Steady-state electrolytic cell power adjusting system and method based on controllable modality

Abstract

The invention relates to the technical field of electrolytic bath power regulation, in particular to a steady-state electrolytic bath power regulation system and method based on a controllable modality, and the method comprises the steps: obtaining the input current and voltage of an electrolytic bath and the state information of a capacitor bank, and carrying out vectorization processing to obtain a current vector, a voltage vector and a switching state vector; determining an optimal capacitor switch combination based on the current vector and the switch state vector by using a pre-constructed deep learning model and taking the situation that the system power is maintained in a preset threshold range as a target; and according to the obtained optimal capacitor switch combination, driving a capacitor group switch to act, and adjusting an equivalent capacitance value to realize power dynamic balance. The capacitor bank comprises a plurality of capacitor units which are connected in parallel, and each capacitor unit adjusts an equivalent capacitance value of an access system through on-off of a corresponding switch; the storage battery is used for energy storage compensation; the inverter is used for conducting bidirectional conversion between the power grid alternating current and the electrolytic bath direct current.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrolytic tank power regulation, in particular to a steady-state electrolytic tank power regulation system and method based on controllable modal. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] In a green hydrogen production electrolytic tank system, the electrolytic tank produces hydrogen by electrolyzing water, and the power stability directly affects the purity of the product gas and the service life of the equipment. In order to ensure the normal operation and efficient operation of the electrolytic tank, the power needs to be stabilized in a reasonable range. In the prior art, when the electrolytic tank is connected to wind power / photovoltaic power, due to the existence of the output power of wind power / photovoltaic power, there is a sharp fluctuation of the output power in seconds to minutes, which causes the input current of the electrolytic tank to have a high-frequency ripple. At the same time, the impedance (resistance) inside the electrolytic tank is not constant, and the resistance of the electrode material changes due to the change of the working temperature of the electrolytic tank, the change of the ion mobility of the electrolyte, etc., and there is an impedance mismatch problem.

[0004] Current fluctuations can cause uneven electrode reaction rates, which can trigger reverse reactions (such as hydrogen-oxygen recombination) or impurity gases, causing a decrease in gas purity. High-frequency current ripples can accelerate electrode corrosion (especially for noble metal catalysts in PEM electrolytic tanks) and membrane material aging, shortening the service life of the electrolytic tank.

[0005] In view of the above problems, the existing regulation system can ensure the relative stability of the voltage of the electrolytic tank, but lacks effective means to deal with the power instability problem caused by current fluctuations. The dynamic response of the electrolytic tank needs to be controlled in seconds or even milliseconds, and manual operation cannot cope with the rapid fluctuations of wind and solar power generation. Moreover, the traditional method based on a fixed parameter control algorithm cannot adapt to the dynamic changes of the impedance of the electrolytic tank. SUMMARY

[0006] In order to solve the technical problems existing in the background art, the present application provides a steady-state electrolytic tank power regulation system and method based on controllable modal, which realizes voltage modal regulation by constructing a reconfigurable capacitor network, and realizes millisecond-level dynamic response by combining a deep reinforcement learning algorithm.

[0007] In order to achieve the above purpose, the present application adopts the following technical solutions: The first aspect of the present application provides a steady-state electrolytic tank power regulation system based on controllable modal, comprising: a current monitoring module for acquiring an input current signal of the electrolytic tank and sending it to a control unit; a voltage monitoring module for acquiring an input voltage signal of the electrolytic tank and sending it to the control unit; An external power supply group includes a capacitor bank, a battery, and an inverter. The capacitor bank includes multiple capacitor units connected in parallel, and each capacitor unit adjusts the equivalent capacitance value of the connected system by turning on and off the corresponding switch. The battery is used for energy storage compensation. The inverter is used to perform bidirectional conversion between the AC power of the grid and the DC power of the electrolyzer. The control unit receives the current and voltage values ​​of the electrolytic cell and converts them into corresponding current vectors and voltage vectors. It also obtains the switch state code of the capacitor group and converts it into a switch state vector. Using a pre-built deep learning model, it determines the optimal capacitor switch combination based on the current vector and switch state vector, with the goal of maintaining the system power within a preset threshold range. This control instruction is then generated and sent to the execution module. The execution module drives the switch action of the capacitor group according to the received optimal capacitor switch combination, adjusts the equivalent capacitance value, and realizes dynamic power balance.

[0008] Furthermore, the control unit includes a data acquisition and conversion module and an algorithm model module. The data acquisition and conversion module obtains the current value, voltage value and switch state coding of the electrolytic cell and performs vector processing.

[0009] Furthermore, the algorithm model module uses a pre-built deep learning model to maintain the system power within a preset threshold range, obtains the optimal capacitor switch combination based on the current vector, voltage vector and switch state vector, and sends it to the execution module.

[0010] Furthermore, the deep learning model includes an input layer, a hidden layer, and an output layer; The input layer is used to obtain the current vector and switch state vector; The hidden layer extracts features and performs nonlinear transformation on the acquired current vector and switch state vector, mining the complex relationship between the current value, the closed state of the capacitor bank switch, and the system power. The output layer determines the optimal switching state combination of the capacitor bank with the goal of maintaining the system power within a preset threshold range.

