Hydrogen production by electrolysis of water under renewable power fluctuations

Abstract

The application provides a hydrogen production method by electrolysis of water under renewable power fluctuation, comprising: performing health degree evaluation on an electrolytic cell for hydrogen production by electrolysis of water, and classifying the electrolytic cell according to the health degree evaluation result; performing operation mode distribution and power baseline acquisition on the classified electrolytic cell; acquiring predicted power information of renewable power, and performing power distribution on the electrolytic cell through an optimization control method according to the predicted power information of renewable power, the operation mode distribution result and the acquired power baseline to obtain a power control instruction, and controlling the electrolytic cell according to the power control instruction to realize hydrogen production by electrolysis of water, and realize power fluctuation fast tracking, low-load safe and efficient operation and equipment life balance.

Description

Technical Field

[0001] This invention relates to the field of water electrolysis hydrogen production technology, and more specifically, to a method for regulating water electrolysis hydrogen production under renewable power fluctuations. Background Technology

[0002] Utilizing renewable energy sources such as wind and solar power to drive water electrolysis for hydrogen production is a key technological pathway for achieving green hydrogen production. However, unlike traditional grid-powered water electrolysis, renewable energy power generation inherently possesses randomness, intermittency, and severe fluctuations, with its output continuously varying on millisecond to minute timescales. Existing control methods suffer from the following technical problems: First, there is a mismatch between the timescale of the control strategy and the physical response capability of the electrolyzer. The thermal dynamic response of an alkaline electrolyzer is on the order of minutes, and the electrochemical response is on the order of seconds, while renewable electricity exhibits power flicker on the order of milliseconds to seconds. Existing methods directly couple minute-level optimization with millisecond-level tracking, lacking a second-level coordination layer in between. This results in the system being unable to effectively track sudden rise and fall events, leading to the accumulation of power tracking deviations and ultimately causing DC bus voltage instability.

[0003] Second, the overall system efficiency is low and safety hazards exist under low-load conditions. When the alkaline electrolyzer operates below 20% of its rated power, the gas purity safety boundary deteriorates sharply—hydrogen diffuses laterally through the diaphragm to the oxygen side, and the hydrogen concentration in the oxygen may exceed the safety limit of 1.8%, significantly increasing the risk of hydrogen-oxygen crosstalk; at the same time, parasitic current losses increase dramatically, and hydrogen production efficiency drops significantly. Existing methods attempt to improve low-load efficiency through low-frequency pulsed current, but the double-layer capacitance is completely discharged during the pulse turn-off period, and additional energy is consumed to rebuild the double layer during the conduction period. As a result, the actual efficiency of industrial-grade alkaline electrolyzers under low load is actually lower than that of DC electrolysis.

[0004] Third, the lifespan distribution in multi-electrolyte arrays is uneven. Traditional power allocation strategies treat all equipment equally, without considering the objective differences in health status, cumulative operating time, and number of start-stop cycles among the various electrolyzers. This results in some equipment suffering overload losses while others remain idle and wasteful, with the overall lifespan of the array limited by the first single unit to fail.

[0005] In summary, existing methods for regulating hydrogen production through water electrolysis under renewable power fluctuations have significant shortcomings in terms of time-scale matching, safe and efficient operation under low loads, and balanced lifespan of multiple cells. There is an urgent need for a regulation method that can systematically solve the above problems. Summary of the Invention

[0006] In view of this, the present invention proposes a method for regulating hydrogen production by water electrolysis under renewable power fluctuations, in order to solve the problems existing in the prior art.

[0007] To achieve the above objectives, this invention proposes a method for regulating hydrogen production through water electrolysis under renewable power fluctuations, comprising: A health assessment is performed on the electrolyzers used for hydrogen production by water electrolysis, and the electrolyzers are classified according to the health assessment results; then, the classified electrolyzers are assigned operating modes and power baselines are obtained. The system obtains the predicted power information of renewable electricity, and based on the predicted power information of renewable electricity, the operation mode allocation results and the obtained power baseline, it allocates power to the electrolyzer through an optimized control method to obtain power control commands. The system then controls the electrolyzer according to the power control commands to achieve hydrogen production by water electrolysis.

[0008] Optionally, the process of conducting a health assessment of the electrolyzer for hydrogen production by water electrolysis includes: Obtain the operating parameters of the electrolyzer, wherein the operating parameters include ohmic impedance, charge transfer resistance, diffusion impedance, cumulative operating time, and cumulative number of start-stop cycles; The health assessment result is calculated using a mapping function based on the operating parameters; wherein the mapping function is a weighted sum function or a neural network.

[0009] Optionally, the process of assigning operating modes includes: The electrolytic cells are classified into healthy cells, medium-quality cells, and aging cells based on the health assessment results. The electrolytic cells are classified and assigned operating modes. The healthy cell, medium cell, and aging cell all operate in DC mode and their rated power range decreases sequentially. The rated power ranges of the healthy cell, medium cell, and aging cell are inclusive of each other.

[0010] Optionally, the power baseline acquisition process includes: Predict the power required to obtain renewable electricity, select the corresponding number of electrolyzers based on the predicted power, and mark the corresponding electrolyzers based on the number of electrolyzers. Power baseline allocation and adjacent time period switching constraints are applied to the predicted power of the marked electrolyzers to obtain the enabled electrolyzers and power baselines.

[0011] Optionally, the process of power distribution in the electrolyzer includes: The system acquires the operating status information of the electrolyzer, pre-allocates the predicted power information to the electrolyzer based on the power baseline, and generates a power reference value. Using the power reference value as the initial variable, the system iteratively optimizes the initial variable using an optimization control method to obtain the optimal power control sequence for each electrolyzer, and generates a power control command based on the optimal power control sequence. The optimization control method mentioned above employs the MPC method.

