An intelligent control method and system for a PEM electrolyzer

Abstract

This invention discloses an intelligent control method and system for PEM electrolyzers, relating to the field of new energy technology. Addressing the problems of existing PEM electrolyzers under fluctuating operating conditions, such as difficulty in dynamically coordinating multiple physical fields, difficulty in sensing internal states, and the inability to balance operating efficiency and equipment lifespan, this invention proposes a multivariate collaborative optimization control based on the Electrolysis Process State Function (EPSF). By calculating the proton conduction efficiency function and the catalytic activity distribution optimization function in real time, the key internal states of the electrolyzer are accurately quantified. Furthermore, a multivariate optimization objective function integrating efficiency and degradation cost is constructed and solved to obtain optimal control parameters such as current density and pressure online. A multi-timescale rolling optimization architecture and a parameter online identification algorithm with a forgetting factor are further introduced to achieve synergy between rapid response, efficiency optimization, and lifespan management. This invention significantly improves the operating efficiency, stability, and durability of PEM electrolyzers under dynamic loads.

Description

Technical Field

[0001] This invention relates to the field of new energy technology, and in particular to an intelligent control method and system for PEM electrolyzers. Background Technology

[0002] Proton exchange membrane (PEM) water electrolysis technology is considered one of the most promising green hydrogen production technologies due to its high efficiency, fast response speed, high purity of product gas, and good compatibility with fluctuating renewable energy sources. The core component of a PEM electrolyzer is the membrane electrode assembly (MEA). Its working principle is that under the action of direct current, water molecules undergo an oxidation reaction at the anode to generate oxygen and protons. The protons migrate through the proton exchange membrane to the cathode, where a reduction reaction occurs to generate hydrogen.

[0003] However, in actual operation, especially when coupled with fluctuating renewable energy sources such as wind and solar power, PEM electrolyzers face multiple complex technical challenges, which severely restrict their efficiency, lifespan, and commercialization. The core of these challenges lies in the complex coupling and mutual constraints of multiple physical processes, including strong electrochemical, thermodynamic, and fluid dynamic processes, within the electrolyzer.

[0004] At the operational control level, existing technologies have the following main limitations: Lack of Dynamic Coordination Control: Most existing control strategies are based on single-variable or simple proportional-integral-derivative (PID) control, such as adjusting current or circulating water flow rate individually. However, the performance of a PEM electrolyzer is the result of highly nonlinear coupling of multiple variables, including current density, temperature, pressure, water flow rate, and membrane water content. For example, when the current is rapidly increased in response to power fluctuations, if the rates of water transport, proton conduction, and heat generation cannot be synchronized, it will lead to localized "drying" or "flooding" of the membrane electrode, causing a sharp increase in proton conduction resistance, an increase in overpotential, and even irreversible membrane damage. Existing methods lack modeling and coordination optimization of the dynamic relationships between multiple variables, resulting in a sharp drop in system efficiency and accelerated degradation under dynamic operating conditions.

[0005] Insufficient state perception and estimation capabilities: Many key state parameters inside the electrolyzer, such as the real-time water content of the proton exchange membrane, the local current density distribution on the catalyst surface, and the gas-liquid two-phase flow state within the electrode pores, are difficult to measure directly, quickly, and in situ using conventional external sensors. Existing systems typically rely on limited macroscopic signals such as inlet and outlet temperatures, pressures, total voltage, and current for control, making them typical "black box" or "gray box" controls. Due to the inability to accurately perceive critical internal microstates, the control system cannot achieve preventative adjustments, often only performing delayed and coarse compensation after performance degradation has occurred (such as hotspot formation or localized water shortage), resulting in poor control accuracy and robustness.

[0006] Improper trade-off between operating efficiency and lifespan: Traditional control aims to maintain constant operating conditions or track a given power curve, with little optimization from a life-cycle cost perspective. For example, operating at high current densities for extended periods in pursuit of instantaneous high efficiency may exacerbate catalyst agglomeration, ionomer degradation, and membrane chemical decay; while overly conservative low-load operation, although beneficial for extending lifespan, reduces equipment utilization and economic efficiency. Existing control strategies lack models that correlate real-time operating status with long-term degradation mechanisms, making it impossible to dynamically find the optimal balance between efficiency and lifespan, i.e., to achieve "economical operation."

