PEM electrolytic cell life attenuation model and life prediction method

Abstract

The invention relates to the technical field of electrolytic cell life prediction, in particular to a PEM electrolytic cell life attenuation model which comprises a multi-physics field coupling simulation module used for executing electrochemical-thermal-fluid three-field coupling simulation based on finite element analysis and simulating the degradation behavior of a membrane electrode assembly by solving a set of control equations, the simulation module is configured to output a local stress concentration coefficient and a proton conductivity attenuation rate for quantifying performance attenuation; the self-adaptive life prediction module is in communication connection with the multi-physics field coupling simulation module and is used for receiving the key characteristic parameters output by the simulation module and combining the voltage fluctuation data, the hydrogen permeability and the film thickness change which are acquired in real time by utilizing a pre-trained random forest regression model; the attenuation mechanism of the electrolytic cell is accurately simulated through a multi-physics field coupling model, the material aging process is quantified in combination with an Arrhenius equation, high-precision life prediction is achieved through a random forest algorithm, the prediction accuracy is improved, and a basis is provided for optimized operation.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of electrolytic cell life prediction, in particular to a PEM electrolytic cell life attenuation model and a life prediction method. BACKGROUND

[0002] With the increasing demand for renewable energy worldwide, water electrolysis hydrogen production as one of the key technologies for clean energy production, the operation state monitoring and life prediction of the core component PEM electrolytic cell become an important link to ensure the efficient and stable operation of the system. According to the search of the publication No. CN119416668B, a PEM water electrolytic cell life prediction method is disclosed, which discloses the technical scheme of "using artificial intelligence technology based on deep learning to analyze the operation voltage data of the PEM water electrolytic cell, decomposing the voltage signal into a plurality of intrinsic mode components, and extracting the frequency domain features of each intrinsic mode component, and then, through kernel feature aggregation processing of the frequency domain features of each intrinsic mode component, the internal core regularity and time sequence change mode of the voltage signal are captured, so as to realize intelligent prediction of the life of the PEM water electrolytic cell", which has the technical effects of "can effectively improve the accuracy and efficiency of PEM water electrolytic cell life prediction, and reduce the high cost and high time consumption caused by traditional laboratory accelerated aging test method"; Although the above-mentioned method uses empirical mode decomposition and frequency domain feature extraction, the voltage signal shows complex non-stationary and multi-component characteristics in actual operation, and the energy distribution is uneven and the time sequence correlation is different between different intrinsic mode functions. It is difficult to accurately represent the essential time sequence mode of the signal by simply aggregating the features, resulting in feature representation ambiguity and loss of discriminative information when constructing the operation data time sequence kernel representation, thereby affecting the accuracy and robustness of the life prediction. SUMMARY

[0003] In view of the deficiencies of the prior art, the present application provides a PEM electrolytic cell life attenuation model and a life prediction method, which accurately simulates the attenuation mechanism of the electrolytic cell through a multi-physical field coupling model, quantifies the material aging process by using the Arrhenius equation, and realizes high-precision life prediction by using the random forest algorithm, thereby improving the prediction accuracy and providing a basis for optimizing the operation.

[0004] To achieve the above purpose, the present application realizes the following technical scheme: a PEM electrolytic cell life attenuation model, comprising: a multi-physical field coupling simulation module, configured to perform electrochemical-thermal-fluid three-field coupling simulation based on finite element analysis, the simulation simulating the degradation behavior of the membrane electrode assembly by solving a set of control equations, and the simulation module being configured to output a local stress concentration coefficient and a proton conductivity attenuation rate for quantifying performance attenuation; An adaptive lifetime prediction module, in communication connection with the multi-physical field coupling simulation module, is configured to receive the key characteristic parameters output by the simulation module, and utilize a pre-trained random forest regression model to combine real-time collected voltage fluctuation data, hydrogen permeation rate and membrane thickness change to output a residual lifetime prediction value; A cooperative processing platform, integrated with an activation energy calculation unit, a decay weight distribution unit and a failure threshold determination unit, is configured to couple and process data streams of the simulation module and the prediction module.

[0005] Preferably, the control equations of the multi-physical field coupling simulation module include an electric field distribution control equation, a two-phase flow mass transfer control equation, an electrode reaction kinetics equation and a heat conduction equation.

