Fuel cell performance prediction method and device and computer program product

Abstract

The invention discloses a fuel cell performance prediction method and device and a computer program product. The method comprises the following steps: acquiring real-time operation data corresponding to a fuel cell in a current time period; based on the real-time operation data, target voltage data corresponding to the fuel cell is obtained by adopting a cell performance prediction model, the cell performance prediction model is obtained through machine learning based on historical operation data in a historical time period and historical voltage data corresponding to the historical operation data, and the historical operation data comprises multiple operation parameters; the multiple operation parameters are screened out from the multiple input parameters based on historical voltage data. According to the invention, the technical problem of low accuracy of a battery performance prediction result caused by many and complex operation parameters used for constructing a model in related technologies is solved.

Description

Technical Field

[0001] The present invention relates to the field of battery performance prediction, and in particular to a fuel cell performance prediction method, device and computer program product. Background Art

[0002] The performance of the fuel cell output voltage is related to the operating parameters. The output voltage of the fuel cell directly affects the life of the battery. In order to ensure that the fuel cell can operate efficiently, safely and stably for a long time, it is necessary to monitor the voltage performance of the fuel cell in actual work in a timely manner.

[0003] In the related art, the battery performance prediction model constructed uses many and complex operating parameters, and fails to filter data such as operating parameters, resulting in inaccurate battery performance prediction models, which in turn leads to inaccurate battery performance prediction results.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] Embodiments of the present invention provide a fuel cell performance prediction method, device and computer program product to at least solve the technical problem of low accuracy of battery performance prediction results caused by the large number and complexity of operating parameters used in building a model in the related art.

[0006] According to one aspect of an embodiment of the present invention, a fuel cell performance prediction method is provided, comprising: obtaining real-time operating data corresponding to a fuel cell in a current time period; based on the real-time operating data, using a battery performance prediction model to obtain target voltage data corresponding to the fuel cell, wherein the battery performance prediction model is based on historical operating data in a historical time period and historical voltage data corresponding to the historical operating data, obtained through machine learning, the historical operating data packet includes multiple operating parameters, and the multiple operating parameters are screened from multiple input parameters based on the historical voltage data.

[0007] Optionally, before obtaining the target voltage data corresponding to the fuel cell by using a battery performance prediction model based on the real-time operating data, the method also includes: obtaining the historical voltage data of the fuel cell within the historical period; obtaining parameter values ​​corresponding to multiple input parameters that affect the operation of the fuel cell within the historical period; determining the correlation coefficients between the multiple input parameters and the voltage of the fuel cell based on the historical voltage data and the parameter values ​​corresponding to the multiple input parameters; screening out the multiple operating parameters from the multiple input parameters based on the correlation coefficients between the multiple input parameters and the voltage of the fuel cell; and obtaining the historical operating data based on the parameter values ​​corresponding to the multiple operating parameters.

[0008] In the above manner, when more eigenvalues ​​are input into the battery performance prediction model, high noise will result, which will reduce the accuracy of the model. Therefore, it is necessary to screen out eigenvalues ​​with greater correlation with historical voltage data from the various eigenvalues ​​in the historical operation data as input for constructing the battery performance prediction model, thereby increasing the correlation between the input and output of the battery performance prediction model and effectively obtaining more accurate output prediction results (target voltage data).

[0009] Optionally, in the case where the historical period includes multiple sampling moments, the correlation coefficients between the multiple input parameters and the voltage of the fuel cell are determined based on the historical voltage data and the parameter values ​​corresponding to the multiple input parameters, including: obtaining the correlation coefficient between any one of the multiple input parameters and the voltage of the fuel cell in the following manner: obtaining multiple parameter values ​​corresponding to any one of the input parameters collected at the multiple sampling moments, and multiple historical voltage data, wherein the multiple parameter values ​​correspond to the multiple sampling moments one-to-one, and the multiple historical voltage data correspond to the multiple sampling moments one-to-one; determining a voltage average based on the multiple historical voltage data; determining the covariance between the multiple parameter values ​​and the voltage average; determining the standard deviation between the multiple parameter values ​​and the voltage average; determining the correlation coefficient between any one of the input parameters and the voltage of the fuel cell based on the covariance and the standard deviation; obtaining the correlation coefficients between the multiple input parameters and the voltage of the fuel cell by adopting the method of obtaining the correlation coefficient between any one of the input parameters and the voltage of the fuel cell.

[0010] In the above way, by calculating the correlation coefficient between each input parameter and the historical voltage data, those features that have little impact on the battery performance prediction model can be excluded, thereby reducing the interference of invalid input parameters. After screening by correlation analysis, the data is processed for dimensionality reduction, the model is simplified, and the calculation efficiency is improved.

[0011] Optionally, obtaining the historical voltage data of the fuel cell within the historical time period includes: obtaining a plurality of historical voltage data corresponding to a plurality of sampling moments of the fuel cell within the historical time period, wherein the plurality of sampling moments correspond one-to-one to the plurality of voltage data; preprocessing the voltage data to obtain a plurality of historical voltage data corresponding to the plurality of sampling moments, wherein the preprocessing includes at least one of the following: erroneous data screening, missing data value completion, and the plurality of sampling moments correspond one-to-one to the historical voltage data.

[0012] Through the above method, the historical voltage data is screened and the data with large deviations in the actual collected data is removed, which can effectively achieve the accuracy of data acquisition, fill in the missing values ​​of the historical voltage data, and effectively ensure the time series continuity of the historical battery data when the fuel cell is in operation.