[0012] Optionally, the iterative optimization process for solving the initial variables includes: Based on the power reference value, the corresponding state parameters are calculated using the electrochemical reaction dynamics, thermal dynamics model, and gas purity model of the electrolytic cell. It is then determined whether the initial variables and state parameters meet the constraints. If the constraints are met, the corresponding cost function is calculated. The cost function is a weighted sum of power tracking deviation, temperature tracking deviation, cumulative cell start-up and shutdown counts, cumulative running time deviation, and hard-switching penalty term. Under constraints, the power reference value is adjusted, the cost function is recalculated based on the adjusted power value, and the power value corresponding to the minimum cost function is selected and retained. The process of adjusting the power value, calculating the state parameters, judging the constraints, and calculating and selecting the cost function is repeated to obtain the optimal power control sequence.

[0013] Optional constraints include: power balance constraints, power upper and lower limit constraints, ramp rate constraints, temperature constraints, gas purity constraints, start-stop state logic constraints, and shutdown priority constraints.

[0014] Optionally, during the process of controlling the electrolytic cell according to the power control command, an energy storage unit is configured; the energy storage unit is used to make up for the power shortage of the electrolytic cell.

[0015] Optionally, during the process of controlling the electrolytic cell according to the power control command, different extreme operating conditions are obtained, and the electrolytic cell is controlled according to the extreme operating conditions.

[0016] On the other hand, the present invention provides a water electrolysis hydrogen production control system under renewable power fluctuations for performing the above-described method.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: First, the control timescale is matched with the physical response capability of the electrolyzer. This invention uses a three-layer control architecture to separate hourly scheduling, second-level model predictive control, and millisecond-level current closed-loop execution. The timescale of each layer is aligned with the thermal inertia, electrochemical response, and power electronic switching of the electrolyzer, respectively. This avoids the gap problem in the prior art where there is no second-level coordination layer between minute-level optimization and millisecond-level tracking, and significantly improves the response capability to sudden rise and fall events.

[0018] Secondly, the system's safety and overall efficiency are improved through DC following and hard-switching strategies. This invention abandons the low-frequency pulse efficiency enhancement scheme, which is engineering-infeasible in existing technologies, and all online electrolyzers operate in DC mode. When the total power is insufficient, a hard-switching strategy is implemented—concentrating limited power to a few electrolyzers, allowing them to operate within the 40%~80% high-efficiency range. Low-load cells are completely shut down instead of operating inefficiently, avoiding the safety risk of hydrogen-oxygen cross-contamination when operating below 20% load, thus significantly improving the overall system efficiency.

[0019] Third, controlled discharge suppresses shutdown corrosion. Before complete shutdown, this invention actively performs voltage clamped controlled discharge, dissipating energy through the internal resistance of the electrolytic cell and the external discharge resistor, neutralizing the residual potential of the electrodes, and fundamentally suppressing reverse current corrosion caused by the galvanic cell effect. This reduces irreversible catalyst layer shedding and membrane damage, and extends electrode life.

[0020] Fourth, array lifespan management is achieved through load balancing. This invention prioritizes each slot based on its cumulative operating time, number of start-stop cycles, and health status, prioritizing the slot with the least cumulative damage to bear the load, thus avoiding excessive wear and tear on a single slot. The start-stop optimization logic reduces unnecessary start-stop cycles, and long-term rotating scheduling (weekly to monthly) balances aging without introducing additional thermal cycling damage, extending the overall lifespan of the system.

[0021] Fifth, external hot standby ensures rapid response and operational safety. This invention uses external auxiliary heating (electric heating belt or hot water circulation) to maintain the standby tank temperature. During hot standby, the electrolytic cell does not supply power or produce gas, fundamentally eliminating the safety hazard of hydrogen-oxygen cross-contamination in traditional electrochemical thermal standby mode. Restarting requires only 10-15 minutes of nitrogen purging and seconds-level ramp-up, far shorter than the 30-60 minutes of cold start, balancing rapid response and inherent safety.

[0022] Sixth, by coordinating energy storage to smooth out second-level fluctuations, the tracking pressure on the electrolyzer is reduced. This invention incorporates energy storage units into the system architecture, with the energy storage handling the high-frequency power components below the second level, while the electrolyzer only tracks the smoothed minute-level power commands. This reduces the ramp-up burden on the electrolyzer and the configuration requirements for energy storage capacity, achieving an economical match between the power supply side and the load side.

[0023] Seventh, intrinsic safety is achieved through online monitoring of gas purity. This invention monitors the oxygen content in hydrogen and the hydrogen content in oxygen in real time under all operating conditions (including startup, operation, shutdown, and hot standby). If the limits are exceeded, nitrogen purging and shutdown protection are triggered, ensuring that the system always operates within the safety boundary and eliminating safety accidents caused by deterioration of gas purity. Attached Figure Description

[0024] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a schematic diagram of the method flow in an embodiment of the present invention. Detailed Implementation

[0025] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0026] Renewable electricity (wind power, photovoltaic) is characterized by randomness, intermittency, and volatility, with its output varying drastically on timescales from milliseconds to minutes. When alkaline electrolyzers operate below 20% of their rated power, the gas purity safety margin deteriorates, the risk of hydrogen-oxygen crosstalk increases, and parasitic current losses surge. Industrial-grade alkaline electrolyzers exhibit a second-level time lag in responding to power changes, making it impossible to quickly track power fluctuations on the order of minutes. Furthermore, frequent start-stop cycles cause reverse current corrosion: during shutdown, galvanic cells form on the electrodes, leading to cathode oxidation and anode reduction; the damage is irreversible and accumulates with each start-stop cycle. In multi-electrolyte arrays, traditional power allocation strategies treat all equipment equally, resulting in underutilization of healthy cells and accelerated wear of aging cells, limiting the overall array lifespan to the earliest decommissioned equipment.