[0011] Furthermore, the control instructions include an optimal capacitor switch combination and a battery state switching instruction, and the battery state switching instruction includes "switch to discharge state", "switch to storage state" and "maintain original state".

[0012] Furthermore, the deep learning model receives the input current vector I, voltage vector V, capacitor bank state vector S through the state input layer. c and its historical window data .

[0013] Further, the deep learning model estimates the value function of a given state through a value function network, and the reward function and target function are used for training. After training, the optimal capacitor switch combination and battery state switching instruction at the current time are output through the policy network.

[0014] Further, the capacitor group includes multiple capacitor units connected in parallel. Different capacitor units are combined to form the capacitor group through the gradient difference of the capacity.

[0015] Further, the execution module includes a microcontroller and a drive circuit. The microcontroller receives the optimal switch combination and uses the drive circuit to provide a drive signal for the switch of the capacitor group to control the switch to be closed or opened.

[0016] The second aspect of the present application provides a controllable mode-based steady-state electrolytic cell power regulation method, including the following steps: Obtain the current signal, voltage signal and state information of the capacitor group of the electrolytic cell, and obtain the current vector, voltage vector and switch state vector through vectorization processing; Through a pre-constructed deep learning model, the system power is maintained within a preset threshold range as the target, and the optimal capacitor switch combination is determined based on the current vector and the switch state vector; According to the obtained optimal capacitor switch combination, the switch action of the capacitor group is driven to adjust the equivalent capacitance value, and dynamic power balance is achieved.

[0017] Compared with the prior art, the above one or more technical solutions have the following beneficial effects: 1. A hybrid buffer unit is formed by using a capacitor-battery-inverter, and the inverter is used to regulate the AC side power input to achieve long-term power balance in the grid interaction mode. The battery is used as an energy pool to deal with minute-level fluctuations, and power balance is achieved in the battery mode. By controlling the connection of different capacitor units to the system, high-frequency ripple and millisecond-level fluctuations are suppressed, and power balance is achieved in the capacitor-dominated mode to deal with the current fluctuations caused by the connection of new energy to the electrolytic cell.

[0018] 2. For the problem of impedance mismatch, the deep learning algorithm model analyzes the correlation between historical current data and capacitor switch state through a hidden layer, and real-time predicts the impedance change trend of the electrolytic cell. The "optimal switch combination" generated by the output layer not only considers the current state, but also optimizes the long-term regulation sequence through reinforcement learning to form a multi-modal regulation strategy, avoiding frequent switch actions.

[0019] 3. Through the inverter, alternating current (such as grid supplement) is converted into direct current that matches the impedance of the electrolytic cell, realizing seamless switching of multiple energy sources. It can also output phase-synchronous direct current to ensure that the release rates of the two polar charges are consistent, preventing local impedance mutations caused by charge accumulation, and achieving electrode charge balance control. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated herein by reference. The illustrations are of exemplary embodiments of the application and explain the principles of the application and are not meant to limit the application.

[0021] Figure 1 is a schematic diagram of a controllable mode-based steady-state electrolytic cell power regulation system structure provided by one or more embodiments of the application; Figure 2 is a schematic diagram of the working principle of a controllable mode-based steady-state electrolytic cell power regulation system provided by one or more embodiments of the application; Figure 3 is a schematic diagram of the controllable mode-based steady-state electrolytic cell power regulation process provided by one or more embodiments of the application. DETAILED DESCRIPTION

[0022] The application will be further described below with reference to the drawings and embodiments.

[0023] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the application. Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the application belongs.

[0024] Embodiment one: The controllable mode-based steady-state electrolytic cell power regulation system converts the originally fluctuating power supply into a controllable variable by introducing a capacitor bank controlled by a controllable port, effectively addressing the uncontrollable changes in current due to wind and light fluctuations and the factors of conductive materials, thereby stabilizing the power of the electrolytic cell.

[0025] The use of deep learning methods for vectorization and parameter fine-tuning of current values and capacitor bank on-off states can more accurately and quickly regulate power compared to traditional manual operation and simple control algorithms, improving the automation level and regulation efficiency of the system.

[0026] The automatic mechanism can adjust the power in real time according to the changes in current, greatly reducing the workload and errors of manual intervention, improving the stability and reliability of the electrolytic cell operation, and being conducive to improving production efficiency and product quality.

[0027] As shown in Figure 1 To achieve multi-modal power regulation function, the system intelligently manages the charging and discharging process of the battery. The system includes a voltage control module, a current monitoring module, an external power supply bank, a data acquisition and conversion module, an algorithm model module, and a control execution module.

[0028] Voltage control module: used for real-time monitoring of the voltage value of the electrolytic cell.

[0029] Current monitoring module: used to monitor the current value of the electrolytic cell in real time.

[0030] External power supply group: composed of capacitor group, battery, and inverter.

[0031] The control logic in the system includes two parts: one is automatic control based on auxiliary judgment logic, and the other is joint optimization output fused into the deep learning model.