[0013] Optionally, the battery performance prediction model includes an input layer, a hidden layer and an output layer, wherein the input layer is used to encode the real-time operation data to obtain first data; the hidden layer is used to perform nonlinear conversion on the first data to obtain nonlinear characteristics of the first data, and feature matching is performed on the nonlinear characteristics to obtain second data; and the output layer is used to decode the second data to obtain the target voltage data.

[0014] Through the above method, real-time operating data is used as input, and based on the battery performance prediction model, the target voltage data is output, which effectively realizes the prediction of battery performance. Based on the above performance prediction, the voltage performance of the fuel cell in actual work can be monitored in time, thereby effectively ensuring that the fuel cell can operate efficiently, safely and stably over a long period of time.

[0015] Optionally, the hidden layer includes a forward long short-term memory neural network, a reverse long short-term memory neural network and a fully connected layer, wherein the forward long short-term memory neural network is used to obtain long-term dependency features in the first data in a first direction to obtain a first feature, wherein the first direction is a direction from the first data to the last data of the first data; the reverse long short-term memory neural network is used to obtain short-term dependency features in the first data in a second direction to obtain a second feature, wherein the second direction is the opposite direction of the first direction; based on preset weights, the fully connected layer is used to perform nonlinear mapping on the first feature and the second feature to obtain the nonlinear feature, and feature matching processing is performed on the nonlinear feature to obtain the second data.

[0016] Through the above method, a forward long short-term memory neural network, a reverse long short-term memory neural network and a fully connected layer are used to obtain features in a first direction of the first data, and at the same time, features in a second direction of the first data are obtained, so that more comprehensive nonlinear features can be obtained, and all nonlinear features in the historical operation data can be effectively extracted.

[0017] Optionally, the fully connected layer includes a first fully connected layer, a second fully connected layer, a third fully connected layer and a fourth fully connected layer, wherein, based on the preset weights, the first fully connected layer is used to perform nonlinear mapping on the first feature and the second feature to obtain the nonlinear feature; the second fully connected layer is used to perform feature matching on the nonlinear feature to determine the predicted value of the second data; the third fully connected layer is used to determine the loss value between the predicted value and the historical voltage data; based on the loss value, the fourth fully connected layer is used to adjust the predicted value to obtain the second data.

[0018] In the above manner, the fully connected layer is set at the end of the battery performance prediction model to combine and integrate the first feature and the second feature extracted from the forward long short-term memory neural network and the reverse long short-term memory neural network, thereby achieving the effect of forming the final output based on the first feature and the second feature, and performing reverse adjustment through multiple fully connected layers, thereby effectively ensuring the accuracy of the output second data.

[0019] According to another aspect of an embodiment of the present invention, a fuel cell performance prediction device is also provided, including: an acquisition module, used to acquire real-time operating data corresponding to the fuel cell in a current time period; a prediction module, used to obtain target voltage data corresponding to the fuel cell based on the real-time operating data and using a battery performance prediction model, wherein the battery performance prediction model is based on historical operating data in a historical time period and historical voltage data corresponding to the historical operating data, and is obtained through machine learning, and the historical operating data includes multiple operating parameters, and the multiple operating parameters are screened from multiple input parameters based on the historical voltage data.

[0020] According to another aspect of an embodiment of the present invention, a non-volatile storage medium is provided, wherein the non-volatile storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing any one of the fuel cell performance prediction methods.

[0021] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the steps of any one of the fuel cell performance prediction methods are implemented.

[0022] In an embodiment of the present invention, real-time operating data corresponding to a fuel cell in a current time period is acquired; based on the real-time operating data, a battery performance prediction model is used to obtain target voltage data corresponding to the fuel cell, wherein the battery performance prediction model is based on historical operating data in a historical time period and historical voltage data corresponding to the historical operating data, obtained through machine learning, the historical operating data packet includes multiple operating parameters, and the multiple operating parameters are screened out from multiple input parameters based on the historical voltage data, thereby achieving the purpose of screening out multiple operating parameters with a greater correlation with the historical voltage data from multiple input parameters to construct a battery performance prediction model, thereby achieving the technical effect of improving the accuracy of the prediction results of the battery performance prediction model, and further solving the technical problem of low accuracy of battery performance prediction results caused by the large number and complexity of operating parameters used in constructing the model in the related technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0024] Figure 1 is a flow chart of a fuel cell performance prediction method according to an embodiment of the present invention;

[0025] Figure 2 is a diagram of a data selection method of a fuel cell performance prediction method according to an embodiment of the present invention;

[0026] Figure 3 is a schematic diagram of an optional battery performance prediction model structure according to an embodiment of the present invention;

[0027] Figure 4 is a flow chart of an optional fuel cell performance prediction method according to an embodiment of the present invention;

[0028] Figure 5 Schematic diagram of a fuel cell performance prediction device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] The operating status and working performance of fuel cells are two important characteristics related to the practical application of fuel cells. The operation of fuel cells involves complex coupled electrochemical processes, and it is difficult to fully grasp their reaction mechanisms; among them, water and heat management in fuel cells is one of the key factors affecting the performance and life of fuel cells, mainly including the temperature, pressure and flow of anode and cathode gases, coolant temperature and pressure, and insulation measures in harsh environments. With the enrichment of hydrogen energy product lines, traditional data analysis methods have long cycles, high costs, and low data utilization. Therefore, there is an urgent need to improve the utilization rate of historical data corresponding to fuel cells based on data-driven, monitor the working performance of fuel cells, and determine the correlation between parameter changes and fuel cell voltage performance.

[0032] First, to facilitate understanding of the embodiments of the present invention, some terms or nouns involved in the present invention are explained below:

[0033] Bidirectional Long Short-Term Memory (BILSTM) is a deep learning model structure used to process sequence data. BILSTM combines the characteristics of Bidirectional RNN and Long Short-Term Memory (LSTM), and can simultaneously consider the past and future information of the input sequence, thereby better capturing the long-distance dependencies in the sequence.