[0027] The core concept of the electrolytic water hydrogen production regulation method proposed in this invention under renewable power fluctuations is to upgrade the regulation from the existing single power amplitude allocation to a three-layer collaborative control with time scale matching, return to the mainstream engineering approach of DC following and hard switching, balance fast response and intrinsic safety through external hot standby, suppress shutdown corrosion through controlled discharge, and achieve array lifetime management through load balancing.

[0028] The method for regulating hydrogen production through water electrolysis under renewable power fluctuations proposed in this invention relies on a water electrolysis hydrogen production system with the following structure: The power generation layer consists of wind or solar power units, which are combined and output to the DC bus to provide power for the entire hydrogen production system. Energy storage units (batteries or supercapacitors) are configured on the DC bus side to absorb power fluctuations on the order of seconds.

[0029] The converter layer equips each group of electrolyzers with an independent programmable power electronic converter, employing a multiphase interleaved parallel Buck topology, with a total output current reaching several thousand amperes and an output current ripple of less than 1%. The electrolyzer layer consists of an array of N alkaline electrolyzers, each equipped with a temperature sensor, a voltage sensor, a current sensor, and an online analyzer for the oxygen content in hydrogen and the hydrogen content in oxygen.

[0030] The sensing layer is responsible for collecting state data from each electrolyzer, including real-time voltage, current, temperature, gas purity, and electrochemical parameters obtained by fitting ohmic resistance and polarization curves based on microsecond-level current pulse measurements.

[0031] The decision-making layer adopts a three-layer controller architecture: the top layer performs hourly scheduling with a period of 1 to 4 hours to formulate start-up and shutdown plans and power baselines; the middle layer performs model predictive control with a period of 1 second to solve the optimal power allocation sequence for the next 30 to 60 seconds; and the bottom layer performs current closed-loop control with a period of 10 to 100 milliseconds.

[0032] The present invention provides a method for regulating hydrogen production through water electrolysis under renewable power fluctuations, such as... Figure 1 As shown, it includes the following steps: Step 1: State Perception and Multidimensional Health Assessment Temperature sensors are installed at the electrolyte inlet, outlet, anode and cathode sides in the middle of the tank, near the diaphragm, and bottom of each electrolyzer to collect the average tank temperature and temperature gradient. Current and voltage sensors are installed at the power input of each electrolyzer. Hydrogen oxygen content analyzers and oxygen hydrogen content analyzers are installed on the gas output pipeline of each electrolyzer, with a range of 0–5%, accuracy of ±0.05%, and response time ≤5 seconds, ensuring sufficient measurement accuracy and response speed near the alarm thresholds (HTO ≤ 1.5%, OTH ≤ 2.0%).

[0033] The health assessment employs a multi-dimensional fusion approach: (1) Ohmic resistance measurement: Apply a current pulse lasting less than 100 microseconds, sample the instantaneous voltage change ΔU within the window before activation polarization is established, and calculate the true ohmic resistance using ΔU / ΔI. This measurement is performed during the initial stage of constant current charging after the electrolyzer has been shut down and purged, to avoid interference from gas evolution bubbles. This measurement is performed weekly.

[0034] (2) Polarization curve fitting: During the start-up or periodic maintenance of the electrolytic cell, the steady-state voltage and current data of at least 5 power points are scanned within the range of 30% to 100% of the rated power. The ohmic impedance, activation overpotential coefficient and concentration polarization coefficient are separated by nonlinear fitting to establish a complete set of electrochemical characteristic parameters.

[0035] (3) Electrochemical impedance spectroscopy diagnosis: Perform electrochemical impedance spectroscopy scanning, and estimate the charge transfer resistance by equivalent circuit fitting. (Corresponding to the mid-frequency region, reflecting catalyst activity) and diffusion impedance (Corresponding to the low-frequency region, reflecting the membrane mass transfer capability), jointly evaluating the catalyst activity state and membrane permeability. The low-frequency scan (0.01Hz~1Hz) is performed when the electrolyzer is in a stable operating state, and the current fluctuation is kept less than ±2% of the rated current during the measurement to ensure the validity of the low-frequency data.

[0036] (4) Operation history record: The cumulative operating time and cumulative number of start-ups and shutdowns of each electrolytic cell are recorded through the operation log.

[0037] (5) Multi-dimensional SOH fusion: f represents the mapping function, which uses a weighted sum function or other neural network or large model processing methods. When using a weighted sum function, the above parameters are normalized and then weighted sum is calculated. The weights of each dimension are calibrated through historical data, and the comprehensive score range is 0~100%.

[0038] Step Two: Electrolytic Cell Classification and Priority Ranking: Based on the multi-dimensional health score and operational history from step one, all electrolytic cells are comprehensively scored and sorted from highest to lowest score. The top 30% of electrolytic cells are classified as healthy cells, given priority for load and fluctuation tracking; the middle 40% are classified as medium-weight cells, operating under base load; and the bottom 30% are classified as aging cells, activated only when most necessary and only under stable low loads. An absolute scrapping threshold is also set: if an electrolytic cell's comprehensive score falls below 40%, regardless of its ranking, it is directly marked as scrapped and must not be put into operation, requiring maintenance or replacement. The activation priority and load type for each category are recorded in the database and updated during each top-level scheduling.

