value of the cooling water inlet temperature, the actual value of the cooling water inlet and outlet temperature difference, and the actual value of the anode nitrogen content.

[0091] More specifically, in step S130, the actual operating condition data of the fuel cell can be input into the prediction model. The prediction model can then output the current and temperature of the individual cells in the fuel cell based on the input actual operating condition data, thereby further determining the current density distribution and temperature distribution of the individual cells in the fuel cell.

[0092] In this embodiment, the artificial neural network model is trained using test data under various test conditions to obtain a prediction model. This prediction model, along with the actual test conditions, can then be used to predict and determine the current and temperature of each cell within the fuel cell. In practical applications, by inputting the test conditions of each fuel cell into the prediction model, the current and temperature of each cell within the fuel cell can be obtained, thereby enabling the measurement of the temperature and current of each cell within the fuel cell.

[0093] This approach eliminates the need to configure and install an S++ device for current and temperature measurement on every fuel cell. Instead, a minimum of one S++ device is sufficient to test the current and temperature of the fuel cell under various operating conditions. Therefore, this prediction method can reduce equipment investment costs.

[0094] Furthermore, the prediction method in this embodiment can predict current and temperature under various operating conditions. When there is uneven current density and temperature distribution in a single cell, the operating parameters can be simulated and adjusted within the prediction model. For example, by adjusting the operating parameters input to the prediction model, the uniformity of the current density and temperature distribution of the single cell can be adjusted to the user's desired level. This allows the user to refer to the simulated operating parameters in the prediction model when actually adjusting the unevenly distributed single cell. This improves the convenience for users when adjusting the current and temperature of a single cell.

[0095] Furthermore, in the prediction method of this embodiment, the prediction model is an artificial neural network model trained using the K-fold cross-validation algorithm. By using test data for multiple cross-training and validation iterations, compared to using only the test data for a single training and validation, the number of training iterations can be increased, which helps reduce the probability of underfitting due to insufficient test data. Moreover, the weights and bias coefficients of the artificial neural network model are determined using the bat algorithm. By using the bat algorithm to find the optimal solution to adjust the weights and bias coefficients of the artificial neural network model, the difference between the model's prediction results and the actual results can be minimized, thus reducing test result error and improving the prediction accuracy of the prediction model.

[0096] In some exemplary implementations, step S110 determines a test condition combination based on the test range of the test equipment. For example, it may specifically include: determining a test condition combination based on a preset orthogonal array and the test range, wherein the test condition combination covers the test range of the test equipment.

[0097] For example, orthogonal experimental design is a design method that uses orthogonal arrays to efficiently screen combinations of multiple factors and levels of operating conditions. The core idea is to cover the main effects and interactions of all factors with the fewest number of trials.

[0098] Orthogonal arrays are a fundamental tool for orthogonal experimental design, denoted by L. For example, L9 represents an orthogonal array with a total of 9 trials.

[0099] In this embodiment, nine operating parameters affect the temperature and current density distribution of a single cell. These nine parameters are the nine factors in the orthogonal array. Assuming that each factor in this embodiment considers four levels, the number of levels refers to the number of different values ​​or states for each factor. For example, a level number of four means that each factor has at least four different values ​​or states considered, meaning that each factor can select four levels within its corresponding operating range, such as high, medium-high, medium-low, and low. As another example, if the anode metering ratio ranges from 1.2 to 1.8, the four levels that can be selected are 1.2, 1.4, 1.6, and 1.8.

[0100] Orthogonal arrays can be selected based on the number of factors and levels. For example, in this embodiment, the number of factors is 9, and the preset orthogonal array can be L40(8^8 10^1). Each column of the orthogonal array represents a factor, and each row represents an experimental combination. 8^8 indicates that there are 8 columns with 8 levels, and 10^1 indicates that there is 1 column with 10 levels. Each of the first 8 columns can accommodate one factor with 8 levels. The 9th column can accommodate one factor with 10 levels. Orthogonality is satisfied between any two columns, meaning the combinations of factor levels are evenly distributed, effectively reducing the number of experiments and allowing for the analysis of main effects. For example, this preset orthogonal array has 9 columns and 40 rows, requiring a total of 40 experiments.

[0101] Then, the values ​​of each level corresponding to each factor are filled into a preset orthogonal table, with each row corresponding to a test condition, thus obtaining 40 different test condition combinations. These 40 test condition combinations cover the test range of the test equipment. For example, the preset orthogonal table L40 (8^810^1) of this embodiment can be shown in Table 1 below.