[0034] A fully connected layer is a type of layer in a deep learning neural network, also known as a densely connected layer or a fully connected layer. In a fully connected layer, each neuron is connected to all neurons in the previous layer, and each input value is multiplied by the weight of each neuron, and then the results are added and processed through an activation function. Fully connected layers usually appear in the last few layers of a neural network. They can integrate and process the features extracted by the previous convolutional layer or pooling layer, and finally output the prediction results of the network. Fully connected layers can also increase the nonlinear fitting ability of the network.

[0035] According to an embodiment of the present invention, a method embodiment of a fuel cell performance prediction is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0036] Figure 1 is a flow chart of a fuel cell performance prediction method according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0037] Step S102, obtaining real-time operation data corresponding to the fuel cell in the current period;

[0038] Optionally, the real-time operating data corresponding to the fuel cell refers to the parameters related to the operation of the fuel cell, such as the speed of the pump, the input pressure of the hydrogen fuel, and the input flow rate of the hydrogen fuel during the chemical reaction when the fuel cell is running and hydrogen fuel is input into the chemical reactor through a pump.

[0039] Optionally, adjusting the above real-time operating parameters, i.e. changing the hydrogen fuel input pressure and hydrogen fuel input flow rate, can change the chemical reaction rate in the chemical reactor, thereby changing the output voltage of the fuel cell. By obtaining the real-time operating data corresponding to the fuel cell in the current period, the output voltage of the fuel cell in the current period can be predicted, thereby determining the performance of the fuel cell.

[0040] Step S104, based on the real-time operation data, a battery performance prediction model is used to obtain target voltage data corresponding to the fuel cell, wherein the battery performance prediction model is based on historical operation data in a historical period and historical voltage data corresponding to the historical operation data, and is obtained through machine learning, and the historical operation data includes multiple operation parameters, and the multiple operation parameters are based on the historical voltage data and are screened from multiple input parameters.

[0041] Optionally, a battery performance prediction model is constructed, with real-time operating data as input to obtain target voltage data corresponding to the real-time operating data, wherein the target voltage data can intuitively reflect the voltage performance of the battery, thereby achieving the effect of real-time monitoring of fuel cell performance, and effectively taking timely response measures in the event of battery performance failure, so as to ensure that the fuel cell can operate efficiently, safely and stably over a long period of time.

[0042] In an optional embodiment, before obtaining the target voltage data corresponding to the fuel cell by using a battery performance prediction model based on real-time operating data, the method also includes: obtaining historical voltage data of the fuel cell in a historical period; obtaining parameter values ​​corresponding to multiple input parameters that affect the operation of the fuel cell in the historical period; determining the correlation coefficients between the multiple input parameters and the voltage of the fuel cell based on the historical voltage data and the parameter values ​​corresponding to the multiple input parameters; screening out multiple operating parameters from the multiple input parameters based on the correlation coefficients between the multiple input parameters and the voltage of the fuel cell; and obtaining historical operating data based on the parameter values ​​corresponding to the multiple operating parameters.

[0043] Optionally, a battery performance prediction model is constructed based on historical operating data and historical voltage data. Specifically, since the historical operating data that affect the voltage output are of various types and complex, after more historical operating data are input into the battery performance prediction model, the battery performance prediction model cannot determine the data with a high correlation with the historical voltage data. Therefore, it is necessary to screen out the input parameters with a high correlation with the historical voltage data from the various input parameters in the historical operating data as the input for constructing the battery performance prediction model, thereby increasing the correlation between the input and output of the battery performance prediction model, and effectively obtaining a more accurate output prediction result (target voltage data).

[0044] Specifically, the correlation coefficients between the parameter values ​​corresponding to the various input parameters and the historical voltage data are determined. The larger the correlation coefficient, the greater the correlation between the input parameters and the historical voltage data, that is, the input parameter has a greater impact on the output historical voltage data. Using input parameters with larger correlation coefficients as historical operating data to construct a battery performance prediction model can effectively improve the accuracy of the construction of the battery performance prediction model and the accuracy of the output prediction results.

[0045] In an optional embodiment, when a historical period includes multiple sampling moments, the correlation coefficients between the multiple input parameters and the voltage of the fuel cell are determined based on the historical voltage data and the parameter values ​​corresponding to the multiple input parameters, including: obtaining the correlation coefficient between any input parameter and the voltage of the fuel cell among the multiple input parameters in the following manner: obtaining multiple parameter values ​​corresponding to any input parameter collected at multiple sampling moments, and multiple historical voltage data, wherein the multiple parameter values ​​correspond to the multiple sampling moments one-to-one, and the multiple historical voltage data correspond to the multiple sampling moments one-to-one; determining the voltage average value based on the multiple historical voltage data; determining the covariance between the multiple parameter values ​​and the voltage average value; determining the standard deviation between the multiple parameter values ​​and the voltage average value; determining the correlation coefficient between any input parameter and the voltage of the fuel cell based on the covariance and the standard deviation; obtaining the correlation coefficients between the multiple input parameters and the voltage of the fuel cell by adopting the method of obtaining the correlation coefficient between any input parameter and the voltage of the fuel cell.