[0039] Step 3: Operation mode allocation and power baseline determination: Based on the classification results from step two, different operating modes are assigned to different types of electrolyzers: The health tank undertakes the main power tracking task, is assigned a DC operating mode, and allows it to be adjusted within the range of 30% to 100% of the rated power.

[0040] The medium-sized tank operates under base load and is allocated DC operating mode, allowing it to be adjusted within the range of 40% to 90% of rated power.

[0041] The aging tank is only activated when most necessary, and is assigned a DC operation mode, allowing it to operate stably only within 50% to 80% of its rated power range, without being responsible for fluctuation tracking.

[0042] The standby tank is equipped with an external hot standby mode. The electrolytic cell does not receive power or produce gas. The temperature of the cell is maintained at 50~60°C by an external electric heating belt or hot water circulation system.

[0043] All online electrolytic cells operate in DC mode with an output current ripple of less than 1%. No pulse waveforms are assigned; the differences between cells lie in the power allocation ratio and start / stop priority.

[0044] After classification, top-level hourly scheduling (update cycle: 1~4 hours) is carried out. Based on the renewable energy power forecast for the next 4~24 hours, the top-level system formulates the start-up and shutdown plan and power baseline of the electrolyzers in each hourly segment, which serves as the reference trajectory for the intermediate layer MPC.

[0045] (1) Start-up and shutdown plan formulation: For each hourly period, firstly, in order of priority (shortest cumulative running time → fewest start-up and shutdown times → highest health), the average or fixed percentage value of the stable operating power of each cell is accumulated until the accumulated value is not lower than the predicted total power of that hourly period. The number of cells accumulated is the minimum number of electrolytic cells that need to be operated during that period. Then, the online cells are selected from all electrolytic cells in the following order of priority: the cells with the shortest cumulative running time are given priority; if the running times are similar, the cells with the fewest cumulative start-up and shutdown times are given priority; if they are still similar, the cells with the highest health are given priority.

[0046] The selected electrolytic cells are marked as online cells for that hour period, and the remaining electrolytic cells are marked as shut down or externally available for standby.

[0047] (2) Power baseline allocation: For the electrolytic cells marked as online within the hourly period, the predicted total power is allocated to the power baseline value of each cell according to the rated power ratio of each cell. After allocation, each cell is checked to see if it is within its allowable operating range (set according to the classification results: the range is widest for healthy cells and narrowest for aging cells). If it exceeds the limit, power transfer correction is performed between the online cells in the hourly period to ensure that the baseline value of each cell falls within the allowable range.

[0048] (3) Adjacent time period switching constraints: Between adjacent hourly periods, if a cell switches from operation to shutdown, its continuous operating time must not be less than a set value (e.g., 2 hours); if it switches from shutdown to operation, its continuous shutdown time must not be less than a set value (e.g., 1 hour). Switches that do not meet the constraints are automatically blocked by the system and postponed to the next highest-scoring electrolytic cell for replacement. This is to obtain the start-up and shutdown status of the electrolytic cells.

[0049] (4) Instruction issuance: The start-up and shutdown status and power baseline values ​​of different electrolyzers in each hour are packaged and issued to the intermediate layer MPC as the reference boundary for its second-level rolling optimization. Based on this, MPC makes corrections according to ultra-short-term predictions.

[0050] Step 4: Multi-target power allocation: The prediction model of MPC uses a state-space description of an alkaline electrolyzer, which includes two coupled subsystems: electrochemical reaction dynamics and thermal dynamics.

[0051] The equation of state for the electrochemical subsystem is:

[0052] in Slot voltage, It is a reversible voltage. To activate the overpotential, For Ohmic polarization (i is the current density, (for ohm resistance) This is concentration overpotential.

[0053] The thermal dynamics subsystem adopts a lumped-parameter thermal model:

[0054] in T represents the heat capacity of the electrolytic cell, and T represents the temperature of the cell. For electrochemical heat generation, To reduce heat loss to the environment, The heat removed by the cooling system.

[0055] Gas purity is determined using an empirical transfer function model:

[0056] Where HTO is the oxygen content in hydrogen gas, and K is the gain coefficient. The time constant (on the order of minutes) represents the residence time of the gas in the gas-liquid separator. It is the operating current of the electrolytic cell. The Laplace transform form of , where s is the complex frequency.

[0057] The electrochemical response (milliseconds to seconds) and the thermal dynamic response (minutes) are described uniformly within the MPC framework, but the prediction time domain is set to 30–60 seconds—only covering the initial stages of rapid electrochemical response and thermal dynamic changes, to avoid control failure due to long-term integral errors in the thermal model. The prediction window is selectively truncated based on the actual physical time constant of the alkaline electrolyzer.

[0058] The optimization variable for MPC is the power of each online slot. (i is the slot number, k is the step index in the prediction time domain). The constraints are: (1) Power balance: ,in This is an ultra-short-term power prediction sequence. The ultra-short-term power prediction sequence is obtained by predicting time-series data using methods such as LSTM. The output is the power and related operating data at the current and previous fixed time points, and the output is the power prediction results for the ultra-short-term (30-60s) period.