[0007] Model-object mismatch: Some advanced control studies attempt to introduce predictive control based on fixed-parameter models. However, the performance parameters of PEM electrolyzers change slowly over time due to operating time (catalyst activity decay, membrane characteristic changes) and external conditions (water quality fluctuations). Fixed-parameter models cannot accurately describe these time-varying characteristics, leading to a decrease in model prediction accuracy over operating time, a gradual deterioration in control effectiveness, and the need for frequent manual intervention and model calibration.

[0008] In summary, existing PEM electrolyzer control technologies have failed to fundamentally address core issues such as dynamic coordination under strong multi-physics coupling, accurate internal state perception, efficiency-lifetime multi-objective optimization, and model adaptation. This has resulted in PEM electrolyzers not achieving ideal operating efficiency, reliability, durability, and economic efficiency in fluctuating power supply scenarios, becoming one of the key technological bottlenecks restricting the large-scale development of the green hydrogen energy industry. Therefore, there is an urgent need for a novel control method and system that can deeply integrate multi-physics process mechanisms, achieve intelligent multi-variable coordination, and possess state perception and adaptive capabilities. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent control method and system for PEM electrolyzers. The aim is to achieve dynamic collaborative optimization control of multiple variables of the electrolyzer by establishing and solving a mathematical model that integrates key internal states and global efficiency, rather than traditional single-variable or empirical control.

[0010] To achieve the above objectives, this application proposes an intelligent control method for PEM electrolyzers, comprising the following steps: Collect multi-physics sensor data during the operation of the PEM electrolyzer, wherein the sensor data includes at least temperature, pressure, current and voltage; Based on the sensor data, the state function of the electrolysis process is calculated. The state function of the electrolysis process includes at least a proton conduction efficiency function for characterizing the real-time proton conduction efficiency of the membrane electrode and a catalytic activity distribution optimization function for characterizing the catalyst utilization rate. Based on the state function of the electrolysis process, a multivariate optimization objective function J is constructed and solved to obtain the optimal set of control parameters; Based on the optimal set of control parameters, control commands are generated and the actuators of the PEM electrolyzer are adjusted. Wherein, the multivariate optimization objective function J is the efficiency maximization objective function, and its expression is: In the formula, For Faraday efficiency, For voltage efficiency, For power density, This is the performance degradation cost function based on changes in system state.

[0011] As a further solution, the formula for calculating the proton conduction efficiency function is: Where α is a material property constant, λ(t) is the real-time membrane water content, and λmin and λmax are the minimum and maximum water content of the membrane, respectively.

[0012] As a further solution, the formula for calculating the catalytic activity distribution optimization function is as follows: in, For local current density, For rated current density, β represents the local overpotential deviation, and β is the catalyst attenuation coefficient.

[0013] As a further solution, the formula for calculating the performance degradation cost function is as follows: Where k1 and k2 are weighting coefficients, and ΔTmax is the maximum temperature difference. This represents the magnitude of the current density gradient.

[0014] As a further solution, the optimal control parameter set includes at least the optimal current density and the optimal pressure difference between the anode and cathode; wherein, the formula for calculating the optimal current density is: In the formula, Let γ be the temperature effect function, and γ be the response coefficient. Δη is the proton conductivity threshold, and Δη is the width of the transition region.

[0015] As a further solution, an online parameter identification step is also included, in which the recursive least squares method with a forgetting factor is used to estimate and update the time-varying parameter vector θ in the state function of the electrolysis process in real time. The parameter vector θ includes at least the material characteristic constant α in the proton conduction efficiency function and the catalyst attenuation coefficient β in the catalytic activity distribution optimization function.

[0016] As a further solution, the step of constructing and solving the multivariate optimization objective function J is performed using a multi-timescale rolling optimization architecture, including: In the first millisecond timescale, a linearized model based on the system state is used to solve the fast dynamic compensation problem; On the second-second time scale, based on a nonlinear prediction model, the cost optimization problem is solved within the finite prediction time domain. At the third hour timescale, the lifetime management optimization problem is solved based on the performance degradation model.

[0017] As a further solution, the dynamic model of membrane water content λ(t) used in calculating the state function of the electrolysis process considers capillary pressure-driven two-phase flow transport, and its liquid water transport equation includes Darcy's law terms based on capillary pressure: Among them, v l Let K be the velocity of liquid water, and k be the permeability. rl μ represents relative permeability. l P is the dynamic viscosity of water. l P is the pressure of the liquid water. c It is capillary pressure.