[0102] Table 1 Preset Orthogonal Array

[0103] 0 0 0 0 0 1 0 7 7 0 2 5 7 1 6 7 7 4 0 3 1 1 7 2 1 5 1 0 5 3 6 3 3 4 6 9 0 6 2 3 6 7 1 0 8 1 1 7 2 4 0 7 4 7 1 2 7 4 6 5 5 6 0 1 4 4 7 5 1 4 1 1 1 5 4 1 2 4 2 6 6 1 7 2 4 1 4 0 0 4 2 0 6 2 3 7 6 1 2 2 3 6 5 0 2 2 1 5 2 4 1 4 4 5 3 5 3 2 5 7 1 5 6 4 7 8 2 7 4 6 6 1 3 3 9 3 0 2 5 7 0 5 3 5 3 1 1 7 2 7 7 0 6 3 3 0 2 5 6 1 2 2 3 6 3 3 0 0 6 2 3 3 6 5 0 3 3 2 4 0 4 1 2 7 4 7 2 6 0 4 1 4 4 7 4 6 0 3 4 4 7 5 2 2 1 0 2 4 6 6 0 5 3 7 2 6 4 7 5 2 0 4 5 1 5 5 0 3 1 1 5 3 1 9 5 2 0 6 2 2 4 5 8 5 3 6 3 3 1 3 7 3 5 4 1 2 7 6 2 3 5 5 7 1 5 4 3 6 3 2 6 0 5 3 6 0 0 2 4 6 2 3 6 5 0 2 4 6 6 3 3 0 2 5 4 3 1 6 5 0 3 1 1 5 4 0 6 6 0 5 3 4 7 6 7 7 1 5 4 1 3 1 2 8 7 2 4 1 4 7 4 4 9 7 4 6 6 0 6 1 7 1 7 5 2 0 6 2 7 5 4 7 7 7 7 7 5 0 5 7

[0104] In related technologies, the controlled variable method can be used to enumerate test condition combinations. For example, with 9 test condition parameters, each factor is calculated with at least 4 levels, which would require 4 to the power of 9, or 262,144 experiments. In actual testing, this would be difficult to complete due to the excessive time cost. In this embodiment, however, an orthogonal experimental design is used to select a small number of test condition combinations (40), which can cover the entire range of test conditions. Then, through a small number of test experiments (40 experiments), training data can be provided for the subsequent prediction model, effectively reducing the experimental testing time and cost.

[0105] After determining the test condition combinations, the S++ equipment can be used to measure the current and temperature of a single cell under each test condition. Furthermore, since the current and temperature differ under different dielectric densities, this embodiment measures the current and temperature of a single cell under each test condition for each dielectric density value. For example, if the dielectric density ranges from 0.1 to 1.6, the current and temperature under 40 test conditions are measured when the dielectric density is 0.1. The current and temperature under 40 test conditions are measured when the dielectric density is 0.2, and so on.

[0106] Once the current and temperature of a single cell under each test condition are measured, i.e., the test data for each test condition is obtained, a predictive model can be established based on the test data.

[0107] In some embodiments, in step S120 above, referring to Figure 2 Based on the operating data of each test condition and the test data of the current and temperature of the single cell corresponding to each test condition, a prediction model for predicting the current and temperature of the single cell is established, which may specifically include the following steps S121-S123.

[0108] Step S121: Combine the data of each working condition and the corresponding test data respectively, divide the combined data into training data and validation data, and preprocess the training data.

[0109] Specifically, in step S121, the test area of ​​the testing equipment can be evenly divided into multiple units according to a preset division rule. This preset division rule can be based on a preset number of rows and columns, dividing the test area into units equal to the preset number of rows * preset number of columns. For example, refer to... Figure 3 , Figure 3 An example of a test range being divided is shown. Figure 3 In the example, the preset number of rows is 11 and the preset number of columns is 17, so the test range is divided into 11*17 blocks, which is 187 units.

[0110] Each unit corresponds to a position coordinate (m, n), where m is the row number and n is the column number.

[0111] In step S121, the test condition data and test data are combined. Specifically, this may include: splitting the test data at the location of each unit; combining the test data at the location of each unit, the coordinates of that unit, the test condition corresponding to that test data, and the electrical density corresponding to that test data. For example, one of the combined data is: electrical density 0.1 - condition a1 - position (m1, n1) - test data z1. Wherein, the test data z1 is the current and temperature measured at the position m1th row and n1th column when the electrical density is 0.1 and the condition is a1.

