A fuel cell health status estimation method and system based on compensated GPR model

Abstract

The present invention belongs to the technical field of fuel cell health state estimation, and specifically relates to a fuel cell health state estimation method and system based on a compensated GPR model, including obtaining characteristic parameters of the fuel cell under discharge conditions, using a self-sparse autoencoder to extract target features from the characteristic parameters; using the target features as the input of a pre-constructed Gaussian process regression GPR model, using the battery SOH as the output of the GPR model, using the Kepler planetary optimization algorithm to perform hyperparameter optimization on the GPR model, and the optimized GPR model outputs a first predicted value of SOH; setting a compensation function to compensate and optimize the first predicted value of SOH to obtain a second predicted value of SOH; selecting the mean absolute percentage error, mean square error, and maximum error as evaluation indicators to evaluate the second predicted value of SOH. The present invention can realize the SOH estimation of fuel cells under different discharge currents with high accuracy, and the method can adapt to the SOH estimation of fuel cells under different discharge currents, and has good robustness.

Description

Technical Field

[0001] The present invention belongs to the technical field of fuel cell health state estimation, and in particular relates to a fuel cell health state estimation method and system based on a compensated GPR model. Background Art

[0002] The durability problem of fuel cells limits their large-scale application. Accurately estimating the health state SOH of fuel cells is of great significance to improving their safety and service life. Since the aging of fuel cells is a nonlinear process coupled by multiple factors, accurately estimating the SOH of fuel cells is a major challenge. Generally speaking, when the power of the battery stack degrades to 90% of the rated power, the fuel cell reaches the end of its life.

[0003] The SOH estimation methods of fuel cells can be divided into model-based methods, experimental methods and data-driven methods. Model-based methods require a deep understanding of the working mechanism of fuel cells, and in practical applications require a large amount of computing resources to perform model simulation and estimation; experimental methods have high requirements for experimental data and take a long time to collect and analyze data; data-driven methods use machine learning and data mining techniques to discover complex relationships between data without the need for complex system modeling, but have high requirements for data quality and poor interpretability.

[0004] In practical applications, as the discharge current of the fuel cell changes, the SOH estimation method based on the data at a certain discharge current will have a significantly reduced accuracy in estimating the SOH of the fuel cell at other discharge currents. Due to limited computing resources and label data, it is unrealistic to establish different data-driven models based on data at different discharge currents. Summary of the invention

[0005] The purpose of the present invention is to provide a fuel cell health status estimation method and system based on a compensated GPR model to solve the problems raised in the background technology.

[0006] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0007] In a first aspect, the present invention proposes a method for estimating the health status of a fuel cell based on a compensated GPR model, the method comprising:

[0008] Acquire characteristic parameters of the fuel cell under discharge conditions, and use a self-sparse autoencoder to extract target features related to the battery SOH from the characteristic parameters;

[0009] The target feature is used as the input of a pre-built Gaussian process regression GPR model, the battery SOH is used as the output of the GPR model, and the Kepler planetary optimization algorithm is used to perform hyperparameter optimization on the GPR model to obtain a trained GPR model;

[0010] The target feature of the fuel cell to be predicted is input into the trained GPR model, a first SOH prediction value is output, a compensation function is set to compensate and optimize the first SOH prediction value, and a second SOH prediction value is obtained.

[0011] As a further preferred solution, the pre-built Gaussian process regression GPR model includes:

[0012] Regression is performed by fitting the target feature to a Gaussian process uniquely determined by a mean function and a covariance function, setting the mean function μ(x) and the kernel function of the GPR model, and setting the mean function μ(x) of the GPR model to 0; the kernel function of the GPR model is composed of radial basis functions, as shown in the following formula:

[0013] y(x)~N(μ(x),k(x,x'))

[0014] Among them, μ(x) represents the mean function; k(x,x') represents the kernel function;

[0015]

[0016] Among them, δ and l are hyperparameters;

[0017]

[0018] Among them, y * |y means that given the SOH observation value y, the unknown variable y * The probability distribution of x and x * Respectively represent y and y * Corresponding target features; K = k(x,x * ), K * = k(x,x * ), K ** = k(x * ,x * ).

[0019] As a further preferred solution, the setting of a compensation function to compensate and optimize the first predicted value of SOH to obtain a second predicted value of SOH includes:

[0020] The compensation function is set to compensate and optimize the first predicted value of SOH of the optimized GPR model, as shown in the following formula:

[0021] SOH ult=GPR(F)+f(I,GPR(F))

[0022] f(I,GPR(F))=a(I)*GPR(F)+b(I)

[0023] Among them, SOH ult represents the second predicted value of SOH; GPR(F) represents the first predicted value of SOH, and F represents the target feature set; f(I, GPR(F)) is a function of the discharge current I and GPR(F), which is used to compensate the output of the GPR model. a(I) and b(I) vary with the discharge current.