[0018] As a further solution, the acquired sensor data includes: Temperature and strain fields are measured using fiber Bragg grating sensors embedded in bipolar plate channels. Charge transfer resistance and double-layer capacitance are obtained by online electrochemical impedance spectroscopy measurement and are used as state variables in the calculation of the state function of the electrolysis process.

[0019] On the other hand, the present invention also provides an intelligent control system for a PEM electrolyzer, for implementing an intelligent control method for a PEM electrolyzer as described in any of the preceding claims, the system comprising: The sensor layer comprises multiple sensors arranged in a distributed manner to collect temperature, pressure, current, and voltage data. The state assessment layer is communicatively connected to the sensor layer and is configured to calculate the state function of the electrolysis process based on the collected data. The collaborative control layer, which is communicatively connected to the state evaluation layer, is configured to construct and solve a multivariate optimization objective function and calculate the optimal set of control parameters. The actuator layer, which is communicatively connected to the collaborative control layer, includes an adjustable power supply, a water pump, and valves, and is configured to adjust the operating state of the PEM electrolyzer according to the optimal set of control parameters.

[0020] Compared with related technologies, the intelligent control method and system for PEM electrolyzers provided by this invention have the following advantages: 1. This invention achieves coordinated dynamic optimization of current, temperature, pressure, and water flow by constructing a multi-physics coupled electrolysis process state function (EPSF) and a multivariate optimization objective. This solves the problem of multi-parameter mismatch under power fluctuations and improves dynamic efficiency and stability.

[0021] 2. This invention utilizes a multi-timescale rolling optimization architecture based on a mechanistic model, combining millisecond-level dynamic compensation with second-level economic optimization. This reduces the critical control response time to within 100 milliseconds, an order of magnitude improvement over traditional second-level response, and provides precise and smooth adjustment. 3. This invention innovatively introduces a quantitative performance degradation cost function into the optimization objective, simultaneously optimizing instantaneous efficiency and factors that suppress material degradation (such as hot spots and uneven reactions), synergistically optimizing efficiency and lifespan, and reducing the total life cycle cost.

[0022] 4. This invention integrates an online parameter identification algorithm to update key model parameters in real time, overcoming the model mismatch problem caused by catalyst aging and changes in membrane properties, enabling the system to maintain optimal performance over a long period of time, eliminating the need for frequent manual calibration and possessing adaptive robustness.

[0023] 5. This invention transforms the complex internal processes of an electrolyzer into a precise calculation problem based on state functions and optimization objectives, providing a universally applicable quantitative framework and tools for the design, control, and evaluation of PEM electrolyzers, and establishing a quantifiable control theory.

[0024] 6. The core of this invention is based on an interpretable mathematical model and optimized calculation, without relying on a black-box AI model. It is easy to deploy in industrial controllers, providing a practical solution for the intelligentization of PEM hydrogen production equipment, and has high engineering feasibility. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0027] Figure 1 A schematic diagram illustrating the steps of an intelligent control method for a PEM electrolyzer provided by the present invention; Figure 2 This invention provides a schematic diagram of an intelligent control system structure for a PEM electrolyzer. The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0029] Example 1 Please see Figure 1 This embodiment provides an intelligent control method for PEM electrolyzers, including the following steps: Collect multi-physics sensor data during the operation of the PEM electrolyzer, wherein the sensor data includes at least temperature, pressure, current and voltage; Based on the sensor data, the state function of the electrolysis process is calculated. The state function of the electrolysis process includes at least a proton conduction efficiency function for characterizing the real-time proton conduction efficiency of the membrane electrode and a catalytic activity distribution optimization function for characterizing the catalyst utilization rate. Based on the state function of the electrolysis process, a multivariate optimization objective function J is constructed and solved to obtain the optimal set of control parameters; Based on the optimal set of control parameters, control commands are generated and the actuators of the PEM electrolyzer are adjusted. Wherein, the multivariate optimization objective function J is the efficiency maximization objective function, and its expression is: In the formula, For Faraday efficiency, For voltage efficiency, For power density, This is the performance degradation cost function based on changes in system state.