[0112] After the data combination is completed, when determining the training data and verification data, it may specifically include: determining the combined data corresponding to each test condition determined by the preset orthogonal array in step S110 as the training data; selecting an additional set of test conditions from other conditions as verification test conditions, testing the test data corresponding to each electrical density under the verification test conditions, and combining the test data as verification data.

[0113] For example, in step S110 above, operating conditions 1-40 were selected using a preset orthogonal array. The test data corresponding to these operating conditions 1-40 were combined to serve as training data. Operating condition 41 was also selected. Under operating condition 41, a test data point was measured for each electrical density. The test data obtained under operating condition 41 were then combined to serve as verification data.

[0114] In addition, after the training data is determined, it needs to be preprocessed. Preprocessing can include data cleaning, standardization, and normalization to reduce noise, remove erroneous data, improve model accuracy, and speed up convergence.

[0115] After preprocessing the training data, the preprocessed training data can be uniformly input into the prediction model for training, so as to establish a prediction model for predicting the current and temperature of a single cell.

[0116] Step S122: Based on the preprocessed training data and validation data, the weights and bias coefficients of the artificial neural network model are determined using the bat algorithm. Then, the artificial neural network model is trained using the K-fold cross-validation algorithm until the optimal artificial neural network model is output.

[0117] Specifically, in step S122, for each unit within the test range, a corresponding artificial neural network model (hereinafter referred to as a unit neural network model) can be trained using the Bat Algorithm and K-fold cross-validation algorithm. These unit neural network models are then combined to obtain the optimal artificial neural network model. For example, if the test range is divided into 187 units, for each unit, using its training and validation data, a unit neural network model is trained using the Bat Algorithm and K-fold cross-validation algorithm, resulting in 187 unit neural network models. Combining these 187 unit neural network models yields the optimal artificial neural network model.

[0118] In step S122, as an exemplary implementation, combined with Figure 4 As shown, step S122, training the prediction model, may include steps S1221-S1224 as described below.

[0119] Step S1221: Initialize the parameters of the bat algorithm, establish a bat population, and determine the initial position of the bat population as the weights and bias coefficients of the artificial neural network model.

[0120] For example, in step S1221, the parameters of the bat algorithm are initialized to establish a bat population. Specifically, this may include setting the number of bats in the population n, the initial pulse intensity A, the pulse frequency range f, the initial pulse emission rate r, the pulse enhancement coefficient γ and the sound wave attenuation coefficient α, and randomly defining an initial position (x0, y0) for each bat in the bat population.

[0121] The Bat Algorithm is a swarm intelligence-based optimization algorithm that solves optimization problems by simulating the echolocation and hunting behavior of bats. Each individual bat represents a potential solution. Bats locate prey by emitting high-frequency sound waves and receiving echoes, searching for the optimal location in the solution space to find the optimal solution. During the search process, the fitness value of each bat's location is evaluated to find the optimal fitness value. The search continues towards the location corresponding to the optimal fitness value, gradually approaching the optimal position.

[0122] Among them, the pulse frequency f determines the search step size of the bat; the initial pulse intensity A controls the loudness of the sound wave; the initial pulse emission rate r represents the probability of emitting a pulse, for example, set between 0.5 and 1.0; the pulse enhancement coefficient γ is used to update the pulse intensity, for example, set to 0.9; and the sound wave attenuation coefficient α controls the sound wave attenuation rate, usually set to 0.9-0.98.

[0123] Specifically, in step S1221, when establishing bat populations, a bat population can be established for each unit. For example, if the test area is divided into 187 units, establishing a bat population for each unit is equivalent to establishing 187 bat populations in step S1221. For example, unit b corresponds to one bat population b. Each bat population includes multiple bats. For example, the bat population corresponding to unit b includes bats b1 to b20.

[0124] In step S1221, after establishing the bat population, the position of the bat population corresponding to the unit is determined as the weight and bias coefficient of the unit neural network model corresponding to the unit. By using each bat in the bat population corresponding to the unit, the optimal position is continuously searched to obtain the optimal weight and bias coefficient of the unit neural network model corresponding to the unit.

[0125] For example, for unit b, the position of bat population b is determined as the weight and bias coefficient of the unit neural network model b corresponding to unit b. The search is performed using bats b1 through b20 to find the optimal position, thereby obtaining the optimal weight and bias coefficient of the unit neural network model b. The implementation method for other units can be referenced from that of unit b, and will not be elaborated here.