[0024] As a further preferred solution, the mean absolute percentage error MAPE, the mean square error MSE and the maximum error Max-AE are selected as evaluation indicators to evaluate the second predicted value of SOH, as shown in the following formula:

[0025]

[0026]

[0027]

[0028] in, represents the second predicted value of SOH, y i represents the observed value of SOH, and n is the number of samples.

[0029] In a second aspect, the present invention proposes a fuel cell health status estimation system based on a compensated GPR model, which is applied to perform the above estimation method, and the system comprises:

[0030] A feature acquisition module, used to acquire feature parameters of the fuel cell under discharge conditions, and to extract target features related to the battery SOH from the feature parameters using a self-sparse autoencoder;

[0031] A model building module is used to use the target feature as the input of a pre-built Gaussian process regression (GPR) model, and the battery SOH as the output of the GPR model, and use the Kepler planetary optimization algorithm to perform hyperparameter optimization on the GPR model to obtain a trained GPR model;

[0032] The model compensation module is used to input the target characteristics of the fuel cell to be predicted into the trained GPR model, output the first SOH prediction value, set the compensation function to compensate and optimize the first SOH prediction value, and obtain the second SOH prediction value.

[0033] The beneficial effect of the present invention is that compared with the prior art, the present invention uses a sparse autoencoder to extract features from the monitored current, voltage, temperature, flow rate, humidity and other parameters, and inputs them into the GPR model to obtain the SOH estimation result, and uses a compensation function to compensate the SOH estimation result, thereby improving the SOH estimation accuracy. The present invention can realize the SOH estimation of fuel cells under different discharge currents with high accuracy, and the method can adapt to the SOH estimation of fuel cells under different discharge currents, and has good robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is an execution flow chart of the fuel cell health status estimation method based on the compensated GPR model in the present invention.

[0035] Figure 2 It is a framework diagram of the fuel cell health status estimation method based on the compensated GPR model in the present invention.

[0036] Figure 3 This is a flow chart of the Kepler planet optimization algorithm in the present invention. DETAILED DESCRIPTION

[0037] The present application is further described in detail below in conjunction with the accompanying drawings. It is necessary to point out here that the following specific implementation methods are only used to further illustrate the present application and cannot be understood as limiting the scope of protection of the present application. Technical personnel in this field can make some non-essential improvements and adjustments to the present application based on the above application content.

[0038] Example 1

[0039] like Figure 1-3 As shown, this embodiment proposes a method for estimating the health status of a fuel cell based on a compensated GPR model, the method comprising: obtaining characteristic parameters of the fuel cell under discharge conditions, and using a self-sparse autoencoder to extract target features related to the battery SOH in the characteristic parameters; using the target features as the input of a pre-constructed Gaussian process regression GPR model, and the battery SOH as the output of the GPR model, and using the Kepler planetary optimization algorithm to perform hyperparameter optimization on the GPR model to obtain a trained GPR model; inputting the target features of the fuel cell to be predicted into the trained GPR model, outputting a first predicted value of SOH, and setting a compensation function to perform compensation optimization on the first predicted value of SOH to obtain a second predicted value of SOH.

[0040] More specifically, combined Figure 2In this embodiment, a sparse autoencoder (SAE) is used to extract features from the monitored current, voltage, temperature, flow rate, humidity and other parameters, and input them into the GPR model to obtain the SOH estimation result, and the SOH estimation result is compensated by a compensation function, thereby improving the SOH estimation accuracy. The present invention can realize the SOH estimation of the fuel cell under different discharge currents with high accuracy and almost does not increase the calculation load.

[0041] Furthermore, the characteristic parameters of the fuel cell under discharge conditions include: cell voltage, stack voltage, current density, current, air inlet pressure, air outlet pressure, air inlet flow rate, air outlet flow rate, fuel inlet temperature, fuel outlet temperature, fuel inlet pressure, fuel outlet pressure, fuel inlet flow rate, fuel outlet flow rate and fuel humidity.

[0042] The health status estimation method in this embodiment is described below from the aspects of feature acquisition, GPR model, model training, compensation GPR model and model evaluation.