[0030] It should be noted that this method first collects multi-physical field data such as temperature, pressure, current, and voltage during the operation of the electrolyzer using sensors. Then, based on this data, two key state functions are calculated: a "proton conduction efficiency function" for real-time evaluation of the proton exchange membrane's conductivity and a "catalytic activity distribution optimization function" for evaluating the uniformity of catalyst reaction. Next, based on these two functions, a "multivariable optimization objective function" aimed at maximizing system efficiency is constructed and solved, thereby obtaining a set of optimal control parameters. Finally, instructions are generated based on these parameters to adjust the electrolyzer's power supply, water pumps, and other actuators in real time. This embodiment achieves dynamic, coordinated optimization control of the electrolyzer through the establishment and solution of a mathematical model that integrates key internal states and global efficiency, rather than traditional single-variable or empirical control.

[0031] Furthermore, the formula for calculating the proton conduction efficiency function is as follows: Where α is a material property constant, λ(t) is the real-time membrane water content, and λmin and λmax are the minimum and maximum water content of the membrane, respectively.

[0032] Specifically, the core of this function lies in linking the proton conduction efficiency, which is difficult to measure directly, with the crucial membrane water content state through an exponential model with the natural constant e as its base. The parameters in the formula have clear physical meanings. This formulaic definition enables the control system to quantitatively and continuously calculate the real-time state of the proton conduction efficiency based on the estimable water content λ(t), thus providing accurate and crucial internal state inputs for subsequent multivariate optimization. This solves the control lag and inaccuracy problems caused by the inability to perceive the internal water state of the membrane in traditional control.

[0033] Furthermore, the calculation formula for the catalytic activity distribution optimization function is as follows: in, For local current density, For rated current density, β represents the local overpotential deviation, and β is the catalyst attenuation coefficient.

[0034] Specifically, this function aims to quantify the real-time catalyst utilization at a specific location (x, y) on the catalyst layer plane of an electrolyzer. Its formula consists of two multiplied parts: the first term is the ratio of the local current density i_local to the nominal current density i_nominal, directly reflecting the immediate reaction intensity at that location; the second term is an exponential decay term with overpotential deviation ΔE as the variable, which penalizes local kinetic deviations caused by inhomogeneous reactions. By introducing spatial coordinates (x, y), this function achieves a real-time, quantitative assessment of the microscopic state of catalyst activity distribution. The catalyst decay coefficient β in the exponential term links the overpotential deviation to the catalyst decay rate, enabling the function to not only assess the current utilization but also predict the impact of local reaction conditions on long-term lifetime. This provides crucial mathematical basis for optimizing current density distribution, suppressing local hotspots, and mitigating decay.

[0035] Furthermore, the formula for calculating the performance degradation cost function is as follows: Where k1 and k2 are weighting coefficients, and ΔTmax is the maximum temperature difference. This represents the magnitude of the current density gradient.

[0036] Specifically, this function aims to quantify the key physical factors affecting the long-term lifespan of PEM electrolyzers into a mathematical cost term that can be directly calculated in real-time optimization. Its calculation formula contains two core parts: first, the square of the maximum temperature difference ΔT_max, used to penalize "hot spots" formed on the membrane electrode due to poor thermal management, as excessive temperature differences exacerbate thermomechanical stress and chemical degradation of the material; second, the current density gradient modulus | The squared term of i| is used to penalize excessively uneven current distribution caused by reactant unevenness or flow field design, which can accelerate local catalyst deactivation. This is achieved by summing these two physical quantities, directly related to the degradation mechanism, in quadratic form and assigning weighting coefficients k1 and k2, to C. deg The (t) function enables the optimization algorithm to actively suppress degradation modes that shorten the lifespan while pursuing immediate high efficiency, thus achieving coordinated optimization of operating efficiency and equipment lifespan at the mathematical level.

[0037] Furthermore, the optimal control parameter set includes at least the optimal current density and the optimal pressure difference between the anode and cathode; wherein, the formula for calculating the optimal current density is: In the formula, Let γ be the temperature effect function, and γ be the response coefficient. Δη is the proton conductivity threshold, and Δη is the width of the transition region.