[0126] More specifically, in step S1221, determining the location of the bat population as the weights and bias coefficients of the unit artificial neural network model may include:

[0127] For each unit, multiple small-unit neural network models are established based on the number of bats in the corresponding bat population, with each bat corresponding to a small-unit neural network model. The initial position of the bat is used as the initial weights and initial bias coefficients of the small-unit neural network model corresponding to that bat. For example, when training the unit neural network model b corresponding to unit b (located in row mb and column nb) using the bat population b1-b20, 20 small-unit neural network models are established for unit b. The initial position of bat bn is used as the initial weights and initial bias coefficients of the small-unit neural network model bn. For example, the initial position of bat b1 is used as the initial weights and initial bias coefficients of small-unit neural network model b1, and the initial position of bat b2 is used as the initial weights and initial bias coefficients of small-unit neural network model b2.

[0128] In artificial neural networks (ANNs), weights determine the strength of the input signal's influence on neuron activation; larger weights correspond to a stronger influence of the feature on the output. Bias is used to adjust the activation threshold of neurons, controlling their sensitivity. By determining the location of a bat population as the weights and bias coefficients of the artificial neural network model, it's equivalent to using the bat population to intelligently optimize the neural network's training process.

[0129] Step S1222: Using the preprocessed training data, train the artificial neural network model using the K-fold cross-validation algorithm to obtain the target training model.

[0130] Step S1223: Use the accuracy of the target training model as the fitness value of the bat algorithm, and iteratively update the position of the bats in the bat population. In each iteration, update the weights and bias coefficients of the artificial neural network model according to the updated bat position, and calculate the fitness value of the bat algorithm in each iteration.

[0131] Step S1224: When the iteration of the bat population reaches the preset requirements, determine and output the optimal artificial neural network model based on the target training model after the current iteration.

[0132] Specifically, continuing with the example in step S1221, when using the bat population of bats b1-b20 to train the unit neural network model corresponding to the prediction unit b in step S1221, the initial position of each bat is used as the initial weight and initial bias coefficient of the small unit neural network model corresponding to each bat. Then, step S1222 is entered to use the pre-processed training data corresponding to the unit b and the K-fold cross-validation algorithm to train each small unit neural network model corresponding to the unit b. After each small unit neural network model is trained, the target training model can be obtained, that is, the 20 target training models corresponding to the unit b are obtained.

[0133] For example, in step S1222, the training data corresponding to unit b (the test data at the position of row mb and column nb, the test conditions and electrical density corresponding to the test data) and the K-fold cross-validation algorithm are used to train each small unit neural network model (small unit neural network model b1-b20) to obtain 20 target training models (target training models b1-b20) corresponding to unit b.

[0134] In some embodiments, refer to Figure 5 In step S1222, the artificial neural network model is trained using training data and the K-fold cross-validation algorithm to obtain the target training model. Specifically, this process may include steps S410-S450.

[0135] Step S410: Divide the preprocessed training data into K folds of data on an average basis.

[0136] Step S420: Select one fold of data from the K-fold data as the validation set, and use the remaining folds of data as the training set.

[0137] Step S430: Train the artificial neural network model using the training set.

[0138] Step S440: Validate the trained artificial neural network model using a validation set and evaluate the performance of the trained artificial neural network model.

[0139] Repeat steps S420-S440 above, that is, repeat the selection of the validation set and training set, as well as the validation and evaluation process of the artificial neural network model, until the validation set has traversed every fold of data.

[0140] Step S450: Select the optimal model from the trained artificial neural network models and use it as the target training model.

[0141] Specifically, taking a small-unit neural network model as an example (e.g., small-unit neural network model b1), this explains the process of iteratively training this small-unit neural network model b1 using training data, K-fold cross-validation algorithm, and bat algorithm to obtain the target training model b1 corresponding to this small-unit neural network model b1:

[0142] In step S410, the training data is divided into K folds on average to obtain K-fold data. For example, the test data corresponding to each test condition and each electrical density of the unit is divided into 5 folds to obtain 5-fold data.

[0143] Then, in step S420, one fold of data is selected from the K-fold data in a preset order as the validation set, and the remaining folds of data are used as the training set. For example, the test data are sorted in a preset order, and the first fold of data is selected as the validation set, while the second, third, fourth, and fifth folds of data are used as the training set.

[0144] Then, in step S430, the artificial neural network model is trained using the training set selected in step S420, that is, the small unit neural network model b1 is trained. For example, the model is trained using the data from the second, third, fourth, and fifth folds.

[0145] After training is completed, proceed to step S440 to validate the artificial neural network model b1.1 trained in step S430 using the validation set selected in step S420, and evaluate the performance of the trained artificial neural network model b1.1. For example, the model trained in step S430 is validated using the first fold data to obtain the artificial neural network model b1.1, and the performance of the artificial neural network model b1.1 is evaluated.