[0006] Preferably, the multi-physical field coupling simulation module is parameterized by experimental data, and the key parameters calibrated include an ohmic resistance r1=0.001 Ω·cm² and an overvoltage coefficient s1=0.5.

[0007] Preferably, the activation energy calculation unit works based on the Arrhenius equation k=A·e^(-Ea / RT), wherein for an iridium catalyst, the pre-exponential factor A is determined to be 1.2×10 6 h⁻¹, and the activation energy Ea is 50 kJ / mol.

[0008] Preferably, the multi-physical field coupling simulation module is configured to operate in a working condition range of a current density of 1-3 A / cm², a temperature of 60-80℃ and a pressure of 0.1-3 MPa.

[0009] Preferably, the input features of the adaptive lifetime prediction module include the membrane resistance change rate and the catalyst activity decay coefficient characteristic parameters output by the simulation module.

[0010] The application further discloses a PEM electrolytic cell lifetime prediction method, specifically including the following steps: S1, data acquisition and simulation: deploying a sensor network to collect operating parameters of the electrolytic cell, and inputting the parameters into a multi-physical field coupling simulation module to obtain key characteristic parameters output by simulation; S2, equivalent aging time conversion: based on the Arrhenius equation, utilizing an activation energy calculation unit to convert a real-time state under the operating parameters into an equivalent aging time for correcting the output of the multi-physical field coupling simulation module; S3, decay weight distribution: through a decay weight distribution unit, the corrected key characteristic parameters are weighted processed; S4, model prediction: input the key feature parameters and the real-time collected voltage fluctuation data, hydrogen permeation rate and membrane thickness change into the random forest regression model in the adaptive life prediction module; the random forest model is trained by not less than 1000 groups of historical failure data sets, and 5-fold cross-validation and Bayesian hyperparameter optimization are adopted; S5, residual life output: according to the output of the random forest regression model, the residual life prediction value of the PEM electrolyzer is obtained.

[0011] Preferably, in step S2, the construction of the random forest regression model is performed by generating training subsets by Bootstrap sampling, selecting the optimal split point from the random feature subset when splitting the node, and predicting the regression result by integrating the average value of all decision trees.

[0012] The present application provides a PEM electrolyzer life decay model and a life prediction method. Compared with the prior art, the following beneficial effects are achieved: 1. By constructing an electrochemical-heat transfer-mass transfer multi-physical field coupling model, the internal physical field of the electrolyzer is accurately simulated; the model considers the interaction of the multi-physical processes of electric field distribution, two-phase flow mass transfer, electrode reaction kinetics and heat conduction, and can accurately simulate the degradation behavior of the membrane electrode assembly under different working conditions; and by calibrating the key parameters through experimental data, the simulation accuracy of the model under complex working conditions is improved, providing a reliable theoretical tool for electrolyzer performance evaluation and structure optimization.

[0013] 2. The accelerated aging model based on the Arrhenius equation establishes a quantitative relationship between temperature and material degradation rate; by calibrating the intrinsic parameters of the material through experiments, the acceleration effect of temperature on the chemical degradation process of the material can be accurately quantified; by updating the degradation parameters online, the adaptability and accuracy of life prediction are improved, and key technical support is provided for life management of the electrolyzer under non-steady state operation.

[0014] 3. The random forest regression algorithm is adopted to integrate multi-dimensional feature data and realize high-precision life prediction; the algorithm improves the generalization ability and prediction stability of the model through feature random selection and sample resampling strategy; the model has good interpretability and can identify the key factors affecting the life, providing data-driven basis for identification of life decay dominant factors and optimization of operation strategy. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The method steps in the present application are shown in the block diagram; Figure 2 The three-layer structure cooperative working mechanism in the present application is shown in the block diagram; Figure 3 The system integration logic flowchart in the present application is shown in the block diagram. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0017] Please refer to Figure 1 Figure 3 The present application provides a technical solution: a PEM electrolyzer life attenuation model, comprising: A multi-physics coupling simulation module is configured to perform electrochemical-thermal-fluid three-field coupling simulation based on finite element analysis, simulate the degradation behavior of the membrane electrode assembly by solving a set of control equations, and output a local stress concentration coefficient and a proton conductivity attenuation rate for quantifying performance attenuation; An adaptive life prediction module is in communication connection with the multi-physics coupling simulation module, configured to receive the key feature parameters output by the simulation module, and output a remaining life prediction value by using a pre-trained random forest regression model in combination with real-time collected voltage fluctuation data, hydrogen permeation rate and membrane thickness change; A collaborative processing platform is integrated with an activation energy calculation unit, an attenuation weight distribution unit and a failure threshold determination unit, and is configured to couple and process the data streams of the simulation module and the prediction module.