[0043] 1. Feature acquisition

[0044] The current, voltage, temperature, pressure, flow rate, humidity and other parameters of the fuel cell during discharge are collected (see Table 1 for details), and the current and voltage are smoothed using the Savitzky-Golay filter. The power of the fuel cell is calculated according to formula (1) and used as an indicator to measure its aging degree.

[0045] P=UI#(1)

[0046] Wherein, P represents power; U and I represent stack voltage and stack current respectively.

[0047] Table 1 Parameters used for feature extraction

[0048]

[0049] SAE is used to automatically extract features from the parameter variables in the table and input into the GPR model to estimate the SOH of the fuel cell.

[0050] 2. GPR Model

[0051] The pre-built Gaussian process regression GPR model includes: performing regression by fitting the target feature to a Gaussian process uniquely determined by a mean function and a covariance function, setting the mean function μ(x) and kernel function of the GPR model, and the available data points are samples from the Gaussian process, that is, they follow a Gaussian distribution, as shown in formula (2). Given that the properties of the Gaussian process are mainly determined by the kernel function (i.e., the covariance function), the mean function μ(x) of the GPR model is set to 0; the kernel function of the GPR model is composed of radial basis functions, as shown in the following formula:

[0052] y(x)~N(μ(x),k(x,x'))#(2)

[0053] Among them, μ(x) represents the mean function; k(x,x') represents the kernel function;

[0054]

[0055] Among them, δ and l are hyperparameters;

[0056]

[0057] Among them, y * |y means that given the SOH observation value y, the unknown variable y * The probability distribution of x and x * Respectively represent y and y * Corresponding target features; K = k(x,x * ), K * = k(x,x * ), K ** = k(x * ,x * ).

[0058] 3. Model training

[0059] Combination Figure 3 ,Choosing appropriate hyperparameters for the GPR model is crucial to accurately estimate the SOH of the fuel cell.,Therefore, the Kepler Planet Optimization Algorithm (KOA) is used to tune the hyperparameters of the GPR model to obtain the best performance and ensure accurate SOH prediction.,The KOA algorithm uses Kepler's laws of planetary motion to simulate the laws of planetary motion.

[0060] Furthermore, the Kepler planetary optimization algorithm is used to optimize the hyperparameters of the GPR model. The Kepler planetary optimization algorithm is a new physics-based metaheuristic algorithm that simulates planetary motion by using Kepler's laws of planetary motion. In the Kepler planetary optimization algorithm, each planet position is used as a candidate solution, and the sun (optimal solution) and the planets (candidate solutions) that revolve around it in an elliptical orbit represent the search space. The algorithm can avoid falling into the local optimal value by changing the motion position of the planet, which can effectively improve the performance of the GPR model.

[0061] 4. Compensation GPR Model

[0062] In order to improve the SOH estimation accuracy and improve the versatility of the model, the present invention applies a compensation function to compensate and optimize the estimation result of the GPR model, and sets the compensation function to compensate and optimize the first predicted value of SOH to obtain the second predicted value of SOH, including:

[0063] The compensation function is set to compensate and optimize the first predicted value of SOH of the optimized GPR model, as shown in the following formula:

[0064] SOH ult =GPR(F)+f(I,GPR(F))#(5)

[0065] f(I,GPR(F))=a(I)*GPR(F)+b(I)#(6)

[0066] Among them, SOH ult represents the second predicted value of SOH; GPR(F) represents the first predicted value of SOH, and F represents the target feature set; f(I, GPR(F)) is a function of the discharge current I and GPR(F), which is used to compensate the output of the GPR model. a(I) and b(I) vary with the discharge current.

[0067] 5. Model Evaluation

[0068] Furthermore, the present invention selects the mean absolute percentage error MAPE, the mean square error MSE and the maximum error Max-AE as evaluation indicators to evaluate the second prediction value of SOH, as shown in the following formula:

[0069]

[0070]

[0071]

[0072] in, represents the second predicted value of SOH, y i represents the observed value of SOH, and n is the number of samples.

[0073] Based on the same inventive concept, this embodiment also proposes a fuel cell health status estimation system based on a compensated GPR model, which is applied to execute the above estimation method. The system includes:

[0074] A feature acquisition module, used to acquire feature parameters of the fuel cell under discharge conditions, and to extract target features related to the battery SOH from the feature parameters using a self-sparse autoencoder;

[0075] A model building module is used to use the target feature as the input of a pre-built Gaussian process regression (GPR) model, and the battery SOH as the output of the GPR model, and use the Kepler planetary optimization algorithm to perform hyperparameter optimization on the GPR model to obtain a trained GPR model;

[0076] The model compensation module is used to input the target characteristics of the fuel cell to be predicted into the trained GPR model, output the first SOH prediction value, set the compensation function to compensate and optimize the first SOH prediction value, and obtain the second SOH prediction value.