[0146] In some exemplary embodiments, evaluating the performance of the trained artificial neural network model in step S440 may specifically include: calculating the accuracy of the trained artificial neural network model based on a validation set. This accuracy may specifically include the fitting coefficients and / or prediction accuracy of the artificial neural network model.

[0147] Specifically, the fitting coefficient (R²) 2 The parameter is used to quantify how well the model fits the training data. The closer its value is to 1, the better the model fits. Its value is: 1 - (sum of squares of model prediction errors / sum of squares of actual values ​​and their mean).

[0148] Prediction accuracy is a parameter used to measure a model's ability to predict new data. Commonly used parameters for prediction accuracy include mean squared error, root mean square error, and mean absolute error.

[0149] When calculating the accuracy of the trained artificial neural network model based on the validation set, the test conditions and electrical parameters in the validation set can be input into the trained artificial neural network model. Based on the prediction results output by the artificial neural network model and combined with the actual values ​​of the validation set, the fitting coefficient and prediction accuracy can be calculated.

[0150] After evaluating the performance of the trained artificial neural network model in step S440, it can be determined whether the validation set has traversed every fold of data. If the validation set has not traversed every fold of data, return to step S420, select the next fold of data as the validation set according to a preset order, use the remaining folds of data as the training set, and continue with steps S430 and S440. Use the current training set to train the model, and use the current validation set to calculate the performance of the artificial neural network model, until the validation set has traversed every fold of data. After the validation set has traversed every fold of data, proceed to step S450.

[0151] For example, in the first iteration, the first fold of data is selected as the validation set, and the second, third, fourth, and fifth folds of data are used as the training set for model training. After validating the trained model using the first fold of data, the artificial neural network model b1.1 is obtained, and the process returns to step S420. In the second iteration, the second fold of data can be selected as the validation set, and the first, third, fourth, and fifth folds of data can be selected as the training set. After training and validation, the artificial neural network model b1.2 is obtained, and the process returns to step S420 again. In the third iteration, the third fold of data can be selected as the validation set, and the first, second, fourth, and fifth folds of data can be selected as the training set. After training and validation, the artificial neural network model b1.3 is obtained, and the process returns to step S420 again. This loop continues until each of the five folds of data is used as a validation set to validate the trained model, resulting in the artificial neural network model b1.5. The loop then ends, and the process proceeds to step S450.

[0152] In step S450, the best-performing artificial neural network model among the trained models is identified as the optimal model and used as the target training model. For example, the best-performing artificial neural network model among the five models obtained after five training and validation iterations is used as the target training model. For instance, if artificial neural network model b1.2 has the best performance, then artificial neural network model b1.2 is the target training model b1 corresponding to the small unit neural network model b1.

[0153] Furthermore, for each unit, the neural network models corresponding to each small unit are trained through steps S410-S450 to obtain their respective target training models. Then, step S1223 is entered to calculate the accuracy of each target training model and use each accuracy as the fitness value of its corresponding bat. For example, for unit b, the 20 small unit neural network models corresponding to bats b1-b20 are trained through steps S410-S450 to obtain 20 target training models. Then, the accuracy of target training model b1 is calculated; the accuracy of target training model b1 is used as the fitness value of bat b1. The accuracy of target training model b2 is calculated and used as the fitness value of bat b2, and so on, to calculate the accuracy of target training model bn, which is the fitness value of bat bn.

[0154] It is worth noting that the accuracy of the target training model can be the fitting coefficient and / or prediction accuracy determined in step S440 above, which will not be elaborated here.

[0155] Next, the bat with the best fitness value is selected from among all bats to find the optimal bat position. Then, based on the current optimal bat position, the positions of each bat are iterated. The iterated bat positions are used as new weights and bias coefficients for their respective small-unit neural network models. This process is iterated continuously, using multiple bats to continuously search for the optimal bat position, thus continuously determining the optimal weights and bias coefficients of the unit neural network model. After each iteration, the bat positions gradually approach the optimal position, causing the weights and bias coefficients of the neural network model to also approach their optimal values.

[0156] In some embodiments, the formula for updating the bat's location can specifically be the following formula:

[0157]

[0158] Where t is the current time, v i Let x be the speed of the i-th bat. * f is the current globally optimal position. i Let x be the frequency emitted by the i-th bat, and x be the bat's position.