[0038] Specifically, the core of this embodiment lies in the calculation formula for the optimal current density. It is not a fixed setpoint, but rather a dynamic feedback result based on the real-time system state. This formula uses a base current density as a benchmark and responds to changes in real-time proton conduction efficiency through a term containing a hyperbolic tangent function. This ensures that the current command can be smoothly increased when efficiency is good and cautiously limited when efficiency is insufficient. Simultaneously, it multiplies by a temperature influence function to guarantee that the operating temperature is always guided towards the optimal range. This mathematical structure allows the control command to automatically and continuously adapt to changes in the water and thermal states inside the electrolyzer, solving the problems of impact and efficiency loss caused by traditional step-like or segmented adjustments, and achieving adaptive and fine optimization of the current density.

[0039] Furthermore, it also includes an online parameter identification step, which uses a recursive least squares method with a forgetting factor to estimate and update the time-varying parameter vector θ in the state function of the electrolysis process in real time; The parameter vector θ includes at least the material characteristic constant α in the proton conduction efficiency function and the catalyst attenuation coefficient β in the catalytic activity distribution optimization function.

[0040] This embodiment adds a crucial online parameter identification step to the method flow. The purpose of this step is to address the technical challenge that, during long-term operation of a PEM electrolyzer, the properties of its core materials (such as proton exchange membranes and catalysts) slowly change with operating time and conditions, leading to the gradual inaccuracy of the state function model constructed based on fixed parameters.

[0041] This step specifically employs a recursive least squares method with a forgetting factor to estimate and update the key time-varying parameter vectors θ in the state function (e.g., the material constant α in the proton conduction efficiency function and the catalyst decay coefficient β in the catalytic activity distribution optimization function) in real time. By continuously comparing the real-time system output with the model's predicted output and automatically adjusting the parameter estimates using an algorithm, the mathematical model can "follow" the actual aging or state evolution process of the electrolyzer. This is equivalent to endowing the control system with "self-learning" and "adaptive" capabilities, ensuring the long-term accuracy of state assessment, and thus guaranteeing the continuity and robustness of the entire multivariate optimization control effect, overcoming the inherent defect of performance degradation due to parameter drift in fixed models.

[0042] Furthermore, the step of constructing and solving the multivariate optimization objective function J is executed using a multi-timescale rolling optimization architecture, including: In the first millisecond timescale, a linearized model based on the system state is used to solve the fast dynamic compensation problem; On the second-second time scale, based on a nonlinear prediction model, the cost optimization problem is solved within the finite prediction time domain. At the third hour timescale, the lifetime management optimization problem is solved based on the performance degradation model.

[0043] Specifically, this embodiment introduces a collaborative control framework with multi-timescale rolling optimization. This framework decomposes the complex global optimization problem into three collaborative layers working in the time dimension: the first millisecond-level layer focuses on handling rapid fluctuations and disturbances in signals such as current and voltage, achieving instantaneous stability; the second-level layer, based on updated system models and states, solves a multivariate economic optimization problem within a finite prediction time domain, calculating the optimal setpoint; the third hour-level layer, starting from long-term operating data, optimizes slowly varying parameters or strategy weights related to lifetime decay. These three layers work collaboratively in a rolling, nested manner, ensuring that the control system can respond agilely to rapid dynamics, perform refined energy efficiency optimization, and simultaneously consider long-term equipment health management. This resolves the technical contradiction that traditional single control modes cannot simultaneously balance response speed, operating efficiency, and long-term reliability.

[0044] Furthermore, in calculating the state function of the electrolysis process, the dynamic model of the membrane water content λ(t) considers the capillary pressure-driven two-phase flow transport, and its liquid water transport equation includes a Darcy's law term based on capillary pressure: Among them, v l Let K be the velocity of liquid water, and k be the permeability. rl μ represents relative permeability. l P is the dynamic viscosity of water. l P is the pressure of the liquid water. c It is capillary pressure.

[0045] Specifically, this model abandons traditional macroscopic empirical formulas and innovatively introduces a refined physical description of two-phase flow transport driven by capillary pressure. Its liquid water transport velocity, v_l, is characterized by a modified Darcy's law incorporating capillary pressure gradients, which more realistically reflects the actual transport mechanism of water within the porous catalytic layer and proton exchange membrane microstructure.