[0046] Optional, such as Figure 2 As shown, Figure 2 2 is a diagram of data selection for a fuel cell performance prediction method according to an embodiment of the present invention. Since there are many historical operating data related to historical voltage data, the data set range for developing a battery performance prediction model can be preliminarily selected based on expert knowledge through expert scoring methods and other means. Specifically, seven operating parameters (Blowout Preventer, BOP) are screened out, namely, air pressure entering the stack, hydrogen pressure entering the stack, coolant pressure entering the stack, coolant temperature entering the stack, air flow, water pump speed, hydrogen pump speed. In addition, there is one operating condition parameter, namely, load current, and one environmental parameter, namely, ambient temperature. Based on the parameter values ​​corresponding to the above nine input parameters and the correlation coefficient target between the historical voltage data, the historical operating data is determined.

[0047] Specifically, the hundreds of collected data are carefully classified according to the BOP operating parameters, status, environment and working conditions. The Pearson correlation coefficient between the various input parameters and the voltage of the fuel cell can be calculated by the Pearson correlation coefficient method as the correlation coefficient between the various input parameters and the voltage of the fuel cell. By calculating the correlation coefficient between each input parameter and the historical voltage data, those features that have little effect on the battery performance prediction model can be excluded, thereby reducing the interference of invalid input parameters. After secondary screening of correlation analysis, the data is processed for dimensionality reduction, the model is simplified, the calculation efficiency is improved, and the parameter values ​​corresponding to the input parameters that have a greater impact on performance are retained to determine the historical operation data.

[0048] Optionally, before calculating the correlation coefficient, based on expert knowledge, a plurality of parameters can be preliminarily screened by professionals in the field, and a plurality of parameters unrelated to voltage can be screened, a plurality of input parameters related to voltage can be retained, and a secondary screening feature can be added to the input parameters related to voltage. A correlation analysis algorithm is used, that is, the Pearson correlation coefficient (Pearson) between the preliminarily screened features and the voltage of the fuel cell is calculated, and the preliminarily screened features are further screened to obtain a plurality of operating parameters with a large correlation with the voltage of the fuel cell.

[0049] Optionally, the specific method for calculating the Pearson correlation coefficient is as follows:

[0050] The covariance between any input parameter and the voltage is calculated as follows: Where cov(x,y) is the covariance between any input parameter and the voltage, x i is the parameter value corresponding to any input parameter at any sampling time, is the average value of any input parameter, y i is the historical voltage data of the fuel cell voltage at any sampling time, is the average voltage corresponding to the fuel cell voltage.

[0051] The standard deviation of any input parameter is calculated as follows: Where S1 is the standard deviation of any input parameter, x i is the parameter value corresponding to any input parameter at any sampling time, is the average value of any input parameter.

[0052] The standard deviation of the voltage is calculated as follows: Where S2 is the standard deviation of voltage, y i is the historical voltage data of the fuel cell voltage at any sampling time, is the average voltage corresponding to the fuel cell voltage.

[0053] The Pearson correlation coefficient between any input parameter and the voltage of the fuel cell is calculated as follows: W = cov(x, y) / (S1*S2).

[0054] Specifically, the value of the Pearson correlation coefficient ranges between -1 and 1, where 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no linear relationship. The Pearson correlation coefficient is used to measure the relationship between two continuous variables and is calculated by calculating the covariance between the two variables (input parameter and voltage) divided by the product of their standard deviations.

[0055] In an optional embodiment, historical voltage data of a fuel cell in a historical time period is obtained, including: obtaining multiple initial voltage data corresponding to multiple sampling moments of the fuel cell in the historical time period, wherein the multiple sampling moments correspond one-to-one to the multiple initial voltage data; preprocessing the multiple initial voltage data to obtain multiple historical voltage data corresponding to the multiple sampling moments, wherein the preprocessing includes at least one of the following: erroneous data screening, missing data value completion, and the multiple sampling moments correspond one-to-one to the multiple historical voltage data.

[0056] Optionally, the historical voltage data corresponding to the fuel cell is processed by means of data collection, protocol analysis, XOR verification, and invalid value screening, and the historical voltage data is stored in a time series database. Specifically, the basic state of the fuel cell power system and the historical voltage data (i.e., the single-chip battery voltage statistics) that can be collected during the historical period are statistically analyzed to visualize the key parameters of the fuel cell stack and auxiliary system corresponding to the fuel cell. In view of the fact that there are continuous load changes, unstable load points in steady-state operation, abnormal communication interference and other factors in the actual historical voltage data, which lead to data missing, the collected historical voltage data is cleaned based on data rationality and data continuity.

[0057] Optionally, the upper and lower limits of the fuel cell voltage value are determined based on the electrochemical characteristics of the fuel cell. For a hydrogen-oxygen fuel cell at room temperature and pressure, the Gibbs free energy is determined to be ΔG=-237.3 kJ / mol, the Faraday constant is 96400, and the upper limit of the fuel cell voltage value is determined based on the open circuit voltage theory as follows:

[0058]

[0059] Among them, E is the upper limit range of the fuel cell voltage value, the lower limit range of the fuel cell is determined based on the upper limit range, the fuel cell voltage value should be in the range of -1.23V to 1.23V, ΔG is the Gibbs free energy, specifically, Gibbs free energy is the free energy of a system under constant temperature and pressure conditions, it represents the energy available for work in the system at constant temperature and pressure. Gibbs free energy can be used to determine whether a chemical reaction will occur in the system, and the direction of the reaction. The smaller the Gibbs free energy of the system, the more stable the system, and the more likely the reaction will occur. N is the number of N monolithic stacks in the fuel cell, and F is the Faraday constant, which represents the energy required to transfer 1 mole of electrons in an electrochemical reaction.

[0060] Optionally, based on the range of the above-mentioned fuel cell voltage value, abnormal data with voltage values ​​exceeding the range are screened out to obtain historical voltage data at multiple sampling moments, wherein multiple sampling moments are discontinuous due to the removal of multiple voltage data exceeding the range, so it is necessary to fill in the missing values ​​at the discontinuous positions based on the above-mentioned data that does not exceed the range value.