[0159] Where, x *This refers to the position of the bat with the optimal fitness value determined based on the fitness values ​​of each bat after the current iteration. For example, among bats b1-b10, bat b2 has the optimal fitness value. The current position of bat b2 is taken as the current global optimal position and substituted into the above formula to update the positions of bats b1-b20, thereby updating the weights and bias coefficients of each small unit neural network model. Then, iterative training is performed again according to steps S1222 and S1223 until the iteration of the bat population reaches the preset requirements.

[0160] When the iteration of the bat population corresponding to unit b reaches the preset requirement, proceed to step S1224, compare the accuracy of the 20 target training models corresponding to the currently iterated unit b, and select the target training model with the best accuracy. The best target training model is the trained unit neural network model b of unit b.

[0161] In some embodiments, the preset requirements include the bat population reaching a maximum number of iterations, or the target training model reaching a preset accuracy standard. The maximum number of iterations and the preset accuracy standard can be set according to actual circumstances and will not be elaborated here.

[0162] For example, using 20 bats for searching, and obtaining 20 target training models after several iterations, it is determined whether the accuracy of the 20 target training models meets the preset accuracy standard, and whether the number of iterations has reached the maximum number of iterations. If the accuracy of any target training model meets the preset accuracy standard, or the number of iterations reaches the maximum number of iterations, it indicates that the iteration of the bat population has met the preset requirements. Then, in step S1224, the accuracy of these 20 target training models is compared, and the target training model with the best accuracy is selected. This optimal target training model is the trained unit neural network model corresponding to unit b.

[0163] For the unit neural network models of other units, training is performed simultaneously according to steps S1221-S1223 above to obtain the optimal target training model for each unit, i.e., the trained unit neural network model. Then, in step S1224, the iteratively obtained optimal target training models for each unit, i.e., the trained unit neural network models, are combined to obtain the optimal artificial neural network model, which is then output. This output optimal artificial neural network model serves as the prediction model for predicting the current and temperature of a single battery.

[0164] Step S123: Use the output optimal artificial neural network model as a prediction model for predicting the current and temperature of a single cell.

[0165] In the above embodiments, after obtaining the optimal artificial neural network model using the bat algorithm and K-fold cross-validation, the optimal artificial neural network model is the prediction model for predicting the current and temperature of a single battery.

[0166] Continue as Figure 6 As shown, Figure 6 The prediction process of the prediction model used to predict the current and temperature of a single cell is shown. Figure 6 The left-hand block diagram includes nine operating parameters and the coordinates of the location points. These nine parameters and coordinates are input into the trained prediction model in the middle block diagram, allowing the model to output the current and temperature of the single battery. Figure 6 The middle block diagram in the image illustrates the architecture of the predictive model (ANN model), which includes an input layer, hidden layers, and an output layer. The input layer receives operating parameters such as gas stoichiometry, temperature, and pressure; the hidden layer consists of multiple layers of nodes and is responsible for processing and learning the complex relationships between the data; the output layer outputs the prediction results.

[0167] In this embodiment, the artificial neural network model is trained using test data under various test conditions to obtain a prediction model. This prediction model, along with the actual operating conditions, can then be used to predict and determine the current and temperature of individual cells within the fuel cell. In practical applications, the operating conditions of each fuel cell can be input into the prediction model to obtain the current and temperature of individual cells within each fuel cell, thereby achieving the measurement of the temperature and current of individual cells in each fuel cell. This approach eliminates the need to configure and install an S++ device for each fuel cell; at a minimum, only one S++ device is required to test the current and temperature of the fuel cell under various operating conditions and obtain test data. Therefore, this prediction method can reduce equipment investment costs.

[0168] Furthermore, in the prediction method of this embodiment, the prediction model can be used to predict current and temperature under various operating conditions. When there is uneven current and temperature distribution in a single cell, the operating parameters can be simulated and adjusted within the prediction model. For example, by adjusting the operating parameters input to the prediction model, the uniformity of the current and temperature distribution of the single cell can be adjusted to the user's desired level. This way, when actually adjusting a single cell with uneven distribution, the user can refer to the operating parameters simulated in the prediction model for actual adjustments, without needing to adjust the actual operating conditions based on experience. This improves the convenience for users when adjusting the current and temperature of a single cell.

[0169] Furthermore, in the prediction method of this embodiment, the prediction model is an artificial neural network model trained using the K-fold cross-validation algorithm. By using test data for multiple cross-training and validation iterations, compared to using only the test data for a single training and validation, the number of training iterations can be increased, which helps reduce the probability of underfitting due to insufficient test data. Moreover, the weights and bias coefficients of the artificial neural network model are determined using the bat algorithm. By using the bat algorithm to find the optimal solution to adjust the weights and bias coefficients of the artificial neural network model, the difference between the model's prediction results and the actual results can be minimized, thus reducing test result error and improving the prediction accuracy of the prediction model.

