Alkaline electrolytic cell life prediction method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of hydrogen energy, and discloses an alkaline electrolytic cell service life prediction method and device, computer equipment and a storage medium, and the method comprises the steps: collecting multi-source operation data of an alkaline electrolytic cell; based on the multi-source operation data, adopting a preset neural network model to construct a data driving model; based on the multi-source operation data, constructing a voltage mathematical model of the alkaline electrolytic cell by adopting an electrochemical theory of the alkaline electrolytic cell; and fusing the output of the data driving model and the output of the voltage mathematical model by adopting a preset fusion mode, and generating a life prediction index based on the fused output, the life prediction index being used for predicting the residual life of the alkaline electrolytic cell. The service life attenuation mechanism of the electrolytic cell is defined, the finally generated service life prediction index can comprehensively and accurately reflect the remaining service life of the electrolytic cell, dynamic optimization can be achieved depending on multi-source operation data, the long-term prediction precision is stable, and a scientific and reliable decision basis is provided for efficient operation and maintenance and advanced planning and replacement of the electrolytic cell.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydrogen energy, in particular to an alkaline electrolyzer life prediction method and device, computer equipment and a storage medium. BACKGROUND

[0002] Under the background of global energy transformation and carbon neutralization, hydrogen energy plays a key role in the new power system, has the characteristics of fully absorbing renewable energy, helping the flexibility of the power system and peak shaving. However, due to the volatility, intermittency and seasonality of renewable energy, the alkaline electrolyzer hydrogen production system cannot respond to the fluctuations of wind and light in time, resulting in frequent start-stop and power fluctuations, accelerating the aging of alkaline electrolyzer materials and seriously reducing the operating life of alkaline electrolyzers. The life of alkaline electrolyzers is affected by the activity decay of electrode catalysts, the aging of diaphragms, the corrosion of alkaline solution and the uneven heat and mass transfer, and the degradation mechanism of its life is not clear; the traditional life test method relies on long-term actual operation data, with a cycle of tens of thousands of hours, high cost and long time, which cannot meet the rapid iteration of modern industrialization; the existing life prediction model is based on the static operation assumption, without fully considering the complexity of the operating conditions, and it is difficult to determine the life decay mechanism, resulting in inaccurate prediction of the remaining life of the electrolyzer. SUMMARY

[0003] Therefore, the present application provides an alkaline electrolyzer life prediction method, device, computer equipment and storage medium to solve the problem that the electrolyzer operating life prediction model in the prior art cannot determine the life decay mechanism, resulting in inaccurate prediction of the remaining life of the electrolyzer.

[0004] In a first aspect, the present application provides an alkaline electrolyzer life prediction method, which comprises:

[0005] Collecting multi-source operating data of the alkaline electrolyzer;

[0006] Based on the multi-source operating data, a data-driven model is constructed using a preset neural network model;

[0007] Based on the multi-source operating data, an alkaline electrolyzer voltage mathematical model is constructed using the electrochemical theory of the alkaline electrolyzer;

[0008] The outputs of the data-driven model and the voltage mathematical model are fused using a preset fusion method, and a life prediction index is generated based on the fused output, which is used to predict the remaining life of the alkaline electrolyzer.

[0009] The alkaline electrolytic tank life prediction method provided by the application is designed through multi-source data support, a voltage mathematical model and data fusion driving, relies on an electrochemical theory of the alkaline electrolytic tank to construct the voltage mathematical model, ensures that the prediction result has clear physical interpretability, and avoids the blindness of a pure data-driven model deviating from the actual mechanism, utilizes a preset neural network model to construct a data-driven model, accurately captures complex attenuation factors such as electrode corrosion and catalyst deactivation that are difficult to cover by a mechanism model, and makes up for the scene adaptation limitation of a pure mechanism model, integrates the outputs of the two types of models through a preset fusion mode, takes into account the mechanism interpretability and complex working condition adaptability, clearly defines the life attenuation mechanism, and finally generates a life prediction index that can comprehensively and accurately reflect the residual life of the electrolytic tank and can be dynamically optimized based on multi-source operation data, has stable long-term prediction accuracy, provides a scientific and reliable decision basis for efficient operation and maintenance and early planning of replacement of the electrolytic tank, and solves the problem in the prior art that an electrolytic tank operation life prediction model is difficult to clearly define the life attenuation mechanism, resulting in inaccurate prediction of the residual life of the electrolytic tank.

[0010] In an optional embodiment, based on multi-source operation data, a data-driven model is constructed using a preset neural network model, including:

[0011] Data preprocessing of data cleaning and data denoising is performed on the multi-source operation data;

[0012] A plurality of data features are selected from the preprocessed multi-source operation data, and the plurality of data features are fused with preset physical features to obtain a high-dimensional feature matrix;

[0013] The high-dimensional feature matrix is dimensionally reduced and standardized to obtain a reduced time series feature matrix;

[0014] The data-driven model is constructed based on the reduced time series feature matrix, a preset neural network model and a preset physical constraint loss function.

[0015] The alkaline electrolytic tank life prediction method provided by the application is designed through data cleaning and denoising preprocessing, effectively eliminates outliers and high-frequency noise in the original multi-source operation data, lays a high-quality data foundation for the model, further fuses the selected data features with preset physical features to construct a high-dimensional feature matrix, covers time series and working condition information at the data level, and also incorporates physical mechanism correlation to avoid one-sidedness of a single feature, subsequent dimension reduction processing can reduce feature redundancy and improve model training efficiency, and standardization eliminates the interference of dimension differences on model learning and ensures the consistency of input features, finally, the reduced time series feature matrix, the preset neural network model and the preset physical constraint loss function are combined, the complex time series dependence is captured by relying on the neural network, and the model is prevented from deviating from the electrochemical law through the physical constraint, the preliminary combination of data driving and physical mechanism is realized, and the constructed data-driven model has precision, stability and physical reliability.

[0016] In an optional embodiment, based on the multi-source operation data, a preset neural network model is used to construct a data-driven model, and the data-driven model further comprises the following steps:

[0017] In the training process, the improved adaptive moment estimation algorithm is used to optimize the parameters of the data-driven model, and the Bayesian optimization is used to search the hyperparameters of the data-driven model.