[0061] Optionally, based on the historical voltage data and historical operation data obtained above, a time series graph of the stack voltage, hydrothermal management, cathode and cathode gas temperature, cathode and cathode gas pressure, and cathode and cathode gas flow rate can be created to display the data change trend, abnormal values, and data loss collected by the fuel cell system. The relationship between the fuel cell voltage and the cathode and cathode gas temperature, cathode and cathode gas pressure, and cathode and cathode gas flow rate in the fuel cell chemical reactor can be displayed using a scatter plot, so that the changes in the operating state of the fuel cell can be observed intuitively. At the same time, based on the scatter plot, the missing values ​​of the data during the actual operation of the fuel cell can be identified, and the differences in the steady-state points of different fuel cells, the diverse load frequency, and the presence of abnormal communication interference can be clarified.

[0062] Optionally, based on the above scatter plot, a linear interpolation method can be used to fill in the missing values ​​of the historical voltage data to ensure the temporal continuity of the historical battery data when the fuel cell is running. Specifically, a standardized score (Z-Score) is used for interpolation substitution processing.

[0063] In an optional embodiment, the battery performance prediction model includes an input layer, a hidden layer and an output layer, wherein the input layer is used to encode and convert real-time operation data to obtain first data; the hidden layer is used to perform nonlinear conversion on the first data to obtain nonlinear characteristics of the first data, and feature matching processing is performed on the nonlinear characteristics to obtain second data; and the output layer is used to decode the second data to obtain target voltage data.

[0064] Optionally, the battery performance prediction model can be constructed by stacking a bidirectional long short-term memory network (BidirectionalLong Short-Term Memory, BiLSTM) model with a fully connected neural network, wherein the number of hidden layers can be one or more, for example, the battery performance prediction model can include an input layer, 6 hidden layers, and an output layer. Specifically, the real-time operation data is used as input, input into the input layer, and the encoding conversion is performed at the input layer to obtain the first data recognizable by the hidden layer, that is, converted into data recognizable by the neural network. The hidden layer receives the first data and performs a nonlinear conversion on the first data based on the activation function to obtain the nonlinear feature required for identification, which can be used to predict the matching relationship between multiple input parameters and voltage, thereby accurately obtaining the second data related to the voltage. The output layer receives the second data and decodes it into target voltage data that can be viewed by the staff.

[0065] Optionally, in order to achieve a certain amount of prediction advance and increase the accuracy of the prediction, the above real-time operating data can be standardized and subjected to sliding window processing between inputs, with the input three seconds ahead of the output, so that the output of the battery performance prediction model is the target voltage data of the fuel cell. Among them, sliding window processing is a method for processing input data, which can process continuous sequence data, that is, real-time operating data corresponding to multiple sampling moments. Sliding window processing divides the input sequence (real-time operating data) into windows of fixed length (or called sliding windows), that is, slicing processing, and then uses each window as the input of the battery performance prediction model. This method can retain the sequential information of the input data and can better capture the dependencies between sequence data when training the battery performance prediction model.

[0066] Optionally, during the training of the battery performance prediction model, the regularization parameter (dropout), reverse parameter (recurrent dropout) and learning rate and other hyperparameters can be adjusted according to the fitting performance of the historical operating data, and the parameter curve can be observed. When the error drops to a certain preset value, a model that can accurately predict the performance of the fuel cell is obtained. It should be noted that Dropout is a parameter value in the regularization technology. Specifically, during the training of the battery performance prediction model, the output of a part of the neurons is randomly set to 0, thereby reducing the dependency between neurons and preventing overfitting. By adjusting the dropout ratio, the complexity of the neural network can be controlled and the generalization ability of the model can be improved. The dropout value is between 0.2 and 0.5, and the selection of the specific value needs to be adjusted according to the actual situation.

[0067] In an optional embodiment, the hidden layer includes a forward long short-term memory neural network, a reverse long short-term memory neural network and a fully connected layer, wherein the forward long short-term memory neural network is used to obtain long-term dependency features in the first data in a first direction to obtain a first feature, wherein the first direction is the direction from the first data of the first data to the last data; the reverse long short-term memory neural network is used to obtain short-term dependency features in the first data in a second direction to obtain a second feature, wherein the second direction is the opposite direction of the first direction; based on preset weights, the fully connected layer is used to perform nonlinear mapping on the first feature and the second feature to obtain a nonlinear feature, and feature matching processing is performed on the nonlinear feature to obtain the second data.

[0068] Optionally, the hidden layer includes a forward long short-term memory neural network, a reverse long short-term memory neural network and a fully connected layer. Among them, the forward long short-term memory neural network obtains the long-term dependency features in the first data in the direction from the first data of the first data to the last data. Since the historical operation data is a sequence with a time order, the time order is formed according to the order of collection at multiple sampling moments. The above time series has a forward and a reverse direction. The forward direction is the direction from the first data to the last data (the first direction). When the above historical operation data is input into the battery performance prediction model according to the time series, the order of the above sequence is still maintained in the battery performance prediction model. Obtaining long-term dependency features in time order can well capture the time correlation in the historical operation data and better obtain long-term dependency features. Specifically, the forward long short-term memory neural network includes 512 neural network nodes. Based on the above neural network nodes, feature extraction of historical operation data can effectively achieve the effect of comprehensive acquisition of long-term dependency features. At the same time, the forward long short-term memory neural network uses the hyperbolic tangent activation function (Tanh) to map the input value to [-1,1], which helps to reduce the gradient disappearance problem and can help the neural network better learn complex nonlinear relationships.