[0170] Furthermore, in the prediction method of this embodiment, an orthogonal experimental design is used to select a small number of test condition combinations (e.g., the 40 combinations in the above embodiment) to cover the entire test condition range. Then, through a small number of test experiments (40 experiments), training data can be provided for the subsequent prediction model. This avoids the need to enumerate test condition combinations using the controlled variable method as in related technologies (enumerating the equivalent of 4 to the power of 9, or 262,144 combinations, corresponding to 262,144 experiments), effectively reducing experimental testing time and cost.

[0171] Furthermore, in the prediction method of this embodiment, when training the prediction model, it is also possible to train a separate ANN model for each location point, and combine the prediction models of the individual points to form the entire prediction model of the single cell. This method has higher reliability and is conducive to improving prediction accuracy and reliability.

[0172] Example 2

[0173] This embodiment relates to a device for predicting the operating parameters of a fuel cell, which can implement the prediction method in Embodiment 1 to predict the current and temperature of a single cell within the fuel cell under different operating conditions. Furthermore, combined with... Figure 7 As shown, the prediction device comprises a test module 510, a model building module 520, and a prediction module 530.

[0174] The aforementioned test module 510 is used to determine the combination of test conditions based on the test range of the test equipment, and to test the current and temperature of the single cell under each test condition using the test equipment. Furthermore, the aforementioned test equipment is used to measure the current and temperature of the single cell within the fuel cell. In specific implementation, as mentioned in Embodiment 1, the test equipment can, for example, be an S++ device.

[0175] The aforementioned model building module 520 is used to establish a predictive model for predicting the current and temperature of a single battery based on the operating data of each test condition and the test data of the current and temperature of the single battery corresponding to each test condition.

[0176] The aforementioned prediction module 530 is used to predict the current and temperature of individual cells within the fuel cell based on actual operating data during fuel cell operation, using a prediction model. The prediction model is an artificial neural network model trained using a K-fold cross-validation algorithm, and the weights and bias coefficients of the artificial neural network model are determined using the bat algorithm.

[0177] In detail, in the specific implementation of the prediction device of this embodiment, the above-mentioned modules can be existing module products with data transmission, storage or computing functions. At the same time, the above-mentioned modules in this embodiment can be set up separately, or preferably, they can be integrated into electronic devices.

[0178] In practical applications, the prediction device of this embodiment establishes an artificial neural network model and uses the artificial neural network model and actual operating conditions to predict the operating parameters of the fuel cell. For details, please refer to the relevant description in Embodiment 1, which will not be repeated here.

[0179] The prediction device of this embodiment, through the above modules and by executing the prediction method in Embodiment 1, can predict and determine the current and temperature of individual cells within a fuel cell using a prediction model and actual operating conditions. It eliminates the need to configure and install an S++ device for each fuel cell; at a minimum, only one S++ device is required to test the current and temperature of individual cells within the fuel cell under various operating conditions to obtain test data. Therefore, the prediction device of this embodiment can reduce equipment investment costs.

[0180] Example 3

[0181] This embodiment relates to an electronic device, combined with Figure 8 As shown, the electronic device includes a memory 610 and a processor 620. The memory 610 stores computer-readable instructions that, when executed by the processor 620, enable the prediction method for fuel cell operating parameters in Embodiment 1.

[0182] The electronic device in this embodiment, by executing the prediction method in Embodiment 1, can predict and determine the current and temperature of individual cells within a fuel cell using a prediction model and actual operating conditions. It eliminates the need to configure and install an S++ device for each fuel cell; at a minimum, only one S++ device is required to test the current and temperature of the fuel cell under various operating conditions, thus reducing equipment investment costs.

[0183] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting fuel cell operating parameters, used to predict the current and temperature of a single cell within a fuel cell under different operating conditions, characterized in that, The prediction method includes: Based on the testing range of the testing equipment, a combination of testing conditions is determined, and under each testing condition, the current and temperature of the single battery are tested using the testing equipment. Based on the operating data of each test condition and the test data of the current and temperature of the single battery corresponding to each test condition, a prediction model for predicting the current and temperature of the single battery is established. During the operation of the fuel cell, the current and temperature of the individual cells within the fuel cell are predicted using the prediction model based on the actual operating data of the fuel cell. The testing equipment is used to measure the current and temperature of the single cell in the fuel cell, and the prediction model is an artificial neural network model trained using the K-fold cross-validation algorithm, with the weights and bias coefficients of the artificial neural network model determined by the bat algorithm.