[0018] In the present embodiment, the multi-physics coupling simulation module is based on the COMSOL Multiphysics platform, and by simultaneously solving the Ohm's law equation describing the electric field distribution, the Navier-Stokes equation and the continuity equation describing the gas-liquid two-phase flow, the Butler-Volmer equation describing the electrochemical reaction, and the heat conduction equation including the Joule heat and the reaction heat, the stress distribution and the proton conduction characteristic change of the membrane electrode assembly are accurately simulated in the typical working condition range of 1-3 A / cm² current density, 60-80℃ temperature and 0.1-3 MPa pressure, and the key physical field parameters such as the local stress concentration coefficient and the proton conductivity attenuation rate are output. The adaptive life prediction module receives the above parameters, combines the real-time collected voltage fluctuation, hydrogen permeation rate and membrane thickness data, inputs the random forest regression model trained by 1000 groups of historical data, using Bootstrap sampling and feature random selection, and outputs the remaining life estimate value through the prediction results of the integrated multiple decision trees. The collaborative processing platform then calculates the activation energy based on the Arrhenius equation (parameter A=1.2×10 6 ​(h⁻¹, Ea=50kJ / mol) realizes temperature-lifetime mapping and equivalent aging time conversion. The characteristic parameters are optimized and weighted by considering the change of interface contact resistance through the decay weight allocation unit. Finally, the failure threshold judgment unit integrates multi-parameter coupling analysis to realize accurate judgment of the life end, forming a complete prediction chain of "physical simulation-chemical correction-data optimization", so that the overall prediction error is controlled within 8%.

[0019] Specifically, the control equations of the multiphysics coupled simulation module include the electric field distribution control equation, the two-phase mass transfer control equation, the electrode reaction kinetics equation, and the heat conduction equation.

[0020] In this embodiment, the electric field distribution follows Ohm's law, and its governing equation is ∇⋅(σ∇ϕ)=0, where ϕ represents the electric potential (unit: V), and σ is the local conductivity (unit: S / m). This conductivity is not a constant value, but a function affected by the coupling of multiple physical fields, specifically expressed as σ=f(T,λ,C_H+,ε), where T represents the temperature (unit: K), λ is the hydration degree of the proton exchange membrane, C_H+ is the proton concentration (unit: mol / m³), and ε is the porosity of the porous medium. This equation accurately describes the spatial distribution of the electric potential in the complex electrolytic cell environment by considering the comprehensive influence of multiple factors such as temperature, moisture, proton transport, and material structure on conductivity. The mass transfer process in two-phase flow is described by the simultaneous Navier-Stokes equations and the continuity equation. The continuity equation ∂ρ / ∂t + ∇⋅(ρu) = 0 embodies the law of conservation of mass, where ρ is the fluid density (kg / m³), u is the velocity vector (m / s), and ∇⋅(ρu) represents the mass flow rate difference between the outflow and inflow of the control volume per unit time. The Navier-Stokes equation ρ(∂u / ∂t + u⋅∇u) The equation ∇∇p + μ∇²u + F completely describes the conservation of momentum in the fluid: the left side of the equation represents the inertial force, and the right side consists of the pressure gradient term (∇p, unit Pa), the viscous diffusion term (μ∇²u, where μ is the dynamic viscosity in Pa·s), and the volume force F (such as gravity or electromagnetic force, unit N / m³). This set of equations accurately reproduces the complex transport behavior of the gas-liquid two-phase fluid in the porous electrode within the electrolyzer by coupling the velocity field and the pressure field. The electrode reaction kinetics are described by the Butler-Volmer equation j = j0[exp(αnFη / RT) - exp(-(1-α)nFη / RT)]. This equation establishes the constitutive relationship between overpotential and reaction rate through kinetic parameters; where j represents the electrode current density (A / m²), j0 is the exchange current density characterizing the intrinsic activity of the electrode (A / m²), η is the overpotential (V) driving the reaction, α is the charge transfer coefficient describing the symmetry of the reaction energy barrier, n represents the number of electrons transferred in the reaction, F and R are the Faraday constant (96485 C / mol) and the gas constant (8.314 J / (mol·K)), respectively, and T is the system temperature (K). This equation characterizes the competitive relationship between the anodic and cathodic reactions through two exponential terms, and can accurately describe the kinetic characteristics of the entire process from the linear response in the low overpotential region to the Tafel behavior in the high overpotential region. The heat transfer process is described by the heat conduction equation, which includes a heat source term: ρc p ∂T / ∂t+∇⋅(uρc p T) = ∇⋅(k) e ff∇T)+ +Qᵣ ea c t ᵢ on The first term on the left side of the equation is ρc. p ∂T / ∂t represents the transient heat storage effect of the system (where ρ is the fluid density, c...). p For specific heat capacity), the second term ∇⋅(uρc) p T) characterizes the energy transport caused by fluid convection and heat transfer; the first term on the right, ∇⋅(k) e ff∇T) describes the efficient heat conduction (k) between the porous medium framework and the fluid. e (where ff is the effective thermal conductivity); the key to the equation lies in its complete coupling of two main heat sources: one is... That is, Joule heating generated by electric current conduction, the intensity of which is... =j² / σ e ff determines (j is the current density, σ) e ff is the effective conductivity); the second is Qᵣ ea c t ᵢ on The heat of reaction associated with the hydrogen evolution / oxygen evolution electrode reaction is directly related to the enthalpy change of the electrochemical reaction. This equation, by coupling fluid flow, heat conduction and electrochemical heat generation, achieves an accurate simulation of the temperature field inside the electrolyzer.