[0018] The alkaline electrolyzer life prediction method provided by the application introduces the improved adaptive moment estimation algorithm and the Bayesian optimization in the data-driven model construction, which significantly improves the effectiveness and efficiency of the model training: the combination of the two enhances the data-driven model in training speed, prediction accuracy and generalization ability, and better adapts to the time sequence characteristics and decay law of the multi-source data of the alkaline electrolyzer.

[0019] In an optional embodiment, the multi-source operation data comprises process parameters and performance parameters; the process parameters comprise reversible voltage, operating temperature, operating voltage, alkali flow, alkali concentration, electrolysis current and effective electrolysis area; and the performance parameters comprise alkaline electrolyzer voltage.

[0020] The voltage mathematical model of the alkaline electrolyzer is represented by the following formula:

[0021]

[0022] Wherein, U is the theoretical voltage of the alkaline electrolyzer, U rev is the reversible voltage, T is the operating temperature, p is the operating pressure, v is the alkali flow, C is the alkali concentration, I is the electrolysis current, A is the effective electrolysis area, a1, b1, c1, d1, a2, b2, c2, d2, t1, t2, t3 and s are undetermined coefficients.

[0023] The alkaline electrolyzer life prediction method provided by the application comprehensively captures the correlation information of the working condition change and the equipment performance state by using the multi-source operation data which simultaneously covers the process parameters (reversible voltage, operating temperature, pressure, alkali flow and concentration, electrolysis current, effective electrolysis area) reflecting the operating conditions of the electrolyzer and the performance parameters (alkaline electrolyzer voltage) representing the core performance, provides rich and relevant input support for the model, and avoids the one-sidedness of a single data dimension; and the alkaline electrolyzer voltage mathematical model is strictly constructed based on the electrochemical mechanism, the key process parameters affecting the voltage are integrated into the formula in the form of an explicit mathematical relationship, the theoretical voltage calculation has clear physical interpretability, and through the design of undetermined coefficients, the model can be fitted based on the actual multi-source operation data, which not only retains the universality of the mechanism model, but also accurately adapts to electrolyzers of different specifications and different operating scenarios, and provides a reliable mechanism calculation benchmark for the subsequent fusion with the data-driven model and life prediction.

[0024] In an optional embodiment, the output of the data-driven model and the output of the voltage mathematical model are fused in a preset fusion manner, and a life prediction index is generated based on the fused output, including:

[0025] The theoretical voltage value of the alkaline electrolyzer is calculated based on the voltage mathematical model;

[0026] The actually collected multi-source operation data is obtained, and the actually collected multi-source operation data is input into the data-driven model to obtain a voltage future offset prediction value;

[0027] The theoretical voltage value of the alkaline electrolyzer and the voltage future offset prediction value are fused in a preset fusion manner to obtain a fused prediction voltage;

[0028] A life prediction index is generated based on the fused prediction voltage.

[0029] The alkaline electrolyzer life prediction method provided by the application first calculates the theoretical voltage value of the alkaline electrolyzer based on electrochemical theory, lays a physical mechanism foundation for prediction, ensures the scientificity and explainability of voltage calculation, and avoids deviating from the actual operation law; then the actual multi-source operation data is processed by a data-driven model to accurately capture the voltage future offset (for example, complex attenuation factors such as electrode corrosion and catalyst deactivation that mechanism models cannot cover), making up for the scene adaptation limitations of pure mechanism models; then the theoretical voltage value and the future offset are integrated in a preset fusion manner, so that the fused prediction voltage not only follows the electrochemical law and ensures the rationality, but also integrates the actual attenuation characteristics and improves the accuracy; finally, a life prediction index is generated based on the fused prediction voltage, and the index can comprehensively reflect the real running state and attenuation trend of the electrolyzer, providing a reliable basis for the remaining life prediction.

[0030] In an optional embodiment, the life prediction index is generated based on the fused prediction voltage, including:

[0031] The fused prediction voltage at the current time is taken as the input at the next time, and a recursive prediction mode is used for continuous prediction until the fused prediction voltage reaches a preset failure threshold, so as to obtain a life prediction index, and the life prediction index is a voltage decay rate or a remaining life cycle.

[0032] The alkaline electrolytic tank life prediction method provided by the application can continuously track the dynamic attenuation process of the electrolytic tank voltage over time, avoid one-sidedness of isolated moment prediction, and fully capture the continuity and relevance of the voltage time sequence change; meanwhile, the fusion prediction voltage reaching a preset failure threshold is used as a termination condition, so that the prediction is strictly anchored to the actual failure standard of the electrolytic tank, and the life prediction indexes (voltage attenuation rate, residual life cycle) are highly consistent with the engineering actual demand, which can not only quantify the rate of voltage attenuation (voltage attenuation rate), but also directly give the landing residual service time (residual life cycle), and provide intuitive and accurate basis for electrolytic tank operation and maintenance decision (such as advance planning of maintenance and replacement).

[0033] In an optional embodiment, the alkaline electrolytic tank life prediction method further comprises:

[0034] Based on the fluctuation frequency of the operating conditions of the alkaline electrolytic tank, a preset online learning algorithm is used to periodically update the parameters of the data-driven model.

[0035] The alkaline electrolytic tank life prediction method provided by the application can accurately match the dynamic operating characteristics of the electrolytic tank, can shorten the update period to quickly adapt to new changes when the operating conditions (such as temperature, pressure, and alkali concentration) fluctuate frequently, and can prolong the period to avoid invalid power consumption when the operating conditions are stable, thereby realizing the pertinence and efficiency of parameter updating, effectively solving the problem of prediction accuracy decay caused by changes in the operating conditions of the traditional offline model. The online learning algorithm does not need to retrain all historical data, but only iteratively optimizes the model parameters based on new operating data, which can save computing resources, improve update efficiency, continuously enhance the data-driven model's ability to capture complex decay factors (such as electrode corrosion and catalyst deactivation), and ensure that the model maintains high prediction accuracy for a long time, thereby providing dynamic support for continuous and reliable prediction of the residual life of the alkaline electrolytic tank.

[0036] In a second aspect, the application provides an alkaline electrolytic tank life prediction device, which comprises:

[0037] A multi-source operating data acquisition module is configured to acquire multi-source operating data of the alkaline electrolytic tank.