[0059] (2) Power upper and lower limits: The lower and upper power limits of the health tank are respectively =30%× , =100%× The medium-sized tank is =40%× , =90%× The aging tank is =50%× , =80%× The above lower limits are all not lower than the minimum safe power (20% ×) corresponding to the gas purity safety boundary. When the rated power of each slot in the array is different, the absolute values ​​of the upper and lower power limits of each slot are determined according to their respective... The scaling factor is adjusted proportionally. Furthermore, the lower limit is increased based on the aging degree of each tank to prevent aging tanks from experiencing accelerated performance degradation when operating in low-load areas, while simultaneously providing healthy tanks with a wider adjustment range for fluctuation tracking. This is the rated power.

[0060] (3) Slope rate: ,in , ; (4) Temperature: ;Calculated using a lumped-parameter thermal model; (5) Gas purity: (OTH represents the hydrogen content in oxygen). Meanwhile, because the response time for gas purity (on the order of minutes) is longer than the MPC prediction time (30-60 seconds), the MPC does not directly predict the absolute value of HTO within the prediction window. Instead, the MPC calculates the rate of change of HTO at the current operating point and, combined with the current measured HTO value, predicts whether HTO will approach the alarm threshold (1.5%) within the next 30 seconds. If dHTO / dt > 0 and the current HTO > 1.2%, the MPC automatically and temporarily raises the lower power limit of the tank to 30%P. rated,i (Healthy tank) or higher (medium tank, aging tank) to avoid long-term low-load operation leading to purity deterioration. The OTH change trend adopts the same transfer function form as HTO (only the gain coefficient K is determined separately through field calibration). MPC synchronously calculates dOTH / dt. If dOTH / dt>0 and the current OTH>1.6%, a temporary increase in the lower limit of the tank power is also triggered.

[0061] (6) Start / Stop Status Logic: And it meets the minimum running / downtime constraint.

[0062] (7) Shutdown priority constraint: The shutdown decision variable ∈{0,1}( =0 indicates shutdown. =1 indicates execution) Apply sorting constraints by category:

[0063] in , , This is a set of aging tanks, intermediate tanks, and healthy tanks. This constraint ensures that when MPC needs to be shut down, the aging tanks are shut down first, followed by the intermediate tanks, and finally the healthy tanks.

[0064] The cost function of MPC adopts the following multi-objective weighted form:

[0065] in The number of time-domain steps to predict is 30-60. This represents the current number of online electrolytic cells; the first item is the power tracking deviation. The dynamic temperature setpoint varies with power. Changes: Lower temperatures are used at low power (to reduce heat loss), and higher temperatures are used at high power (to increase the reaction rate), determined in real time by referring to a table based on the electrolytic cell efficiency-temperature curve; The cumulative number of start-stop cycles for the i-th slot; The cumulative running time of the i-th slot, This is the average value across all slots. For hard-switch penalty items, among which The lower limit of the high-efficiency operating range (health tank, medium tank and aging tank are 30%, 40% and 50% of the rated power, respectively). is the penalty coefficient, where ∈ ( ),in, This penalty applies an incremental penalty to power commands below the efficiency range, forcing the MPC to proactively concentrate power on a few slots where feasible, rather than allowing all slots to operate in the inefficient range.

[0066] Weighting coefficient satisfy =1, the initial value is determined by the analytic hierarchy process (AHP), and is adaptively adjusted every 24 hours based on actual operating statistics. Specific adjustment rules: Statistically calculate the normalized actual values ​​of various indicators for the day: root mean square error of power tracking σP (rated power percentage), temperature overshoot σT (°C), total number of start-stop cycles Ctotal, and operating time variance σRT (hours). 2 If any indicator exceeds the preset threshold (σP>3%, σT>2℃, Ctotal>5 times / day, σRT>100 hours), 2 If ∑w = 1, then the corresponding weight increases by 10%, and other weights are reduced proportionally to maintain ∑w = 1. The upper and lower limits of the weights are constrained to wi ∈ [0.05, 0.5], to avoid excessively increasing a certain weight and causing other objectives to be completely ignored.

[0067] Using relaxation variables to handle the hard constraints of temperature and gas purity:

[0068] , This indicates the corresponding excess amount, which is added to the cost function. M is the penalty coefficient, which is added after the cost function. + This ensures that a feasible solution exists under extreme operating conditions.

[0069] The coordination mechanism between hard-switching logic and MPC: when the predicted total power is lower than the current number of online slots. At this time, MPC is not directly shut down, but rather through the aforementioned hard-switch penalty item. The increasing effect of the penalty term worsens the evaluation of the cost function for the "run all slots at low load" option, while giving a better evaluation to the "shut down some slots and run the remaining slots at high load" option. This guides the optimization algorithm to automatically select shutdown decisions. The shutdown order is determined by the health ranking (aging slots are prioritized). By transforming discrete start-up and shutdown decisions into continuous penalty terms in the MPC cost function, the complexity explosion problem caused by mixed integer programming is avoided, while achieving the proactive hard shutdown capability that conventional MPC does not possess.

[0070] The solution employs a sequential quadratic programming algorithm with a control period of 1 second. The rolling time-domain process is as follows: state update → prediction update → optimization solution → instruction issuance → rolling advancement. For large arrays with N>20, a distributed MPC is used—grouping by health status, with the upper-level coordinator responsible for power allocation between groups and the sub-MPC responsible for fine allocation within groups, in order to reduce computational complexity.