[0021] Specifically, the multiphysics coupling simulation module calibrates parameters using experimental data. The key parameters calibrated include ohmic resistance r1 = 0.001 Ω·cm² and overvoltage coefficient s1 = 0.5.

[0022] In this embodiment, to ensure the accuracy and engineering applicability of the multiphysics coupling simulation model, the key parameters of the model were precisely calibrated using experimental data from the system. Based on the voltage-current characteristic curves under different current densities, the ohmic resistance r1 = 0.001 Ω·cm² was obtained by fitting using the nonlinear least squares method. This parameter accurately characterizes the comprehensive ohmic loss of the proton exchange membrane, catalyst layer, and bipolar plate in the membrane electrode assembly. Secondly, by analyzing the nonlinear segment characteristics of the polarization curves under different operating conditions, the overvoltage coefficient s1 = 0.5 was calibrated. This coefficient serves as a key correction parameter for the Butler-Volmer equation, effectively quantifying the coupling effect of charge transport and mass transfer during the electrode reaction process.

[0023] Specifically, the activation energy calculation unit operates based on the Arrhenius equation k = A·e^(-Ea / RT), where for iridium catalysts, the pre-exponential factor A is determined to be 1.2 × 10^- ... 6 h⁻¹, activation energy Ea is 50 kJ / mol.

[0024] In this embodiment, the activation energy calculation unit is constructed based on the classical Arrhenius equation of chemical kinetics, k = A·e^(-Ea / RT). This equation establishes a quantitative relationship between the material degradation rate constant k and the absolute temperature T, thus directly linking the operating temperature of the electrolyzer to its aging rate. For the iridium catalyst used in the model, the system monitors the decay of catalyst activity over time under accelerated aging experimental conditions of 80℃ and 1.5 A / cm², and fits the Arrhenius equation using degradation data at different temperature points, ultimately accurately determining the pre-exponential factor A = 1.2 × 10⁻⁶. 6 h⁻¹, with an activation energy of Ea = 50 kJ / mol, enables this unit to accurately quantify the accelerating effect of temperature on the chemical degradation process of materials.

[0025] Specifically, the multiphysics coupling simulation module is configured to operate within a range of current density of 1-3 A / cm², temperature of 60-80℃, and pressure of 0.1-3 MPa.