[0038] A data-driven model construction module is configured to construct a data-driven model based on the multi-source operating data and using a preset neural network model.

[0039] A voltage mathematical model construction module is configured to construct a voltage mathematical model of the alkaline electrolytic tank based on the multi-source operating data and using the electrochemistry theory of the alkaline electrolytic tank.

[0040] A remaining life prediction module is configured to fuse the output of the data-driven model and the output of the voltage mathematical model by using a preset fusion manner, and generate a life prediction index based on the fused output, the life prediction index being used to predict the remaining life of the alkaline electrolyzer.

[0041] In a third aspect, the present application provides a computer device, comprising a memory and a processor, the memory and the processor being communicatively connected with each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the alkaline electrolyzer life prediction method of the first aspect or any of the corresponding embodiments thereof.

[0042] In a fourth aspect, the present application provides a computer readable storage medium, the computer readable storage medium storing computer instructions, the computer instructions being used to make a computer execute the alkaline electrolyzer life prediction method of the first aspect or any of the corresponding embodiments thereof.

[0043] In a fifth aspect, the present application provides a computer program product, comprising computer instructions, the computer instructions being used to make a computer execute the alkaline electrolyzer life prediction method of the first aspect or any of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0045] Figure 1 is a flowchart of the alkaline electrolyzer life prediction method according to an embodiment of the present application;

[0046] Figure 2 is a flowchart of another alkaline electrolyzer life prediction method according to an embodiment of the present application;

[0047] Figure 3 is a flowchart of still another alkaline electrolyzer life prediction method according to an embodiment of the present application;

[0048] Figure 4 is a flowchart of yet another alkaline electrolyzer life prediction method according to an embodiment of the present application

[0049] Figure 5 is a structural block diagram of the alkaline electrolyzer life prediction device according to an embodiment of the present application;

[0050] Figure 6 is a hardware structure schematic diagram of the computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. 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.

[0052] According to the embodiments of the present application, a kind of alkaline electrolytic cell life prediction method embodiment is provided, it needs to be explained, the steps shown in the flow chart of the accompanying drawings can be executed in computer system such as a group of computer executable instructions, and although logical sequence is shown in flow chart, in some cases, the steps shown or described can be executed in different order from here.

[0053] In the present embodiment, a kind of alkaline electrolytic cell life prediction method can be used for the computer equipment described above, Figure 1 It is the flow chart of the alkaline electrolytic cell life prediction method according to the embodiments of the present application, as Figure 1 As shown in the figure, the flow includes the following steps:

[0054] Step S101, the multi-source operation data of alkaline electrolytic cell is collected.

[0055] Specifically, alkaline electrolytic cell is one of the core equipment in the technology of electrolytic water hydrogen production, its core function is to generate hydrogen H2 and oxygen O2 by applying external electric energy to split water molecules in alkaline electrolyte environment, and it is the key hub connecting renewable energy (such as wind power, photovoltaic) and hydrogen energy industry chain (can convert unstable electric energy into storable hydrogen energy).

[0056] The multi-source operation data of alkaline electrolytic cell includes process parameters (reversible voltage, operating temperature, operating pressure, alkali flow, alkali concentration, electrolysis current, effective electrolysis area, etc.) and performance data (alkaline electrolytic cell voltage, electrolysis efficiency, etc.).

[0057] According to the different characteristics of process parameters and performance parameters, deploy adaptive sensors at key positions of electrolytic cell: such as using thermocouple / thermal resistance to collect operating temperature T, pressure transmitter to monitor hydrogen / oxygen side pressure p, electromagnetic flowmeter to record alkali circulation flow v, online densimeter or concentration sensor to obtain alkali concentration C, Hall current sensor and voltage transmitter to collect electrolysis current I and cell voltage U, and determine effective electrolysis area A through electrolytic cell design parameters; reversible voltage U rev Then, based on real-time collection of operating temperature T and operating pressure p, real-time calculation is carried out through Nernst equation.

[0058] Step S102, based on multi-source operation data, a preset neural network model is used to construct a data-driven model.

[0059] Specifically, the data-driven model can use random forest, time series neural network, ensemble learning algorithm, long short-term memory neural network (LSTM) or bidirectional long short-term memory network (BiLSTM), etc., to predict the long-term trend of voltage offset, and represent the decay factors not covered by the model (such as electrode corrosion, catalyst deactivation, and membrane damage).

[0060] First, the collected process parameters and performance parameters are preprocessed, such as data cleaning, denoising, etc., to eliminate abnormal data; then, key features are selected from the preprocessed data, and physical features such as polarization components are fused to form a high-dimensional feature matrix; after dimension reduction and standardization, it is input into a preset neural network model (such as BiLSTM), trained with a physical constraint loss function, and the model parameters and hyperparameters are adjusted through an optimization algorithm, and finally a data-driven model that can capture the voltage decay law of the electrolytic cell is obtained.

[0061] Step S103, based on multi-source operation data, an alkaline electrolytic cell electrochemical theory is used to construct a voltage mathematical model of the alkaline electrolytic cell.

[0062] Specifically, based on electrochemical theory (such as thermodynamic reversible voltage calculation, polarization effect mechanism), a voltage calculation formula containing reversible voltage U rev , process parameters (operating temperature T, operating pressure p, electrolysis current I, etc.) and undetermined coefficients is constructed; then, the collected multi-source operation data (measured voltage, values of various process parameters) are used to determine the undetermined coefficients in the formula through regression fitting, and finally a mathematical model reflecting the relationship between process parameters and theoretical voltage is obtained, realizing the voltage theoretical calculation based on working conditions.

[0063] Step S104, the output of the data-driven model and the output of the voltage mathematical model are fused using a preset fusion method, and a life prediction index is generated based on the fused output, which is used to predict the remaining life of the alkaline electrolytic cell.

[0064] Specifically, the theoretical voltage of the alkaline electrolytic cell is obtained through the voltage mathematical model, and the future voltage offset (reflecting the decay factors not covered by the mechanism) is obtained through the data-driven model; then the theoretical voltage and the voltage offset are integrated using a preset fusion method (such as weighted summation) to obtain a fused predicted voltage; finally, based on the fused predicted voltage (such as recursive prediction to the voltage reaching the failure threshold), life prediction indexes such as voltage decay rate and remaining life are generated to realize the prediction of the remaining life of the alkaline electrolytic cell.