[0071] In this rolling solution process, the state update corresponds to the sensor measured data in step one and the real-time power feedback in step six, that is, reading the current power, temperature, voltage, current and current measured total power of each tank. This serves as the initial state for MPC optimization calculations. The prediction update corresponds to step four, where, based on the hourly power baseline value at the top level, the power allocated to the electrolyzers is fine-tuned (e.g., adjusted according to the different electrolyzer ratios based on the power baseline value), based on the electrochemical / thermal dynamic state-space model and the ultra-short-term power prediction sequence. Generate a power reference trajectory for the next Np steps; optimization is the core computational process of multi-objective power allocation in step four. Based on the power reference trajectory as the initial variable, and under all constraints such as power balance, power upper and lower limits, ramp rate, temperature, gas purity, and start / stop state logic, iteratively solve the cost function. The minimum power control sequence for each slot is obtained, and the iteration ends when the maximum number of iterations is reached or convergence is achieved. The instruction is issued to the underlying execution stage in step six, where the first element of the solved control sequence is... The control is then sent to each converter for execution; the continuous loop mechanism corresponding to step six is ​​rolled forward, and when the next control cycle arrives, the predicted time domain is rolled forward one step, repeating the above process to achieve closed-loop continuous optimization.

[0072] Step 5: Rotation scheduling and energy storage coordination: At each preset rotation period (weekly or monthly, dynamically adjusted based on the cumulative operating time and number of start-stop cycles for each cell), the roles of the primary operating cell and the standby cell are swapped. The specific selection logic is as follows: the cell with the shortest cumulative operating time, the fewest start-stop cycles, and the highest health level is given priority for the primary operating role; the cell with the longest cumulative operating time, the most start-stop cycles, and the lowest health level is prioritized for maintenance or standby roles. This rotation period is much longer than the thermal time constant of the electrolytic cell, thus avoiding the introduction of additional thermal cycling damage and not involving waveform type switching.

[0073] Energy storage collaborative control logic: Energy storage units (batteries or supercapacitors) are configured on the DC bus side, and their power commands... Determined by the following rules:

[0074] in Providing real-time power for renewable energy, This represents the actual total power of the electrolytic cell array. This refers to the power that is actively abandoned.

[0075] The boundaries of the role of energy storage: (1) Second-level fluctuation smoothing (<1 second) is undertaken by supercapacitors, with a response time of milliseconds, absorbing high-frequency power components; (2) Minute-level fluctuation smoothing (1 second to 1 minute) is undertaken by battery energy storage, smoothing the mid-frequency power components; (3) Power surge buffering: when P_RE surges to more than 30% of the rated total power within 10 seconds, energy storage will first absorb the part of power that exceeds the climbing capacity of the electrolyzer, and release it after the electrolyzer gradually climbs up; (4) Power drop supplementation: when P_RE drops sharply, energy storage will discharge for a short time to supplement the power gap, and buy time for MPC replanning and electrolyzer state switching (30 to 60 seconds of MPC prediction window).

[0076] The state of charge (SOC) of energy storage needs to be maintained within a reasonable range of 40% to 80%. When the SOC approaches the upper limit, the MPC appropriately increases the electrolyzer power or starts the standby cell; when the SOC approaches the lower limit, the MPC appropriately reduces the electrolyzer power or schedules a shutdown. The SOC reference trajectory is incorporated into the optimization as an additional constraint for the MPC.

[0077] All decision commands (DC current command values ​​for each converter) are sent to the local controller of each converter via industrial Ethernet with a latency of less than 1 millisecond.

[0078] Step Six: Real-time Tracking and Dynamic Correction This step is executed continuously, with a control cycle of 100 milliseconds to 1 second (consistent with the MPC update cycle). Within each control cycle (the current time is denoted as k), the current real-time total power is measured. Calculation and prediction of the value in step four (in step four) deviation . This represents the predicted total power at time k, calculated at time k-1.

[0079] MPC re-solves the optimization problem based on the latest measured values ​​in each control cycle, automatically incorporating deviations into the power allocation for the next cycle. Instead of setting a fixed dead zone, it applies soft constraints to deviations through the MPC cost function—automatically smoothing out small deviations and prioritizing responses to larger deviations.

[0080] Power correction does not use a fixed allocation ratio. Instead, the MPC allocates the power surge to the healthy tank with the largest remaining ramp capacity based on the current remaining ramp capacity, temperature margin, and health status of each tank, while satisfying the global ramp rate constraint. The power drop is preferentially allocated to the aging tank or medium-sized tank.

[0081] For all online electrolyzers, power correction is achieved by fine-tuning the DC current command value. The correction amount is calculated using a linearization method based on the slope of the polarization curve. ,in The slope of the polarization curve at the current operating point is calculated online using the electrochemical equation of state. If the power correction is large (exceeding 10% of the current power), an iterative approach is used: first estimate... Substitute into the electrochemical equation of state to calculate the new The system checks whether the actual power change meets the requirements. If not, it continues to fine-tune until the error is less than 1% of the rated power. The maximum number of iterations is set to 10. If convergence is not achieved within 10 iterations, the most recent estimated value is taken as the output. The remaining deviation for this control cycle is temporarily supplemented by the energy storage unit and corrected through MPC feedback in the next control cycle.

[0082] Before outputting the corrected current command, the rate of change of current in each electrolytic cell is limited to ensure it does not exceed the maximum permissible rate of change. (5% of rated current / second). If the rate of change of current exceeds this value, the current command is limited: new current value. ,in This represents the actual current value, and Δt is the control period. This limiting prevents sudden current surges from causing a large number of bubbles to form instantaneously, which could lead to electrolyte splashing and a sudden drop in liquid level.

[0083] The updated DC current command value is written into the PWM register of each converter, driving the IGBT or MOSFET switching devices to output the corresponding DC current, and then entering the next control cycle.