[0026] In this embodiment, within this parameter space, the module can accurately reproduce the mild reaction state under low load conditions of 1 A / cm² to the intense gas-liquid two-phase flow and significant Ohmic thermal effect under high load conditions of 3 A / cm² by solving the electrochemical-thermal-fluid control equations in a coupled manner. The temperature field simulation covers the nonlinear change in membrane proton conductivity from the 60°C start-up stage to the optimal efficiency region of 80°C. The pressure field analysis spans the influence of gas solubility changes on reaction kinetics from atmospheric pressure test conditions to high pressure operation of 3 MPa. This simulation capability covering all operating conditions ensures that the model can accurately capture the evolution of key performance parameters such as stress distribution and proton conductivity decay of the membrane electrode under various actual operating conditions, providing a high-fidelity physical field data foundation for lifetime prediction.

[0027] Specifically, the input features of the adaptive lifetime prediction module include characteristic parameters such as the membrane resistance change rate and the catalyst activity decay coefficient output by the simulation module.

[0028] In this embodiment, the membrane resistance change rate directly reflects the decrease in proton conduction capacity of the proton exchange membrane due to mechanical stress and chemical degradation during long-term operation, and its value is obtained by coupling the real-time potential field and current density field; the catalyst activity decay coefficient is based on the evolution of the exchange current density in the Butler-Volmer equation, which quantifies the dissolution / agglomeration effect of the iridium catalyst under high pressure and high potential conditions.

[0029] This invention also discloses a method for predicting the lifespan of a PEM electrolyzer, which specifically includes the following steps: S1, Data Acquisition and Simulation: Deploy a sensor network to collect the operating parameters of the electrolyzer and input the parameters into the multiphysics coupled simulation module to obtain the key characteristic parameters of the simulation output; S2, Equivalent Aging Time Conversion: Based on the Arrhenius equation, the activation energy calculation unit is used to convert the real-time state under the operating parameters into the equivalent aging time, which is used to correct the output of the multiphysics coupling simulation module. S3, Attenuation weight allocation: The modified key feature parameters are weighted through the attenuation weight allocation unit; S4, Model Prediction: Key feature parameters, as well as real-time collected voltage fluctuation data, hydrogen permeability, and membrane thickness changes, are input into the random forest regression model in the adaptive lifetime prediction module; the random forest model is trained with no less than 1000 sets of historical fault datasets and uses 5-fold cross-validation and Bayesian hyperparameter tuning. S5, Remaining Lifetime Output: Based on the output of the random forest regression model, the predicted remaining lifetime of the PEM electrolyzer is obtained.

[0030] In this embodiment, in stage S1, operating parameters are collected in real time by a network of temperature, pressure, and current density sensors deployed on the electrolyzer. These parameters are then input into a multiphysics coupled simulation module to obtain key physical field characteristic parameters such as membrane resistance change rate. In stage S2, based on the Arrhenius equation, the activation energy calculation unit converts the real-time operating state into an equivalent aging time, and temperature compensation correction is applied to the simulation output. In stage S3, the characteristic parameters are weighted and optimized based on changes in structural parameters such as interface contact resistance using an attenuation weight allocation unit. In stage S4, the processed characteristic parameters are input together with real-time monitoring data such as voltage fluctuations and hydrogen permeability into a random forest regression model. This model is trained based on 1000 sets of historical fault data, and 5-fold cross-validation and Bayesian hyperparameter tuning are used to ensure prediction accuracy. Finally, in stage S5, the remaining lifetime prediction value is output, and its prediction error is verified to be controlled within 8%. This method, through the organic combination of physical simulation and data-driven approaches, significantly improves the accuracy of PEM electrolyzer lifetime prediction.

[0031] Specifically, in step S2, the random forest regression model is constructed by generating a training subset using Bootstrap sampling, selecting the optimal split point from the random feature subset when splitting nodes, and performing regression prediction by integrating the average of all decision tree prediction results.