[0046] By incorporating capillary pressure, a key microscopic force, into real-time state assessment, this model can more accurately predict the spatiotemporal distribution of moisture within the membrane, especially avoiding the risk of localized "drying out" or "flooding" due to the inaccuracy of traditional models under dynamic load conditions. This fundamentally improves the accuracy of subsequent calculations of key state variables such as the proton conduction efficiency function η_p(t) based on water content λ(t), thus providing a more reliable and physically accurate internal state feedback for the entire multivariable optimization control system. It forms the core model foundation for achieving high-precision, preventative hydrothermal management.

[0047] Furthermore, the collected sensor data includes: Temperature and strain fields are measured using fiber Bragg grating sensors embedded in bipolar plate channels. Charge transfer resistance and double-layer capacitance are obtained by online electrochemical impedance spectroscopy measurement and are used as state variables in the calculation of the state function of the electrolysis process.

[0048] Specifically, two advanced in-situ, online measurement technologies were introduced: one is a fiber Bragg grating (FBG) sensor embedded in the bipolar plate flow channel, used to directly and distributedly measure the temperature field and mechanical strain field of key areas, thereby obtaining the internal physical state that traditional external sensors cannot reach; the other is online electrochemical impedance spectroscopy (EIS) measurement, used to non-invasively obtain key electrochemical state parameters such as charge transfer resistance and double-layer capacitance.

[0049] By fusing these two types of real-time data that directly reflect the core internal state (thermal / mechanical state and electrochemical interface state), the system completely transforms the traditional "black box" model that relies on external macroscopic parameters for inference. This high-dimensional internal state data is directly input into the calculation of the electrolysis process state function, giving the state assessment unprecedented spatial resolution and electrochemical process depth. This significantly improves the perception accuracy, response foresight, and reliability of the entire control system, providing a crucial data foundation for achieving precise preventative control and early fault warning.

[0050] Example 2 Please see Figure 2 This embodiment also provides an intelligent control system for a PEM electrolyzer, used to implement an intelligent control method for a PEM electrolyzer as described in any one of Embodiment 1, the system comprising: The sensor layer comprises multiple sensors arranged in a distributed manner to collect temperature, pressure, current, and voltage data. The state assessment layer is communicatively connected to the sensor layer and is configured to calculate the state function of the electrolysis process based on the collected data. The collaborative control layer, which is communicatively connected to the state evaluation layer, is configured to construct and solve a multivariate optimization objective function and calculate the optimal set of control parameters. The actuator layer, which is communicatively connected to the collaborative control layer, includes an adjustable power supply, a water pump, and valves, and is configured to adjust the operation of the PEM electrolyzer according to the optimal set of control parameters.

[0051] In a specific test embodiment, the hardware system implementation mainly includes: Distributed sensing unit: High-precision temperature sensors (such as PT100), pressure sensors, and flow meters are arranged at the anode and cathode inlets and outlets of the PEM electrolyzer, as well as at key flow channels of the bipolar plates. Specifically, a fiber Bragg grating (FBG) sensor array is embedded within the membrane electrode assembly (MEA) for online monitoring of the temperature distribution and stress state of the MEA. The system electrochemical workstation integrates an online electrochemical impedance spectroscopy (EIS) module for periodically acquiring electrochemical state parameters such as charge transfer resistance (R_ct).

[0052] Multi-core heterogeneous controller: The core processing unit adopts an architecture that combines an ARM Cortex-A series processor with a field-programmable gate array (FPGA). The ARM processor runs a Linux operating system and is responsible for executing complex algorithms such as state evaluation, parameter identification, and multi-timescale optimization; the FPGA is responsible for implementing the underlying fast feedback control law (10ms cycle), digital filtering, and high-speed parallel acquisition of sensor data.

[0053] Intelligent Execution Unit: Power regulation unit: It adopts a DC power supply module based on silicon carbide (SiC) MOSFETs and has a pulse width modulation (PWM) frequency of >10kHz, which can realize millisecond-level precise regulation of electrolytic cell current and optimization of current density distribution.

[0054] Fluid control unit: includes a high-precision metering pump (for deionized water supply) and a piezoelectrically driven high-speed proportional valve (for gas-liquid two-phase flow outlet pressure control), both with a response time of less than 20ms.

[0055] 2. Software Algorithm Implementation Process refer to Figure 2 The software flowchart shown illustrates the following steps that are executed cyclically after the system powers on and initializes: Step S1: Data Acquisition and Synchronization. In each basic control cycle (e.g., 10ms), data from all sensors is synchronously acquired via the FPGA, including: total voltage V. cell (t), total current I(t), temperature T at each measuring point i (t), Import and export pressure P in (t), P out (t), flow rate Q(t), FBG wavelength drift data, and raw EIS spectrum data (when in the measurement period).