[0065] The alkaline electrolyzer life prediction method provided by the embodiment, through multi-source data support, voltage mathematical model, and data fusion driven design, not only relies on the construction of the voltage mathematical model based on the electrochemical theory of the alkaline electrolyzer to ensure that the prediction result has clear physical interpretability and avoid the blindness of the pure data driven model deviating from the actual mechanism, but also uses a preset neural network model to construct a data driven model to accurately capture complex decay factors such as electrode corrosion and catalyst deactivation that cannot be covered by the mechanism model, and make up for the scene adaptation limitations of the pure mechanism model. At the same time, the outputs of the two types of models are integrated through a preset fusion method, taking into account the mechanism interpretability and complex working condition adaptability, and clearly defining the life decay mechanism. The final generated life prediction index can comprehensively and accurately reflect the remaining life of the electrolyzer, and can be dynamically optimized based on multi-source operation data, with stable long-term prediction accuracy, providing a scientific and reliable decision basis for efficient operation and maintenance and advance planning of replacement of the electrolyzer, and solving the problem that the electrolyzer operation life prediction model in the prior art cannot clearly define the life decay mechanism, resulting in inaccurate prediction of the remaining life of the electrolyzer.

[0066] In the embodiment, a kind of alkaline electrolyzer life prediction method can be used for the computer equipment described above, Figure 2 It is the flow chart of the alkaline electrolyzer life prediction method according to the embodiment of the application, as Figure 2 The flow chart includes the following steps:

[0067] Step S201, collecting multi-source operation data of the alkaline electrolyzer. For details, please refer to step S101 of the embodiment shown in Figure 1 The embodiment will not be described here.

[0068] Step S202, based on the multi-source operation data, a data driven model is constructed using a preset neural network model.

[0069] Specifically, the above step S202 includes:

[0070] Step S2021, data preprocessing of data cleaning and data denoising is performed on the multi-source operation data.

[0071] Specifically, the collected process parameters and performance parameters are preprocessed by data cleaning, denoising, etc. Abnormal data is removed. For example, wavelet threshold denoising is performed on the performance parameters (such as the original voltage data of the alkaline electrolyzer) in the multi-source operation data to remove high-frequency noise such as bubble disturbance and measurement error, and retain the degradation trend.

[0072] Step S2022, a plurality of data features are selected from the preprocessed multi-source operation data, and the plurality of data features are feature fused with a preset physical feature to obtain a high-dimensional feature matrix.

[0073] Specifically, data on the changes in the operating conditions of the alkaline electrolyzer were selected from process parameters and performance parameters in multi-source operating data as multiple data features, including the reversible voltage U. rev Operating temperature T, operating pressure p, alkali flow rate v, alkali concentration C, electrolysis current I, effective electrolysis area A, and cell voltage U after noise reduction. clean And the rate of change of voltage ΔU / Δt (reaction decay rate).

[0074] The preset physical characteristics are implemented in the following way: the reversible voltage is calculated based on the operating temperature and operating pressure in the process parameters; the polarization component is obtained by calculating the difference between the original voltage data (i.e., cell voltage) and the reversible voltage in the performance parameters, and the polarization component is used as the preset physical characteristics.

[0075] Integrating physical characteristics (polarization component) with multiple data characteristics (reversible voltage U) rev Operating temperature T, operating pressure p, alkali flow rate v, alkali concentration C, electrolysis current I, effective electrolysis area A, and cell voltage U after noise reduction. clean The high-dimensional feature matrix (number of samples × number of features × time step) is constructed using time alignment and sliding window truncation, along with the voltage change rate ΔU / Δt. The specific construction process is as follows:

[0076] 1. Define the feature set: Integrate all features to be included, including multiple data features (reversible voltage U). rev Operating temperature T, operating pressure p, alkali flow rate v, alkali concentration C, electrolysis current I, effective electrolysis area A, and cell voltage U after noise reduction. clean and voltage change rate ΔU / Δt) and physical characteristics (polarization component V) pol This forms a feature list (the number of features is the length of the list).

[0077] 2. Time alignment: Align all features according to timestamps (e.g., using 1 hour as the smallest time unit) to ensure that all features at the same time (e.g., multiple data features and polarization components at time t) correspond one-to-one, thus solving the problem of differences in the acquisition frequency of multi-source data.

[0078] 3. Sliding window sample generation: Using a fixed time step (e.g., 20 steps, i.e., 20 consecutive time points), temporal features are extracted in a sliding manner.

[0079] First sample: containing t1 to t 20 All features at each time step (one row of feature vectors for each time step);

[0080] The second sample includes t2 to t3. 21 All the characteristics of a moment;

[0081] This process continues until all time-series data has been traversed, resulting in several independent samples.

[0082] 4. Combining into a three-dimensional matrix: Each sample is a two-dimensional matrix of time step × number of features (each row represents all features at a certain moment). After all samples are stacked, a high-dimensional feature matrix of sample number × number of features × time step is formed. The time step dimension preserves the temporal change characteristics of the data, while the number of features dimension integrates data features and physical features, taking into account both dynamic laws and mechanistic constraints.

[0083] Step S2023: After reducing the dimensionality and standardizing the high-dimensional feature matrix, a dimensionality-reduced time series feature matrix is ​​obtained.

[0084] Specifically, Principal Component Analysis (PCA) is used to reduce the dimensionality of the high-dimensional feature matrix, retaining principal components with a cumulative contribution rate ≥95% to reduce redundant information. The specific processing procedure of PCA can be found in relevant technical documents and will not be elaborated here.

[0085] Step S2024: Construct a data-driven model based on the dimensionality-reduced temporal feature matrix, a preset neural network model, and a preset physical constraint loss function.

[0086] Specifically, the preset neural network model in this embodiment adopts the BiLSTM neural network model. The model structure of the BiLSTM neural network model includes: an input layer with the dimension being the number of features after dimensionality reduction; two bidirectional LSTM hidden layers, which extract time-series features in parallel through forward and backward LSTM units; and a fully connected output layer for predicting voltage degradation values.