2. The method for predicting fuel cell operating parameters according to claim 1, characterized in that, The determination of test condition combinations based on the test range of the test equipment includes: Based on the preset orthogonal array and the operating condition test range of the test equipment, the test condition combination is determined, and the test condition combination covers the operating condition test range of the test equipment.

3. The method for predicting fuel cell operating parameters according to claim 1, characterized in that, The step of establishing a predictive model for predicting the current and temperature of a single battery based on the operating condition data of each of the test conditions and the test data of the current and temperature of the single battery corresponding to each of the test conditions includes: The working condition data and the corresponding test data are combined respectively, and the combined data are divided into training data and validation data. The training data is then preprocessed. Based on the preprocessed training data and the validation data, the weights and bias coefficients of the artificial neural network model are determined by the bat algorithm, and then the artificial neural network model is trained by the K-fold cross-validation algorithm until the optimal artificial neural network model is output. The output of the optimal artificial neural network model is used as the prediction model for predicting the current and temperature of the single cell.

4. The method for predicting fuel cell operating parameters according to claim 3, characterized in that, After determining the weights and bias coefficients of the artificial neural network model using the bat algorithm, the artificial neural network model is trained using the K-fold cross-validation algorithm until the optimal artificial neural network model is output, including: Initialize the parameters of the bat algorithm, establish a bat population, and determine the initial position of the bat population as the weights and bias coefficients of the artificial neural network model; Using the preprocessed training data, the artificial neural network model is trained using the K-fold cross-validation algorithm to obtain the target training model; The accuracy of the target training model is used as the fitness value of the bat algorithm, and the position of the bats in the bat population is updated iteratively. In each iteration, the weights and bias coefficients of the artificial neural network model are updated according to the updated bat position, and the fitness value of the bat algorithm is calculated in each iteration. When the iteration of the bat population reaches the preset requirements, the optimal artificial neural network model is determined and output based on the target training model after the current iteration.

5. The method for predicting fuel cell operating parameters according to claim 4, characterized in that, The step of training the artificial neural network model using the preprocessed training data and employing a K-fold cross-validation algorithm to obtain the target training model includes: The preprocessed training data is divided into K folds of data on average. Select one fold of data from the K-fold data as the validation set, and use the remaining folds of data as the training set; The artificial neural network model is trained using the training set; The trained artificial neural network model is validated using the validation set, and the performance of the trained artificial neural network model is evaluated. Repeat the selection of the validation set and the training set, as well as the validation and evaluation process of the artificial neural network model, until the validation set has traversed every fold of data. The optimal model is selected from the trained artificial neural network models and used as the target training model.

6. The method for predicting fuel cell operating parameters according to claim 5, characterized in that, The evaluation of the performance of the trained artificial neural network model includes: Based on the validation set, the accuracy of the trained artificial neural network model is calculated; The accuracy of the artificial neural network model includes the fitting coefficient and / or prediction accuracy of the artificial neural network model.

7. The method for predicting fuel cell operating parameters according to claim 4, characterized in that: The preset requirements include the bat population reaching the maximum number of iterations, or the target training model reaching a preset accuracy standard.

8. The method for predicting fuel cell operating parameters according to claim 4, characterized in that: The velocity update formula for the bats in the bat population is: The formula for updating the position of the bats in the bat population is: Where t is the current time, v i Let x be the speed of the i-th bat. * f is the current globally optimal position. i Let be the frequency emitted by the i-th bat.

9. A device for predicting the operating parameters of a fuel cell, used to predict the current and temperature of a single cell within the fuel cell under different operating conditions, characterized in that, The prediction device includes: The test module (510) is used to determine the test condition combination according to the test condition test range of the test equipment, and to test the current and temperature of the single battery under each test condition by means of the test equipment. The model building module (520) is used to establish a prediction model for predicting the current and temperature of the single battery based on the working condition data of each test condition and the test data of the current and temperature of the single battery corresponding to each test condition. The prediction module (530) predicts the current and temperature of the single cell in the fuel cell based on the actual operating conditions data of the fuel cell during operation. The testing equipment is used to measure the current and temperature of the single cell in the fuel cell, and the prediction model is an artificial neural network model trained using the K-fold cross-validation algorithm, with the weights and bias coefficients of the artificial neural network model determined by the bat algorithm.

10. An electronic device, characterized in that, include: Memory (610) stores computer programs; Processor (620); When the computer program is executed by the processor (620), it implements the method for predicting fuel cell operating parameters according to any one of claims 1 to 8.

Images