[0056] Step S2: State estimation and parameter update Solve the FBG data and separate the temperature field T(x,y,t) and strain field; Online EIS data updates the charge transfer resistance R in real time by fitting an equivalent circuit model.ct and double-layer capacitance C dl ; Call the recursive least squares algorithm with a forgetting factor (λ=0.98) to update the time-varying parameter vector in the state function.

[0057] Step S3: Calculate the electrolysis process state function (EPSF) Based on real-time temperature, current, and pressure data, the membrane water content λ(t) is estimated.

[0058] Calculate the proton conduction efficiency function; The optimal function for catalytic activity distribution was calculated using the distributed temperature and current data. Step S4: Perform multi-timescale rolling optimization (core) Fast layer (executed every 10ms): Based on the current deviation, a feedforward-feedback composite control is used to quickly calculate the current density adjustment Δi to counteract the disturbance. The objective is: min ‖Δx - AΔu‖^2.

[0059] Middle layer (executes every 1 second): Starts the Model Predictive Control (MPC) solver. Using the current state as initial values, it predicts the next 10 seconds (H_p=10). Solves the cost optimization problem; The constraints are discretized heat balance and water transport equations. The optimal control sequence is obtained by solving the equations, and the first element (i.e., the optimal setpoint for the next cycle) is then sent out.

[0060] Slow layer (executed every hour or when the performance metric PI decreases): Evaluate the running data over a period of time and re-optimize the weight coefficients α, β, γ in the objective function of the middle layer MPC to balance efficiency and degradation.

[0061] Step S5: Control Command Execution and Recording. The optimized setpoints are converted into specific commands for each actuator (such as PWM duty cycle and valve opening), and output through the FPGA hardware port. Simultaneously, all key states, control actions, and performance indicators for this cycle are stored in a historical database for long-term trend analysis and model correction.

[0062] Step S6: Fault Diagnosis and Fault Tolerance Handling (Parallel Background Task). Continuously monitor key residual signals, such as r(t) = V_cell_measured - V_cell_model. If J_FD = (1 / N) Σ r^TΣ - If ¹r continuously exceeds the threshold δ_th, a fault is determined to have occurred (such as membrane drying or sensor failure), and the system switches to the preset fault-tolerant control strategy while issuing an alarm.

[0063] 3. Example of initial settings for key parameters Taking a 5kW PEM electrolyzer as an example, some key control parameters are initialized as follows: The parameters of the proton conduction efficiency function are: λ_min=2, λ_max=22, α(0)=0.85.

[0064] Catalytic activity function parameter: β(0) = 0.12; Initial values ​​for MPC weighting coefficients: α=1.0 (efficiency weight), β=0.5 (temperature uniformity weight), γ=0.3 (current distribution weight), δ=10 (terminal weight); Forgetting factor: λ = 0.98; Prediction time domain: H p =10 (corresponds to 10 seconds); Basic control cycle: 10ms.

[0065] 4. Implementation effect verification In a test scenario simulating photovoltaic power fluctuations, an electrolytic cell using the control method of this invention is compared with the traditional PID control method: Dynamic response: When the input power increases by 50%, the system reaches a new steady state in 0.5 seconds (compared to 3 seconds using traditional methods), with no overshoot.

[0066] Efficiency stability: Within a fluctuation range of ±30% of rated power, the system operating efficiency (based on higher heating value) remains within 75% ± 2%.

[0067] State estimation accuracy: Through offline sampling verification, the average error between the online estimated membrane water content λ(t) and the measured value is less than ±0.5.

[0068] Long-term operation: After 1000 hours of accelerated decay test, the voltage rise rate of the electrolytic cell controlled by this method was slowed down by about 40% compared with the traditional control, indicating that the decay was effectively suppressed.