[0084] Abnormal Status Feedback and Top-Level Rescheduling Mechanism: When an online tank is forced to shut down or reduce power due to a fault, excessive temperature, or gas purity alarm, resulting in the current total rated power of the online tank being unable to cover the real-time total power. At that time, the bottom layer immediately reports the abnormal slot status to the intermediate MPC layer. The MPC detects in the next control cycle that the constraint cannot be met ( This immediately triggers a top-level rescheduling request, forcibly opening the highest-scoring standby slot (selected according to the scoring order in step two), and updating the online slot list and power baseline for that hourly period. The rescheduling response time is in the minute range (≤5 minutes), during which the power gap is made up by the energy storage units. If the energy storage SOC is insufficient to support the system for 5 minutes, active power curtailment is initiated until the standby slot is fully operational.

[0085] Step 7: Active Protection under Extreme Operating Conditions When the total power is below the minimum safe power of a single electrolytic cell (20% of the rated power) and there is no indication of recovery within the next 10 minutes, if the predicted power recovery is likely within 10 minutes to 1.5 hours, switch to external hot standby mode to maintain the cell temperature at 50~60°C. Within this time window, the energy consumption of external hot standby is less than the energy consumption of reheating after a cold shutdown. If the predicted recovery time exceeds 1.5 hours or the recovery time is uncertain, execute the complete shutdown procedure. (1) First, with a change rate not exceeding the maximum rate of change. The current is gradually reduced to the minimum safe power value at a certain rate and maintained for 10 seconds to purge the gas from the tank.

[0086] (2) Perform controlled discharge: switch the converter output to voltage clamping mode, and dissipate energy in a controlled manner through the internal resistance of the electrolytic cell and the external discharge resistor to neutralize the residual potential of the electrode for 5~10ms.

[0087] (3) Immediately after controlled discharge, close the inlet and outlet valves of the electrolytic cell, isolate the system, and cut off the main circuit.

[0088] (4) Record this shutdown in the database and increment the cumulative number of shutdowns and restarts by 1.

[0089] If the predicted power recovery time is 1-4 hours, switch to external hot standby mode instead of a complete cold shutdown. The precise calculation method for external hot standby power is as follows: Hot standby power The total heat loss from the tank to the environment is determined by three parts: convection heat dissipation, radiation heat dissipation, and pipe heat dissipation.

[0090] Convection cooling: ,in The convective heat transfer coefficient (related to wind speed and tank surface shape). The outer surface area of ​​the tank. To maintain the temperature (typical value 55℃). The ambient temperature.

[0091] Radiative heat dissipation Where ε is the surface emissivity of the tank (typical value 0.85), and σ is the Stefan-Boltzmann constant (5.67 × 10⁻⁶). -8 W / (m 2 ·K 4 Due to the uneven temperature distribution on the surface of the tank (the temperature at the top is higher than at the bottom), the calculation... The weighted average of multiple surface temperature measurement points is used (the weights are determined based on the area represented by each measurement point), rather than a uniform tank temperature. The surface emissivity ε is calibrated every six months using an infrared thermal imager to correct for changes caused by surface oxidation or dirt accumulation.

[0092] Pipe cooling: Calculations were performed separately for each alkali circulation pipeline and gas pipeline. Let be the thermal resistance of the i-th segment of the pipeline.

[0093] The above parameters are calibrated based on on-site measured heat balance data when the system is put into operation, and are corrected in real time according to ambient temperature and wind speed during operation. Under typical values, It is approximately 1% to 3% of the rated power. The outer wall of the tank is wrapped with a thermal insulation layer with a thermal conductivity ≤0.04 W / (m·K) and a thickness ≥50mm, in order to minimize the energy consumption for thermal standby.

[0094] During hot standby, the electrolytic cell does not supply power or produce gas, and there is no gas safety risk.

[0095] When power is restored, the electrolytic cell in hot standby mode will be quickly restarted in a controlled manner: (1) First, purge the system with nitrogen for 10-15 minutes to replace the gas in the system and ensure that the oxygen content in the hydrogen and the hydrogen content in the oxygen are normal.

[0096] (2) Since the tank temperature has been maintained, start directly with a DC current of 10%~20% of the rated power, and increase the power to the target value within 30 seconds at a rate not exceeding R_max.

[0097] (3) Report the slot number that has been put back into operation to the upper controller.

[0098] Handling Abnormal Operating Conditions: Temperature exceeding limits: If the average temperature of the tank exceeds 90℃, or the temperature gradient is greater than 10℃, or the temperature change rate exceeds 5℃ / minute, immediately reduce the average power of the tank to 70% of the current value and increase the coolant flow rate. If the temperature does not decrease after 3 minutes, execute the complete shutdown procedure in this step.

[0099] Power drop: When the current total power drops by more than 30% of the rated total power within 10 seconds and the predicted value continues to decline, MPC immediately re-solves the shutdown sequence, selects the tank to be shut down in the order of aging tank → medium tank → healthy tank, executes controlled discharge shutdown protection, and returns to step four to redistribute global power. The aging tank is allowed to unload to the minimum safe power at a rate of 5% rated power / second, and immediately executes the controlled discharge shutdown process in step seven after reaching the minimum safe power, no longer subject to the 2% / second normal ramp rate constraint, to ensure that the shutdown priority constraint (7) can be executed quickly in emergency conditions.

[0100] Power Surge: When the current total power increases by more than 30% of the rated total power within 10 seconds, the instantaneous impact is first absorbed by the energy storage unit on the DC bus side. The maximum power absorbed by the energy storage is 80% of its rated power; any excess is handled by active power curtailment. Simultaneously, a timer is activated: If the energy storage SOC exceeds 75% or the absorption time exceeds 10 seconds, the standby cell startup process is immediately triggered—prioritizing the electrolytic cell with the highest temperature in the hot standby state, executing the hot standby startup process (nitrogen purging + ramp-up), without waiting for the MPC to resolve. After the standby cell starts, the energy storage gradually withdraws, transferring power to the electrolytic cell. During startup, if the energy storage SOC exceeds 80% (the upper limit of the normal operating range), the MPC immediately limits the energy storage charging power to 50% of the rated value; if the SOC exceeds 90%, active power curtailment is forcibly initiated.