[0077] According to the embodiments of the present invention, after experimental verification, the estimation error of the fuel cell SOH estimation method based on the compensated GPR algorithm proposed by the present invention is within 3%, the error in most cases is within 2%, and the maximum error does not exceed 4%. Compared with directly using the GPR algorithm to estimate the SOH of the fuel cell, the method proposed by the present invention has improved the average absolute percentage error, root mean square error and maximum absolute error by 91.68%, 90.20% and 87.01% respectively, and the time required for operation has only increased by 0.0017 seconds. In addition, the method can adapt to the estimation of fuel cell SOH under different discharge currents and has good robustness.

[0078] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0079] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0080] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A fuel cell health status estimation method based on a compensated GPR model, characterized in that: The method comprises: Acquire characteristic parameters of the fuel cell under discharge conditions, and use a self-sparse autoencoder to extract target features related to the battery SOH from the characteristic parameters; The target feature is used as the input of a pre-built Gaussian process regression GPR model, the battery SOH is used as the output of the GPR model, and the Kepler planetary optimization algorithm is used to perform hyperparameter optimization on the GPR model to obtain a trained GPR model; Inputting the target feature of the fuel cell to be predicted into the trained GPR model, outputting a first predicted value of SOH, setting a compensation function to compensate and optimize the first predicted value of SOH, and obtaining a second predicted value of SOH; The pre-built Gaussian process regression GPR model includes: Regression is performed by fitting the target feature to a Gaussian process uniquely determined by a mean function and a covariance function, setting the mean function μ(x) and the kernel function of the GPR model, and setting the mean function μ(x) of the GPR model to 0; the kernel function of the GPR model is composed of radial basis functions, as shown in the following formula: y(x)~N(μ(x),k(x,x ′ )) Where μ(x) represents the mean function; k(x,x ′ ) represents the kernel function; Among them, δ and l are hyperparameters; Among them, y * |y means that given the SOH observation value y, the unknown variable y * The probability distribution of x and x * Respectively represent y and y * Corresponding target features; K = k(x,x * ), K * = k(x,x * ), K ** = k(x * ,x * ); The setting of the compensation function to compensate and optimize the first predicted value of SOH to obtain the second predicted value of SOH includes: The compensation function is set to compensate and optimize the first predicted value of SOH of the optimized GPR model, as shown in the following formula: SOH ult =GPR(F)+f(I,GPR(F)) f(I,GPR(F))=a(I)*GPR(F)+b(I) Among them, SOH ult represents the second predicted value of SOH; GPR(F) represents the first predicted value of SOH, and F represents the target feature set; f(I, GPR(F)) is a function of the discharge current I and GPR(F), which is used to compensate the output of the GPR model. a(I) and b(I) vary with the discharge current.

2. A fuel cell health status estimation method based on a compensated GPR model according to claim 1, characterized in that: The characteristic parameters of the fuel cell under discharge conditions include: cell voltage, stack voltage, current density, current, air inlet pressure, air outlet pressure, air inlet flow rate, air outlet flow rate, fuel inlet temperature, fuel outlet temperature, fuel inlet pressure, fuel outlet pressure, fuel inlet flow rate, fuel outlet flow rate and fuel humidity.

3. The method for estimating the health status of a fuel cell based on a compensated GPR model according to claim 1, characterized in that: The mean absolute percentage error MAPE, mean square error MSE and maximum error Max-AE are selected as evaluation indicators to evaluate the second predicted value of SOH, as shown in the following formula: in, represents the second predicted value of SOH, y i represents the observed value of SOH, and n is the number of samples.

4. A fuel cell health status estimation system based on a compensated GPR model, characterized in that: Applied to executing the estimation method according to any one of claims 1 to 3, the system comprises: A feature acquisition module, used to acquire feature parameters of the fuel cell under discharge conditions, and to extract target features related to the battery SOH from the feature parameters using a self-sparse autoencoder; A model building module is used to use the target feature as the input of a pre-built Gaussian process regression (GPR) model, and the battery SOH as the output of the GPR model, and use the Kepler planetary optimization algorithm to perform hyperparameter optimization on the GPR model to obtain a trained GPR model; The model compensation module is used to input the target characteristics of the fuel cell to be predicted into the trained GPR model, output the first SOH prediction value, set the compensation function to compensate and optimize the first SOH prediction value, and obtain the second SOH prediction value.

Images