[0032] In this embodiment, the random forest regression model is constructed using an ensemble learning framework. Bootstrap sampling is used to randomly sample multiple training subsets with replacement from the original training dataset, allowing each decision tree to be trained on different data samples. This effectively enhances the model's generalization ability and provides out-of-bag (OOB) data for unbiased estimation. During node splitting in each tree, the algorithm randomly selects a subset of features from all features and searches for the optimal split point only within this subset. This dual randomness mechanism significantly reduces the correlation between trees and effectively suppresses overfitting. Finally, by integrating the prediction results of all decision trees, a simple averaging method is used to synthesize the remaining lifetime values ​​output by each tree as the final regression prediction result. This fully utilizes the advantages of collective decision-making, improving the model's ability to capture complex nonlinear relationships while maintaining low variance, ensuring the accuracy and stability of lifetime prediction.

[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A PEM electrolyzer lifetime decay model, characterized in that, include: The multiphysics coupling simulation module is used to perform electrochemical-thermal-fluid three-field coupling simulation based on finite element analysis. The simulation simulates the degradation behavior of the membrane electrode assembly by solving a set of governing equations. The simulation module is configured to output the local stress concentration factor and proton conductivity decay rate for quantifying performance degradation. The adaptive lifetime prediction module communicates with the multiphysics coupled simulation module to receive key feature parameters output by the simulation module, and uses a pre-trained random forest regression model, combined with real-time collected voltage fluctuation data, hydrogen permeability and membrane thickness changes, to output the remaining lifetime prediction value. The collaborative processing platform integrates an activation energy calculation unit, a decay weight allocation unit, and a failure threshold determination unit, which is used to couple and process the data streams of the simulation module and the prediction module.

2. The PEM electrolyzer lifetime decay model according to claim 1, characterized in that: The governing equations of the multiphysics coupled simulation module include the electric field distribution control equation, the two-phase mass transfer control equation, the electrode reaction kinetics equation, and the heat conduction equation.

3. The PEM electrolyzer lifetime decay model according to claim 1, characterized in that: The multiphysics coupling simulation module calibrates parameters using experimental data. The key parameters calibrated include ohmic resistance r1 = 0.001 Ω·cm² and overvoltage coefficient s1 = 0.

5.

4. The PEM electrolyzer lifetime decay model according to claim 1, characterized in that: The activation energy calculation unit operates based on the Arrhenius equation k = A·e^(-Ea / RT), where the pre-exponential factor A for iridium catalysts is determined to be 1.2 × 10^- ... 6 h⁻¹, activation energy Ea is 50 kJ / mol.

5. The PEM electrolyzer lifetime decay model according to claim 1, characterized in that: The multiphysics coupling simulation module is configured to operate within a range of operating conditions, including current density of 1-3 A / cm², temperature of 60-80℃, and pressure of 0.1-3 MPa.

6. The PEM electrolyzer lifetime decay model according to claim 1, characterized in that: The input features of the adaptive lifetime prediction module include characteristic parameters such as the membrane resistance change rate and the catalyst activity decay coefficient output by the simulation module.

7. A method for predicting the lifespan of a PEM electrolyzer, characterized in that: The PEM electrolyzer lifetime decay model described in any one of claims 1-6 specifically includes the following steps: S1, Data Acquisition and Simulation: Deploy a sensor network to collect the operating parameters of the electrolyzer and input the parameters into the multiphysics coupled simulation module to obtain the key characteristic parameters of the simulation output; S2, Equivalent aging time conversion: Based on the Arrhenius equation, the activation energy calculation unit is used to convert the real-time state under the operating parameters into an equivalent aging time, which is used to correct the output of the multiphysics coupling simulation module. S3, Attenuation weight allocation: The modified key feature parameters are weighted through the attenuation weight allocation unit; S4, Model Prediction: Key feature parameters, as well as real-time collected voltage fluctuation data, hydrogen permeability, and membrane thickness changes, are input into the random forest regression model in the adaptive lifetime prediction module; the random forest model is trained with no less than 1000 sets of historical fault datasets and uses 5-fold cross-validation and Bayesian hyperparameter tuning. S5, Remaining Lifetime Output: Based on the output of the random forest regression model, the predicted remaining lifetime of the PEM electrolyzer is obtained.

8. The method for predicting the lifespan of a PEM electrolyzer according to claim 7, characterized in that: In step S2, the random forest regression model is constructed by generating a training subset using Bootstrap sampling, selecting the optimal split point from the random feature subset when splitting nodes, and performing regression prediction by integrating the average of all decision tree prediction results.

Images