[0101] Gas purity exceeds limits: In any operating mode, if the oxygen content in hydrogen exceeds 1.5% or the hydrogen content in oxygen exceeds 2.0%, the system will immediately and automatically cut off the power supply to all electrolyzers, open the nitrogen purging valve to purge at the rated flow rate, and simultaneously alarm the upper-level controller. The system can only be restarted after the purity returns to normal.

[0102] This solution constitutes a continuous closed-loop control cycle: the top-level hourly scheduling formulates start-up and shutdown plans, the middle-level second-level MPC executes power optimization allocation, the bottom-level millisecond-level current closed loop achieves precise tracking, and the energy storage unit smooths out sub-second fluctuations. Each layer's time scale is matched and does not interfere with the others. The core logic of this solution is to abandon the physically infeasible low-frequency pulse modulation strategy and return to the mainstream engineering approach of DC following and hard switching. Power tracking is achieved through three-layer control with time-scale matching; lifespan management is achieved through load balancing and long-term rotation; external hot standby ensures both rapid response and intrinsic safety; online gas purity monitoring ensures safety boundaries; and controlled discharge suppresses shutdown corrosion. Based on physical principles, this solution achieves an achievable balance between engineering feasibility and performance improvement.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for regulating hydrogen production through water electrolysis under renewable power fluctuations, characterized in that, include: A health assessment is conducted on the electrolyzers used for hydrogen production by water electrolysis, and the electrolyzers are classified according to the health assessment results; The classified electrolytic cells are assigned operating modes and power baselines are obtained. The system obtains the predicted power information of renewable electricity, and based on the predicted power information of renewable electricity, the operation mode allocation results and the obtained power baseline, it allocates power to the electrolyzer through an optimized control method to obtain power control commands. The system then controls the electrolyzer according to the power control commands to achieve hydrogen production by water electrolysis.

2. The method according to claim 1, characterized in that, The process of conducting a health assessment of the electrolyzer for hydrogen production by water electrolysis includes: Obtain the operating parameters of the electrolyzer, wherein the operating parameters include ohmic impedance, charge transfer resistance, diffusion impedance, cumulative operating time, and cumulative number of start-stop cycles; The health assessment result is calculated using a mapping function based on the operating parameters; wherein the mapping function is a weighted sum function or a neural network.

3. The method according to claim 1, characterized in that, The process of allocating operating modes includes: The electrolytic cells are classified into healthy cells, medium-quality cells, and aging cells based on the health assessment results. The electrolytic cells are classified and assigned operating modes. The healthy cell, medium cell, and aging cell all operate in DC mode and their rated power range decreases sequentially. The rated power ranges of the healthy cell, medium cell, and aging cell are inclusive of each other.

4. The method according to claim 1, characterized in that, The process of obtaining the power baseline includes: Predict the power required to obtain renewable electricity, select the corresponding number of electrolyzers based on the predicted power, and mark the corresponding electrolyzers based on the number of electrolyzers. Power baseline allocation and adjacent time period switching constraints are applied to the predicted power of the marked electrolyzers to obtain the enabled electrolyzers and power baselines.

5. The method according to claim 1, characterized in that, The process of distributing power in an electrolytic cell includes: The system acquires the operating status information of the electrolyzer, pre-allocates the predicted power information to the electrolyzer based on the power baseline, and generates a power reference value. Using the power reference value as the initial variable, the system iteratively optimizes the initial variable using an optimization control method to obtain the optimal power control sequence for each electrolyzer, and generates a power control command based on the optimal power control sequence. The optimization control method mentioned above employs the MPC method.

6. The method according to claim 1, characterized in that, The process of iteratively optimizing the initial variables includes: Based on the power reference value, the corresponding state parameters are calculated using the electrochemical reaction dynamics, thermal dynamics model, and gas purity model of the electrolytic cell. It is then determined whether the initial variables and state parameters meet the constraints. If the constraints are met, the corresponding cost function is calculated. The cost function is a weighted sum of power tracking deviation, temperature tracking deviation, cumulative cell start-up and shutdown counts, cumulative running time deviation, and hard-switching penalty term. Under constraints, the power reference value is adjusted, the cost function is recalculated based on the adjusted power value, and the power value corresponding to the minimum cost function is selected and retained. The process of adjusting the power value, calculating the state parameters, judging the constraints, and calculating and selecting the cost function is repeated to obtain the optimal power control sequence.

7. The method according to claim 1, characterized in that, The constraints include: power balance constraints, power upper and lower limit constraints, ramp rate constraints, temperature constraints, gas purity constraints, start-stop state logic constraints, and shutdown priority constraints.

8. The method according to claim 1, characterized in that, During the process of controlling the electrolytic cell according to the power control command, an energy storage unit is configured; the energy storage unit is used to make up for the power shortage of the electrolytic cell.

9. The method according to claim 1, characterized in that, During the process of controlling the electrolytic cell according to the power control command, different extreme operating conditions are obtained, and the electrolytic cell is controlled according to the extreme operating conditions.

10. A water electrolysis hydrogen production control system under renewable power fluctuations, characterized in that, Used to perform the method described in any one of claims 1-9.

Images