[0087] Physical constraint loss function: A theoretical range constraint term for polarization voltage is introduced into the loss function, expressed by the following formula:

[0088] Loss = MSE(V) pred V true )+λ·∑max(0,∣V pol,pred -V pol,theo |-δ) (1);

[0089] Among them, V pred V is the predicted voltage output of the data-driven model. true V represents the actual tank voltage collected. pol ,pred represents the polarization component predicted by the model, V pol,theoThe polarization voltage range is calculated based on electrochemical theory. λ is the constraint weight, δ is the allowable error range, and model training and hyperparameter optimization are performed. MSE is an abbreviation for Mean Squared Error, which represents the mean square error. This function is based on the fitting error between the predicted and measured values, and incorporates electrochemical mechanism constraints such as polarization voltage range and voltage decay trend (to avoid predictions deviating from physical laws).

[0090] In step S2025, during the training process, an improved adaptive moment estimation algorithm is used to optimize the data-driven model parameters, and Bayesian optimization is used to search for the hyperparameters of the data-driven model.

[0091] Specifically, an improved Adaptive Moment Estimation with Weight Decay (AdamW) algorithm is used to optimize model parameters and suppress overfitting.

[0092] Hyperparameter optimization: Search for optimal parameters using Bayesian optimization (20-60 hidden layer neurons, learning rate 1e). -4 up to 1e -2 With a time step of 10-30, the objective function is to minimize the root mean square error (RMSE) of the validation set partitioned by the dimensionality-reduced time series feature matrix.

[0093] The dimensionality-reduced temporal feature matrix (number of samples × number of features × time step) is used as input to adapt to the temporal input format of the BiLSTM neural network model. The temporal dependencies of the features are captured through the hidden layers (bidirectional LSTM layers) of the neural network. A pre-set physical constraint loss function is then introduced during model training. Finally, the model parameters are iteratively adjusted through optimization algorithms (such as the improved Adam) to minimize the physical constraint loss function, resulting in a data-driven model that can fit the temporal patterns in the dimensionality-reduced features and is constrained by physical mechanisms.

[0094] Step S203: Based on multi-source operating data, construct a voltage mathematical model of the alkaline electrolyzer using the electrochemical theory of alkaline electrolyzers.

[0095] Specifically, the multi-source operating data includes process parameters and performance parameters; process parameters include reversible voltage, operating temperature, operating voltage, alkaline solution flow rate, alkaline solution concentration, electrolysis current, and effective electrolysis area; performance parameters include alkaline electrolytic cell voltage;

[0096] Based on multi-source operating data, mathematical models of voltage, current, temperature, pressure, alkaline solution flow rate, and alkaline solution concentration in alkaline electrolyzers were established using the electrochemical theory of alkaline electrolyzers.

[0097] The mathematical model for the voltage of an alkaline electrolyzer is expressed by the following formula:

[0098]

[0099] Where U is the theoretical voltage of the alkaline electrolytic cell, U rev Here, is the reversible voltage, T is the operating temperature, p is the operating pressure, v is the alkali flow rate, C is the alkali concentration, I is the electrolysis current, A is the effective electrolysis area, and a1, b1, c1, d1, a2, b2, c2, d2, t1, t2, t3 and s are undetermined coefficients. They are calculated by fitting experimental data under theoretical constraints. The core is to determine the physical boundaries of the coefficients by combining electrochemical mechanisms, and then solve them by regression methods using multi-source operating data.

[0100] If the model can be linearized (e.g., ignoring logarithmic terms and simplifying to linear equations), use the linear least squares method (e.g., using MATLAB's polyfit or Python's numpy.linalg.lstsq) to solve for the coefficients;

[0101] For a complete nonlinear model containing logarithmic and exponential terms, nonlinear least squares optimization (such as Python's scipy.optimize.curve_fit or MATLAB's lsqcurvefit) is used to iteratively solve for the coefficients with the objective of minimizing the sum of squared residuals between the calculated voltage and the measured voltage.

[0102] Historical operating data of the alkaline electrolyzer, such as voltage, current, temperature, pressure, alkaline solution flow rate, and alkaline solution concentration, are collected by sensors in the alkaline water electrolysis hydrogen production system.

[0103] Based on the historical operating data collected from the sensor, the above formula (2) is fitted and analyzed using the multivariate nonlinear regression method. At the same time, the historical operating data collected from the sensor is divided into a training set and a validation set. The training set is used to train and correct the above model to obtain an infinitely accurate voltage model. After all parameters are determined, the validation set is used for verification.

[0104] The root mean square error (RMSE) is used to evaluate the prediction results of the above model. The RMSE formula is as follows:

[0105]

[0106] Where m is the number of voltage samples collected, U i U represents the actual voltage collected from the alkaline electrolytic cell. o This is the predicted voltage of the mathematical model of the voltage of the alkaline electrolyzer. The smaller the RMSE value, the higher the accuracy of the mathematical model of the voltage of the alkaline electrolyzer.

[0107] It should be noted that 5-fold cross-validation (the model evaluation method) is used, and the evaluation metrics include root mean square error (RMSE), mean absolute error (MAE), and R². 2 (Determination coefficient). Specific verification methods can be found in relevant technologies and will not be elaborated here.

[0108] Step S204: The outputs of the data-driven model and the voltage mathematical model are fused using a preset fusion method. A lifetime prediction index is generated based on the fused output. This lifetime prediction index is used to predict the remaining lifetime of the alkaline electrolyzer. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0109] The alkaline electrolyzer lifetime prediction method provided in this embodiment utilizes multi-source operating data to simultaneously cover process parameters reflecting the electrolyzer's operating conditions (reversible voltage, operating temperature, pressure, alkali flow rate and concentration, electrolytic current, and effective electrolysis area) and performance parameters characterizing core performance (alkaline electrolyzer voltage). This comprehensively captures the correlation information between changes in operating conditions and equipment performance status, providing rich and relevant input support for the model and avoiding the one-sidedness of a single data dimension. The mathematical model for alkaline electrolyzer voltage is strictly constructed based on electrochemical mechanisms, incorporating key process parameters affecting voltage into the formula with clear mathematical relationships, ensuring that the theoretical voltage calculation has clear physical interpretability. At the same time, through the design of undetermined coefficients, it can be specifically fitted based on actual multi-source operating data, retaining the universality of the mechanism model while accurately adapting to electrolyzers of different specifications and operating scenarios, providing a reliable mechanism calculation benchmark for subsequent fusion with data-driven models and lifetime prediction.