[0069] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A smart control method for PEM electrolyzers, characterized in that, Includes the following steps: Collect multi-physics sensor data during the operation of the PEM electrolyzer, wherein the sensor data includes at least temperature, pressure, current and voltage; Based on the sensor data, the state function of the electrolysis process is calculated. The state function of the electrolysis process includes at least a proton conduction efficiency function for characterizing the real-time proton conduction efficiency of the membrane electrode and a catalytic activity distribution optimization function for characterizing the catalyst utilization rate. Based on the state function of the electrolysis process, a multivariate optimization objective function J is constructed and solved to obtain the optimal set of control parameters; Based on the optimal set of control parameters, control commands are generated and the actuators of the PEM electrolyzer are adjusted. Wherein, the multivariate optimization objective function J is the efficiency maximization objective function, and its expression is: In the formula, For Faraday efficiency, For voltage efficiency, For power density, This is the performance degradation cost function based on changes in system state.

2. The intelligent control method for a PEM electrolyzer according to claim 1, characterized in that, The formula for calculating the proton conduction efficiency function is as follows: Where α is a material property constant, λ(t) is the real-time membrane water content, and λmin and λmax are the minimum and maximum water content of the membrane, respectively.

3. The intelligent control method for a PEM electrolyzer according to claim 1, characterized in that, The formula for calculating the catalytic activity distribution optimization function is as follows: in, For local current density, For rated current density, β represents the local overpotential deviation, and β is the catalyst attenuation coefficient.

4. The intelligent control method for a PEM electrolyzer according to claim 1, characterized in that, The formula for calculating the performance degradation cost function is as follows: Where k1 and k2 are weighting coefficients, and ΔTmax is the maximum temperature difference. This represents the magnitude of the current density gradient.

5. The intelligent control method for a PEM electrolyzer according to claim 1, characterized in that, The optimal control parameter set includes at least the optimal current density and the optimal pressure difference between the anode and cathode; wherein, the formula for calculating the optimal current density is: In the formula, Let γ be the temperature effect function, and γ be the response coefficient. Δη is the proton conductivity threshold, and Δη is the width of the transition region.

6. The intelligent control method for a PEM electrolyzer according to claim 1, characterized in that, It also includes an online parameter identification step, which uses a recursive least squares method with a forgetting factor to estimate and update the time-varying parameter vector θ in the state function of the electrolysis process in real time; The parameter vector θ includes at least the material characteristic constant α in the proton conduction efficiency function and the catalyst attenuation coefficient β in the catalytic activity distribution optimization function.

7. A smart control method for a PEM electrolyzer according to claim 1 or 6, characterized in that, The step of constructing and solving the multivariate optimization objective function J is performed using a multi-timescale rolling optimization architecture, including: In the first millisecond timescale, a linearized model based on the system state is used to solve the fast dynamic compensation problem; On the second-second time scale, based on a nonlinear prediction model, the cost optimization problem is solved within the finite prediction time domain. At the third hour timescale, the lifetime management optimization problem is solved based on the performance degradation model.

8. The intelligent control method for a PEM electrolyzer according to claim 1, characterized in that, When calculating the state function of the electrolysis process, the dynamic model of the membrane water content λ(t) considers the capillary pressure-driven two-phase flow transport, and its liquid water transport equation includes Darcy's law term based on capillary pressure: Among them, v l Let K be the velocity of liquid water, and k be the permeability. rl μ represents relative permeability. l P is the dynamic viscosity of water. l P is the pressure of the liquid water. c It is capillary pressure.

9. The intelligent control method for a PEM electrolyzer according to claim 1, characterized in that, The collected sensor data includes: Temperature and strain fields are measured using fiber Bragg grating sensors embedded in bipolar plate channels. Charge transfer resistance and double-layer capacitance are obtained by online electrochemical impedance spectroscopy measurement and are used as state variables in the calculation of the state function of the electrolysis process.

10. An intelligent control system for a PEM electrolyzer, characterized in that, The system is used to implement the intelligent control method for a PEM electrolyzer according to any one of claims 1-10, the system comprising: The sensor layer comprises multiple sensors arranged in a distributed manner to collect temperature, pressure, current, and voltage data. The state assessment layer is communicatively connected to the sensor layer and is configured to calculate the state function of the electrolysis process based on the collected data. The collaborative control layer, which is communicatively connected to the state evaluation layer, is configured to construct and solve a multivariate optimization objective function and calculate the optimal set of control parameters. The actuator layer, which is communicatively connected to the collaborative control layer, includes an adjustable power supply, a water pump, and valves, and is configured to adjust the operating state of the PEM electrolyzer according to the optimal set of control parameters.

Images