[0069] Optionally, the inverse long short-term memory neural network obtains the short-term dependency features in the first data in the direction from the last data to the first data, and obtains the short-term dependency features based on the reverse direction (second direction), so as to obtain the feature relationship of the reverse time series. Combined with the above-mentioned positive long-term dependency features, more comprehensive nonlinear features can be obtained, and comprehensive feature extraction of historical operation data can be effectively achieved. Specifically, the inverse long short-term memory neural network includes 256 neural network nodes, and feature extraction of historical operation data based on the above-mentioned neural network nodes can effectively achieve the effect of comprehensive acquisition of short-term dependency features. At the same time, the inverse long short-term memory neural network also uses the Tanh function, which can map the input value to between [-1,1], which helps to reduce the gradient vanishing problem and can help the neural network better learn complex nonlinear relationships.

[0070] In an optional embodiment, the fully connected layer includes a first fully connected layer, a second fully connected layer, a third fully connected layer and a fourth fully connected layer, wherein, based on preset weights, the first fully connected layer is used to perform nonlinear mapping on the first feature and the second feature to obtain a nonlinear feature; the second fully connected layer is used to perform feature matching on the nonlinear feature to determine a predicted value of the second data; the third fully connected layer is used to determine a loss value between the predicted value and the historical voltage data; based on the loss value, the fourth fully connected layer is used to adjust the predicted value to obtain the second data.

[0071] Optional, such as Figure 3 As shown, Figure 3 2 is a schematic diagram of the structure of a battery performance prediction model of a fuel cell performance prediction method according to an embodiment of the present invention. The fully connected layer is arranged in the last few layers of the bidirectional long short-term memory network, and is used to combine and integrate the first feature and the second feature extracted from the forward long short-term memory neural network and the reverse long short-term memory neural network to obtain the final output. Specifically, it includes 4 fully connected layers, namely the first fully connected layer, the second fully connected layer, the third fully connected layer and the fourth fully connected layer. In addition, the hidden layer is formed together with the forward long short-term memory neural network and the reverse long short-term memory neural network, that is, the hidden layer includes 6 layers, namely the forward long short-term memory neural network layer, the reverse long short-term memory neural network layer, the first fully connected layer, the second fully connected layer, the third fully connected layer and the fourth fully connected layer.

[0072] Optionally, the first fully connected layer is provided with 200 neural network nodes. The above 200 neural network nodes are used to perform nonlinear mapping on the first feature and the second feature based on preset weights to obtain nonlinear features. Specifically, the first fully connected layer uses a rectified nonlinear activation function (Rectified Linear Unit, ReLU) to allow the first fully connected layer to learn nonlinear patterns and features, thereby improving the expression and fitting capabilities of the battery performance prediction model. At the same time, it can avoid the disappearance or explosion of the gradient during the back propagation process, thereby improving the training effect of the first fully connected layer. It can also output 0 when the input is negative, so that some neurons in the first fully connected layer are in a static state, thereby improving the sparsity and generalization ability of the battery performance prediction model.

[0073] Optionally, the second fully connected layer is provided with 100 neural network nodes; the above 100 neural network nodes are used to perform feature matching on the nonlinear features to determine the predicted value of the second data. Specifically, the second fully connected layer also uses the ReLU activation function to improve the expression and fitting capabilities of the battery performance prediction model, while avoiding the disappearance or explosion of the gradient during the back propagation process.

[0074] Optionally, the third fully connected layer is provided with 50 neural network nodes; using the above 50 neural network nodes, the loss value between the predicted value and the historical voltage data is specifically determined by a loss function, and the loss function expression is as follows:

[0075]

[0076] Where L is the loss value, M is the number of neurons in the third fully connected layer; y ic is a sign function (0 or 1), if the true category c of the observed sample i (historical voltage data) is 1, otherwise it is 0; p ic is the predicted probability that the observed sample i belongs to category c.

[0077] Optionally, the fourth fully connected layer is provided with 10 neural network nodes; using the above 10 neural network nodes, the prediction value is adjusted based on the loss value to obtain the second data.

[0078] Through the above steps S102 to S104, the purpose of screening out multiple operating parameters with a greater correlation with historical voltage data from multiple input parameters and constructing a battery performance prediction model is achieved, thereby achieving the technical effect of improving the accuracy of the prediction results of the battery performance prediction model, and further solving the technical problem of low accuracy of battery performance prediction results caused by the fact that the operating parameters used in constructing the model in related technologies are many and complex.

[0079] Based on the above embodiments and optional embodiments, the present invention proposes an optional implementation mode: Figure 4 is a flow chart of an optional fuel cell performance prediction method according to an embodiment of the present invention, such as Figure 4 As shown, the method includes:

[0080] Step S1, data collection and cleaning: real-time collection of fuel cell data (historical voltage data), and fuel cell data cleaning according to real-time operation data reporting rules and actual transmission conditions.

[0081] Specifically, the operation data of the fuel cell is called from the database interface of the data platform, and then data cleaning is performed based on the data collection situation, including at least erroneous data screening and missing data value completion.

[0082] Step S2, feature value screening: The fuel cell operation data feature screening method can screen the correlation based on expert experience and predict the feature parameter set related to the target.

[0083] Specifically, based on operating data, expert knowledge and correlation analysis methods, the operating condition parameters, load current parameters and environmental parameters corresponding to the subsystems in the fuel cell are feature screened to narrow the parameter range and obtain an effective set of key feature parameters (historical operating parameters).