[0110] This embodiment provides a method for predicting the lifespan of an alkaline electrolyzer, which can be used with the aforementioned computer equipment. Figure 3 This is a flowchart of an alkaline electrolyzer lifetime prediction method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0111] Step S301: Collect multi-source operating data of the alkaline electrolyzer. For details, please refer to [link to relevant documentation]. Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0112] Step S302: Based on multi-source operational data, a data-driven model is constructed using a pre-defined neural network model. For details, please refer to [link to relevant documentation]. Figure 2 Step S202 of the illustrated embodiment will not be described again here.

[0113] Step S303: Based on multi-source operating data, a voltage mathematical model of the alkaline electrolyzer is constructed using the electrochemical theory of alkaline electrolyzers. For details, please refer to [link to relevant documentation]. Figure 2Step S203 of the illustrated embodiment will not be described again here.

[0114] Step S304: The output of the data-driven model and the output of the voltage mathematical model are fused using a preset fusion method, and a lifetime prediction index is generated based on the fused output. The lifetime prediction index is used to predict the remaining lifetime of the alkaline electrolyzer.

[0115] Specifically, step S304 includes:

[0116] Step S3041: Calculate the theoretical voltage value of the alkaline electrolyzer based on the voltage mathematical model.

[0117] Specifically, the theoretical voltage value of the alkaline electrolyzer is calculated based on the voltage mathematical model, i.e., formula (2).

[0118] Step S3042: Obtain the actual collected multi-source operating data and input the actual collected multi-source operating data into the data-driven model to obtain the predicted value of future voltage offset.

[0119] Specifically, acquire actual multi-source operational data, including process parameters (reversible voltage U). rev Operating temperature T, operating pressure p, alkali flow rate v, alkali concentration C, electrolysis current I, effective electrolysis area A) and performance parameters (cell voltage U after noise reduction). clean The voltage change rate (ΔU / Δt) is obtained by sequentially performing the steps in S202 above to obtain a dimensionality-reduced time-series feature matrix (number of samples × number of features × time step). The dimensionality-reduced time-series feature matrix is ​​used as input to adapt to the time-series input format of the BiLSTM neural network model. The time-series dependencies of the features are captured through the hidden layer (bidirectional LSTM layer) of the neural network. Then, a preset physical constraint loss function is introduced in the model training. Finally, the model parameters are iteratively adjusted through an optimization algorithm (such as the improved Adam) to minimize the physical constraint loss function and obtain the predicted value of the future voltage offset.

[0120] Step S3043: The theoretical voltage value of the alkaline electrolyzer and the predicted future voltage offset value are fused using a preset fusion method to obtain the fused predicted voltage.

[0121] Specifically, the preset fusion methods include weighted fusion and dynamic coupling. For example, using the weighted fusion method, the weights of the theoretical voltage value of the alkaline electrolyzer and the predicted future voltage offset value are first determined separately:

[0122] If the prediction error of the data-driven model in historical data is small (e.g., RMSE < 0.05V), then the future voltage offset is assigned a higher weight (e.g., 0.8), and the theoretical voltage is assigned a lower weight (e.g., 0.2).

[0123] If the operating conditions fluctuate drastically (such as sudden temperature changes), the reliability of the short-term predictions of the data-driven model will decrease. Therefore, the weight of the theoretical voltage should be increased (e.g., 0.6), and the weight of the offset should be decreased (e.g., 0.4) to avoid interference from abnormal data.

[0124] The theoretical voltage value of the alkaline electrolyzer and the predicted future voltage offset value are fused together using a fusion method to obtain the fused predicted voltage.

[0125] The formula for calculating the fused predicted voltage is:

[0126] Fusion Predicted Voltage (U) pred = Theoretical voltage value (U) + Future voltage offset (ΔU) pred ).

[0127] For example: Theoretical voltage 1.85V + future voltage offset 0.12V = fused predicted voltage 1.97V.

[0128] Step S3044: Generate lifetime prediction index based on fusion prediction voltage.

[0129] In an optional implementation, step S3044 includes:

[0130] Step a: Use the fusion prediction voltage at the current moment as the input for the next moment, and continue to predict using a recursive prediction method until the fusion prediction voltage reaches the preset failure threshold to obtain the lifetime prediction index, which is the voltage decay rate or the remaining lifetime cycle.

[0131] Specifically, the predicted voltage at time t is used as the input at time t+1. The prediction is recursively continued until the voltage reaches a preset failure threshold (e.g., 1.2 times the initial operating voltage of an alkaline electrolyzer; if the initial voltage is 1.8V, then the failure threshold is set to 2.16V). At this point, the prediction stops, and two types of lifetime prediction indicators are generated:

[0132] Voltage decay rate: Calculates the average increment of the fused predicted voltage per unit time (e.g., 0.005V / day), reflecting the aging speed of the equipment (the higher the decay rate, the faster the aging).

[0133] Remaining lifespan: This is the time from the current moment until the voltage reaches the failure threshold (e.g., 180 days), which directly quantifies the time the equipment can still operate stably.

[0134] The final output voltage decay rate and remaining life cycle are used to predict the life of the alkaline electrolyzer, providing a basis for the operation and maintenance decisions of the electrolyzer.

[0135] Step S305: Based on the fluctuation frequency of the alkaline electrolyzer's operating conditions, the data-driven model parameters are periodically updated using a preset online learning algorithm.

[0136] Specifically, based on the fluctuation frequency of the alkaline electrolyzer's operating conditions (such as operating temperature and electrolysis current), the parameter update cycle of the data-driven model is dynamically set. The cycle is shortened when the operating conditions fluctuate frequently and extended when they are stable. Then, a preset online learning algorithm (incremental gradient descent and adaptive online AdamW algorithm) is used to periodically drive the model parameters based on newly added operating data, such as the gate weights of LSTM and the bias of fully connected layers. Regularization constraints (such as weight decay) are used to prevent the model from forgetting the stable patterns learned in the past (such as the basic correlation between temperature and voltage), while strengthening the learning of new fluctuation features (such as changes in the concentration-voltage relationship caused by new alkaline solutions). This preserves historical patterns and quickly adapts to new operating conditions, ensuring that the model maintains accurate predictive capabilities in the long term.

[0137] The alkaline electrolyzer lifetime prediction method provided in this embodiment first calculates the theoretical voltage value of the alkaline electrolyzer based on electrochemical theory, laying a physical mechanism foundation for the prediction and ensuring the scientific validity and interpretability of the voltage calculation, avoiding deviation from actual operating patterns. Then, it processes actual multi-source operating data through a data-driven model to accurately capture future voltage deviations (e.g., covering complex degradation factors such as electrode corrosion and catalyst deactivation that are difficult to cover by the mechanism model), compensating for the scenario adaptation limitations of the pure mechanism model. Subsequently, it integrates the theoretical voltage value and future deviations in a preset fusion method, so that the fused predicted voltage not only follows electrochemical laws and ensures rationality, but also incorporates actual degradation characteristics and improves accuracy. Finally, it generates a lifetime prediction index based on the fused predicted voltage. The index can comprehensively reflect the actual operating state and degradation trend of the electrolyzer, providing a reliable basis for predicting the remaining lifetime.

[0138] As one or more specific application embodiments of the present invention, combined with Figure 4 The method for predicting the lifespan of an alkaline electrolyzer provided by this invention will be further described in detail, such as... Figure 4 As shown, the specific process is as follows:

[0139] Step 1 involves simultaneously establishing a mathematical model of the alkaline electrolyzer voltage and collecting historical key data from sensors deployed in the alkaline electrolyzer (covering operating conditions, performance parameters, and other information of the electrolyzer).

[0140] Step 2: Based on the established voltage mathematical model and the collected historical data, perform multivariate nonlinear fitting on the model, and carry out training and correction to make the model initially fit the voltage characteristics of the electrolytic cell.

[0141] Step 3: Calculate the root mean square error based on the actual voltage and the theoretical voltage, and use this to quantify the predicted voltage deviation of the voltage mathematical model.

[0142] Step 4: Calculate the root mean square error and the operating parameters of the electrolytic cell, establish and train a data-driven model, and let the model capture the correlation between the error and the operating conditions.

[0143] Step 5: Using the trained data-driven model, predict the voltage decay trend and obtain the voltage change pattern over time, i.e., the future voltage offset.

[0144] Step 6: Incremental learning algorithm is introduced to optimize the parameters of the data-driven model in real time to adapt to the dynamic characteristics of alkaline electrolyzer operation and continuously improve the accuracy of the data-driven model.

[0145] Step 7: Integrate the outputs of the voltage mathematical model and the data-driven model, and finally generate lifetime prediction indicators, including voltage decay rate and remaining lifetime, through weighted or dynamic coupling methods, to provide a basis for the operation and maintenance decisions of the electrolyzer.

[0146] The alkaline electrolyzer life prediction method provided in this embodiment is based on a hybrid model of model and data-driven approach. The voltage mechanism model provides basic parameters, while the data-driven model compensates for dynamic factors not covered by the mechanism, overcoming the limitations and inaccuracies of a single model. Furthermore, implicit decay features are extracted through root mean square error, reducing the complexity of the data-driven model. An online learning algorithm is introduced to realize real-time changes in model parameters, enabling accurate prediction of voltage changes during electrolyzer operation. This allows for accurate prediction of the remaining life of the alkaline electrolyzer, extending the service life of the equipment, preventing sudden failures, reducing long-term costs, and avoiding safety accidents.

[0147] This invention can be used to integrate an intelligent operation and maintenance system for electrolyzers, enabling lifespan prediction and health management of alkaline electrolyzers, reducing operating costs and improving hydrogen production efficiency. Through accurate alkaline electrolyzer lifespan prediction, the health status of the alkaline electrolyzer can be assessed in real time, performance degradation trends can be identified, equipment lifespan can be extended, and operation and maintenance strategies optimized. Simultaneously, by monitoring reversible signals such as voltage and temperature, potential faults can be warned in advance.

[0148] This embodiment also provides an alkaline electrolyzer life prediction device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0149] This embodiment provides a device for predicting the lifespan of an alkaline electrolyzer, such as... Figure 5 As shown, it includes:

[0150] The multi-source operation data acquisition module 501 is used to acquire multi-source operation data of the alkaline electrolyzer.

[0151] The data-driven model building module 502 is used to build a data-driven model based on multi-source running data and a preset neural network model.

[0152] The voltage mathematical model construction module 503 is used to construct the voltage mathematical model of an alkaline electrolyzer based on multi-source operating data and the electrochemical theory of alkaline electrolyzers.

[0153] The remaining life prediction module 504 is used to fuse the output of the data-driven model and the output of the voltage mathematical model using a preset fusion method, and generate a life prediction index based on the fused output. The life prediction index is used to predict the remaining life of the alkaline electrolyzer.

[0154] In some alternative implementations, the data-driven model building module 502 includes:

[0155] The data preprocessing unit is used for data cleaning and noise reduction of multi-source running data.

[0156] The feature fusion unit is used to select multiple data features from the preprocessed multi-source running data and fuse these multiple data features with preset physical features to obtain a high-dimensional feature matrix.

[0157] The dimensionality reduction and standardization unit is used to reduce the dimensionality of a high-dimensional feature matrix and then standardize it to obtain a dimensionality-reduced time series feature matrix.

[0158] The data-driven model building unit is used to build data-driven models based on dimensionality-reduced time-series feature matrices, preset neural network models, and preset physical constraint loss functions.

[0159] In some alternative implementations, the data-driven model building module 502 further includes:

[0160] The parameter optimization unit is used to optimize the data-driven model parameters during training using an improved adaptive moment estimation algorithm and to search for data-driven model hyperparameters using Bayesian optimization.