[0084] Step S3, prediction model:

[0085] Step S31, constructing a performance prediction model of fuel cell operation data (battery performance prediction model), forming a neural network model suitable for the actual operation data of the fuel cell, and performing training and deployment.

[0086] Specifically, the BILSTM network structure is used to predict the average voltage and voltage standard deviation of a single cell in a fuel cell. The specific parameter settings of the bidirectional long short-term memory network include the number of nodes per layer, activation function, and hyperparameter settings to adapt to the complex situations of different models of fuel cells, different data collection situations, and different stage strategies.

[0087] Step S32, in actual application of the battery performance prediction model, the structure of the battery performance prediction model is compared with that of different related performance prediction models, and the accuracy and speed of the constructed battery performance prediction model are compared with other neural network structure models.

[0088] Specifically, relevant deep learning models are collected, and the actual prediction results of five commonly used types of neural network models in performance prediction applications are compared to verify that the structure of the battery performance prediction model has the effect of more accurate prediction results in the actual application of performance prediction.

[0089] Through the above steps S1 to S3, the complex load conditions during the use of the fuel cell can be solved, and the problems of high frequency, large volume, and low information entropy in the actual battery data can be solved, so as to achieve early and accurate prediction of the transient voltage performance of the fuel cell, provide a reference for the calibration of steady-state parameter standard operating conditions and the optimization of load reduction and load reduction strategies, facilitate the analysis of fuel cell historical data, and visualize the performance of the fuel cell stack.

[0090] In this embodiment, a fuel cell performance prediction device is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and the descriptions that have been made will not be repeated. As used below, the terms "module" and "device" can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0091] According to an embodiment of the present invention, there is also provided an embodiment of a device for implementing the above-mentioned fuel cell performance prediction method. Figure 5 is a schematic diagram of the structure of a fuel cell performance prediction device according to an embodiment of the present invention. Figure 5 As shown, the above fuel cell performance prediction device includes: an acquisition module 201 and a prediction module 202, wherein:

[0092] An acquisition module 201 is used to acquire real-time operation data corresponding to the fuel cell in the current period;

[0093] The prediction module 202 is connected to the acquisition module 201, and is used to obtain the target voltage data corresponding to the fuel cell based on the real-time operation data and the battery performance prediction model, wherein the battery performance prediction model is based on the historical operation data in the historical period and the historical voltage data corresponding to the historical operation data, and is obtained through machine learning. The historical operation data includes multiple operation parameters, and the multiple operation parameters are based on the historical voltage data and are screened from multiple input parameters.

[0094] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0095] It should be noted that the acquisition module 201 and the prediction module 202 correspond to steps S102 to S104 in the embodiment, and the examples and application scenarios implemented by the modules and the corresponding steps are the same, but are not limited to the contents disclosed in the embodiment. It should be noted that the modules as part of the device can be run in a computer terminal.

[0096] It should be noted that the optional or preferred implementation of this embodiment can refer to the relevant description in the embodiment, which will not be repeated here.

[0097] The above-mentioned fuel cell performance prediction device may also include a processor and a memory. The above-mentioned acquisition module 201 and prediction module 202 are stored in the memory as program modules, and the processor executes the above-mentioned program modules stored in the memory to realize corresponding functions.

[0098] The processor includes a kernel, which retrieves the corresponding program module from the memory. The kernel may be one or more. The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. The memory includes at least one memory chip.

[0099] According to an embodiment of the present application, an embodiment of a non-volatile storage medium is also provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein when the program is executed, the device where the non-volatile storage medium is located is controlled to execute any of the above-mentioned fuel cell performance prediction methods.

[0100] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group, and the non-volatile storage medium includes a stored program.

[0101] Optionally, when the program is running, the device where the non-volatile storage medium is located is controlled to perform the following functions: obtaining real-time operating data corresponding to the fuel cell in the current time period; based on the real-time operating data, using a battery performance prediction model to obtain target voltage data corresponding to the fuel cell, wherein the battery performance prediction model is based on historical operating data in a historical time period and historical voltage data corresponding to the historical operating data, and is obtained through machine learning, and the historical operating data includes multiple operating parameters, and the multiple operating parameters are based on the historical voltage data and are screened from multiple input parameters.

[0102] According to an embodiment of the present application, an embodiment of a processor is also provided. Optionally, in this embodiment, the processor is used to run a program, wherein when the program is run, any one of the fuel cell performance prediction methods is executed.

[0103] According to an embodiment of the present application, an embodiment of a computer program product is also provided. Optionally, in this embodiment, the computer program product includes a computer program, and when the computer program is executed by a processor, the computer program implements any of the steps of the above-mentioned fuel cell performance prediction method.

[0104] Optionally, the above-mentioned computer program product, when executed on a data processing device, is suitable for executing an initialization program having the following method steps: obtaining real-time operating data corresponding to the fuel cell in the current time period; based on the real-time operating data, using a battery performance prediction model to obtain target voltage data corresponding to the fuel cell, wherein the battery performance prediction model is based on historical operating data in a historical time period and historical voltage data corresponding to the historical operating data, and is obtained through machine learning, and the historical operating data includes a variety of operating parameters, and the various operating parameters are based on the historical voltage data and are screened from a variety of input parameters.

[0105] An embodiment of the present invention provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the following steps are implemented: obtaining real-time operating data corresponding to a fuel cell in a current time period; based on the real-time operating data, using a battery performance prediction model to obtain target voltage data corresponding to the fuel cell, wherein the battery performance prediction model is based on historical operating data in a historical time period and historical voltage data corresponding to the historical operating data, and is obtained through machine learning, and the historical operating data includes a variety of operating parameters, and the various operating parameters are screened from a variety of input parameters based on the historical voltage data.