[0161] In some optional implementations, multi-source operating data includes process parameters and performance parameters; process parameters include reversible voltage, operating temperature, operating voltage, alkali flow rate, alkali concentration, electrolysis current, and effective electrolysis area; performance parameters include the voltage of the alkaline electrolyzer; the mathematical model of the voltage of the alkaline electrolyzer is expressed by the following formula:

[0162]

[0163] Where U is the theoretical voltage of the alkaline electrolytic cell, U revdenoted as reversible voltage, T as operating temperature, p as operating pressure, v as alkali flow rate, C as alkali concentration, I as electrolysis current, A as effective electrolysis area, and a1, b1, c1, d1, a2, b2, c2, d2, t1, t2, t3, and s as undetermined coefficients.

[0164] In some alternative implementations, the remaining lifetime prediction module 504 includes:

[0165] The theoretical voltage calculation unit is used to calculate the theoretical voltage value of the alkaline electrolyzer based on the voltage mathematical model.

[0166] The voltage future offset prediction calculation unit is used to acquire the actual collected multi-source operating data and input the actual collected multi-source operating data into the data-driven model to obtain the voltage future offset prediction value.

[0167] The predicted voltage fusion unit is used to fuse the theoretical voltage value of the alkaline electrolyzer and the predicted future voltage offset value using a preset fusion method to obtain the fused predicted voltage.

[0168] A lifetime prediction index generation unit is used to generate a lifetime prediction index based on the fused prediction voltage. In one optional embodiment, the lifetime prediction index generation unit includes:

[0169] A recursive prediction subunit is used to continuously predict the voltage at the current moment using the fused prediction voltage as the input for the next moment, employing a recursive prediction method until the fused prediction voltage reaches a preset failure threshold, thereby obtaining a lifetime prediction index, which is either the voltage decay rate or the remaining lifetime cycle. In an optional embodiment, the alkaline electrolyzer lifetime prediction device further includes:

[0170] The online learning module is used to periodically update the data-driven model parameters based on the fluctuation frequency of the alkaline electrolyzer's operating conditions using a preset online learning algorithm.

[0171] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0172] In this embodiment, the alkaline electrolyzer life prediction device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0173] This invention also provides a computer device having the above-described features. Figure 5 The device shown is for predicting the lifespan of an alkaline electrolyzer.

[0174] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 6 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 6 Take a processor 10 as an example.

[0175] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0176] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0177] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0178] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0179] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0180] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0181] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0182] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0183] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for predicting the lifetime of an alkaline electrolyzer, characterized by, The method comprises: Collecting multi-source operation data of the alkaline electrolyzer; Based on the multi-source operation data, a data-driven model is constructed using a preset neural network model; Based on the multi-source operation data, an alkaline electrolyzer voltage mathematical model is constructed using an alkaline electrolyzer electrochemical theory; The outputs of the data-driven model and the voltage mathematical model are fused using a preset fusion method, and a life prediction index is generated based on the fused output, which is used to predict the remaining life of the alkaline electrolyzer.

2. The method of claim 1, wherein, The method comprises: Data preprocessing of the multi-source operation data, including data cleaning and data denoising; From the preprocessed multi-source operation data, a plurality of data features are selected, and the plurality of data features are fused with preset physical features to obtain a high-dimensional feature matrix; After dimension reduction and standardization of the high-dimensional feature matrix, a reduced time series feature matrix is obtained; Based on the reduced time series feature matrix, a preset neural network model and a preset physical constraint loss function, a data-driven model is constructed.

3. The method of claim 1, wherein, The method comprises: During the training process, the data-driven model parameters are optimized using an improved adaptive matrix estimation algorithm, and the data-driven model hyperparameters are searched using Bayesian optimization.

4. The method of claim 1, wherein, The multi-source operation data includes process parameters and performance parameters; the process parameters include reversible voltage, operating temperature, operating voltage, alkali flow, alkali concentration, electrolysis current and effective electrolysis area; the performance parameters include alkaline electrolyzer voltage; The voltage mathematical model of the alkaline electrolyzer is represented by the following formula: wherein U is the theoretical voltage of the alkaline electrolyzer, U rev is the reversible voltage, T is the operating temperature, p is the operating pressure, v is the flow rate of the alkaline solution, C is the concentration of the alkaline solution, I is the electrolysis current, A is the effective electrolysis area, a1, b1, c1, d1, a2, b2, c2, d2, t1, t2, t3, and s are undetermined coefficients.

5. The method of claim 1, wherein, The method comprises: Based on the voltage mathematical model, the theoretical voltage value of the alkaline electrolyzer is calculated; The actual collected multi-source operation data is input into the data-driven model to obtain the voltage future offset prediction value; The theoretical voltage value of the alkaline electrolyzer and the voltage future offset prediction value are fused using a preset fusion method to obtain a fused prediction voltage; Based on the fused prediction voltage, a life prediction index is generated.

6. The method of claim 5, wherein, The method comprises: The fused prediction voltage at the current time is taken as the input at the next time, and a recursive prediction method is used for continuous prediction until the fused prediction voltage reaches a preset failure threshold, and a life prediction index is obtained, which is a voltage decay rate or a remaining life cycle.

7. The method of claim 1, wherein, The method further comprises: Based on the fluctuation frequency of the alkaline electrolyzer operating condition, the data-driven model parameters are periodically updated using a preset online learning algorithm.

8. An alkaline electrolyzer lifetime prediction device, characterized by, The device comprises: A multi-source operation data collection module for collecting multi-source operation data of the alkaline electrolyzer; A data-driven model construction module for constructing a data-driven model based on the multi-source operation data using a preset neural network model; The voltage mathematical model construction module is configured to construct a voltage mathematical model of the alkaline electrolyzer based on multi-source operation data and an electrochemical theory of the alkaline electrolyzer. The remaining life prediction module is configured to fuse an output of the data-driven model and an output of the voltage mathematical model in a preset fusion manner, and generate a life prediction index based on the fused output, the life prediction index being used to predict the remaining life of the alkaline electrolyzer.

9. A computer device, comprising: The method comprises the following steps: A memory and a processor are in communication connection with each other, and the memory stores computer instructions. The processor executes the computer instructions to perform the alkaline electrolyzer life prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for enabling a computer to perform the alkaline electrolyzer life prediction method according to any one of claims 1 to 7.

Images