[0106] The above sequence of the embodiments of the present invention is for description only and does not represent the superiority or inferiority of the embodiments.

[0107] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the above modules can be a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, modules or indirect coupling or communication connection of modules, which can be electrical or other forms.

[0109] The modules described above as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0110] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of software functional modules.

[0111] If the above-mentioned integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a non-volatile storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the various embodiments of the present invention. The aforementioned non-volatile storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0112] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A fuel cell performance prediction method, characterized in that: include: Obtain the real-time operation data corresponding to the fuel cell in the current period; Based on the real-time operating data, a battery performance prediction model is used to obtain target voltage data corresponding to the fuel cell, wherein the battery performance prediction model is obtained through machine learning based on historical operating data in a historical period and historical voltage data corresponding to the historical operating data, and the historical operating data includes multiple operating parameters, and the multiple operating parameters are screened from multiple input parameters based on the historical voltage data.

2. The method according to claim 1, characterized in that Before obtaining the target voltage data corresponding to the fuel cell by using the battery performance prediction model based on the real-time operation data, the method further includes: Acquiring the historical voltage data of the fuel cell in the historical period; Obtaining parameter values ​​corresponding to a plurality of input parameters affecting the operation of the fuel cell during the historical period; Determining correlation coefficients between the various input parameters and the voltage of the fuel cell based on the historical voltage data and the parameter values ​​corresponding to the various input parameters; Filtering the plurality of operating parameters from the plurality of input parameters based on correlation coefficients between the plurality of input parameters and the voltage of the fuel cell respectively; The historical operation data is obtained based on the parameter values ​​respectively corresponding to the multiple operation parameters.

3. The method according to claim 2, characterized in that In the case where the historical period includes a plurality of sampling moments, determining the correlation coefficients between the plurality of input parameters and the voltage of the fuel cell based on the historical voltage data and the parameter values ​​corresponding to the plurality of input parameters respectively includes: The correlation coefficient between any input parameter among the multiple input parameters and the voltage of the fuel cell is obtained in the following manner: Acquire multiple parameter values ​​corresponding to any one of the input parameters collected at the multiple sampling moments, and multiple historical voltage data, wherein the multiple parameter values ​​correspond to the multiple sampling moments one-to-one, and the multiple historical voltage data correspond to the multiple sampling moments one-to-one; Determining a voltage average value based on the plurality of historical voltage data; Determining the covariance between each of the plurality of parameter values ​​and the voltage average value; Determine the standard deviations between the plurality of parameter values ​​and the voltage average value; Determining a correlation coefficient between any one of the input parameters and the voltage of the fuel cell based on the covariance and the standard deviation; The correlation coefficients between the plurality of input parameters and the voltage of the fuel cell are obtained by obtaining the correlation coefficient between any one of the input parameters and the voltage of the fuel cell.

4. The method according to claim 2, characterized in that: The acquiring the historical voltage data of the fuel cell in the historical period includes: Acquire a plurality of initial voltage data corresponding to a plurality of sampling moments of the fuel cell in the historical period, wherein the plurality of sampling moments correspond to the plurality of initial voltage data one by one; The multiple initial voltage data are preprocessed to obtain multiple historical voltage data corresponding to the multiple sampling moments, wherein the preprocessing includes at least one of the following: error data screening, missing data value completion, and the multiple sampling moments correspond one-to-one to the multiple historical voltage data.

5. The method according to claim 1, characterized in that The battery performance prediction model includes an input layer, a hidden layer and an output layer, wherein: The input layer is used to perform encoding conversion on the real-time operation data to obtain first data; Using the hidden layer to perform nonlinear transformation on the first data to obtain nonlinear features of the first data, and performing feature matching processing on the nonlinear features to obtain second data; The output layer is used to decode the second data to obtain the target voltage data.

6. The method according to claim 5, characterized in that The hidden layer includes a forward long short-term memory neural network, a reverse long short-term memory neural network and a fully connected layer, wherein: Using the forward long short-term memory neural network to obtain long-term dependency features in the first data in a first direction to obtain a first feature, wherein the first direction is a direction from the first data to the last data of the first data; Using the inverse long short-term memory neural network to obtain short-term dependency features in the first data in a second direction to obtain a second feature, wherein the second direction is the opposite direction of the first direction; Based on the preset weights, the fully connected layer is used to perform nonlinear mapping on the first feature and the second feature to obtain the nonlinear feature, and feature matching processing is performed on the nonlinear feature to obtain second data.

7. The method according to claim 6, characterized in that The fully connected layer includes a first fully connected layer, a second fully connected layer, a third fully connected layer and a fourth fully connected layer, wherein: Based on the preset weights, using the first fully connected layer to perform nonlinear mapping on the first feature and the second feature to obtain the nonlinear feature; Using the second fully connected layer to perform feature matching on the nonlinear feature to determine a predicted value of the second data; Determine a loss value between the predicted value and the historical voltage data using the third fully connected layer; Based on the loss value, the fourth fully connected layer is used to adjust the predicted value to obtain the second data.

8. A fuel cell performance prediction device, characterized in that: include: An acquisition module is used to acquire the real-time operation data corresponding to the fuel cell in the current period; A prediction module is used to obtain target voltage data corresponding to the fuel cell based on the real-time operation data and using a battery performance prediction model, wherein the battery performance prediction model is obtained through machine learning based on historical operation data in a historical period and historical voltage data corresponding to the historical operation data, and the historical operation data includes multiple operation parameters, and the multiple operation parameters are screened from multiple input parameters based on the historical voltage data.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the fuel cell performance prediction method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the fuel cell performance prediction method according to any one of claims 1 to 7 are implemented.

Images