[0032] On the one hand, the control unit is provided with a current threshold judgment mechanism for detecting whether the current input to the electrolytic cell exceeds the set threshold interval. When the current exceeds the upper threshold value (indicating insufficient system power), the battery is triggered to discharge; when the current is lower than the lower threshold value (indicating energy surplus), the battery is triggered to charge; if the current is within the interval, the current state of the battery remains unchanged.

[0033] On the other hand, the charging and discharging state of the battery is one of the control outputs of the deep learning model, that is, the model simultaneously outputs a battery control instruction value while generating the optimal capacitor switch combination, for example: "1" represents discharging; "-1" represents charging; "0" represents maintaining the original state.

[0034] The control instruction is sent to the inverter and battery interface through the control execution module to realize closed-loop control of the direct current output or charging path. The discharge power and duration of the battery can be dynamically adjusted according to the model prediction results, thereby realizing the cooperative work with the capacitor regulation mechanism.

[0035] This mechanism combines the fast ripple suppression dominated by capacitors and the medium-frequency energy balance dominated by batteries to realize cross-scale regulation from seconds to milliseconds, effectively offsetting the current anomalies caused by new energy fluctuations and enhancing the robustness and response speed of the system.

[0036] The battery stores electricity, and when discharging, it realizes dynamic and sustained discharging and adjusts the voltage with the help of the capacitor group. AC power can be converted to DC current through the inverter, ensuring that both electrodes always release the same amount of charge. The capacitor group is composed of multiple capacitors and is provided with multiple controllable ports, each corresponding to the control of the connection or disconnection state of one or more capacitors. By controlling the connection state of the capacitors in the capacitor group, the equivalent capacitance value of the system can be changed.

[0037] Data acquisition and conversion module: connected with the current monitoring module, voltage monitoring module, and capacitor group, used to acquire current value, voltage value, and the on-off state of each controllable port of the capacitor group, and perform vectorization processing on the acquired data.

[0038] Algorithm model module: The algorithm model module is pre-trained with an algorithm model based on deep learning. It is used to receive vectorized data transmitted from the data acquisition and conversion module, fine-tune parameters within the model, and calculate the optimal switch closure state combination of the capacitor group based on the set power stability range threshold to regulate the system power.

[0039] Control execution module: used to receive the optimal switch closing state combination instruction output by the algorithm model module, and control the controllable port of the capacitor group to perform the switch closing or opening operation according to the instruction, thereby realizing the adjustment of the electrolyzer power.

[0040] As a further embodiment, the deep learning algorithm model of the algorithm model module includes an input layer, a hidden layer, and an output layer; The input layer is used to receive the current value vector and the capacitor group switch closing state vector transmitted by the data acquisition and conversion module; The hidden layer extracts features and performs nonlinear transformation on the input data, mining the complex relationship between the current value, the closed state of the capacitor bank switch, and the system power; The optimal switch closing state combination of the output capacitor group of the output layer enables the system power to be maintained at a threshold within the stable range.

[0041] As a further embodiment, the current monitoring module uses a high-precision current sensor to ensure that the current value of the electrolytic cell can be accurately and real-time monitored.

[0042] As a further implementation method, the control execution module includes a microcontroller and a drive circuit. The microcontroller is used to receive instructions from the algorithm model module and perform logical processing. The drive circuit is used to provide sufficient drive signals to the controllable ports of the capacitor group to control the switch to close or open according to the processing results of the microcontroller.

[0043] The system works as follows: The voltage control module is used to collect the voltage signal at the input end of the electrolyzer and send it to the control unit; The current monitoring module is used to collect the input current signal of the electrolytic cell and send it to the control unit; The control unit uses the data acquisition and conversion module to convert the received current value into a current vector, the voltage signal into a voltage vector, and the switch closure state of each controllable port of the capacitor group into a capacitor group switch state vector. All vectors are input into a pre-built deep learning model to generate the optimal capacitor switch combination, so that the system power is maintained within the preset threshold range.

[0044] The algorithm model module is a deep learning model. Based on the deep learning algorithm, according to the pre-trained model and the received vectorized data, it fine-tunes the parameters internally and calculates the optimal switch closure state combination of the capacitor bank that can maintain the system power within the stable range threshold.

[0045] To achieve optimal scheduling of capacitor bank switch combinations and battery charging and discharging, the algorithm model module constructs a deep reinforcement learning model based on the Actor-Critic structure to process multimodal input information and generate combined control strategy outputs.

[0046] The model structure is: State input layer: receives input current vector I, voltage vector V, capacitor bank state vector S c and its historical window data ; Policy network (Actor): Three layers of fully connected (Dense) + ReLU, followed by Softmax + Gumbel-Softmax to output the optimal capacitor switch combination at the current moment; Value Function Network (Critic): A three-layer fully connected network used to estimate the value function of a given state ; Battery Branch: Actor Bypass Output Single Node Stands for "Discharge / Hold / Charge".

[0047] ; Where k is the historical step length. All features are normalized and then input into the network.

[0048] Reward function and optimization objective: Instant Rewards:

[0049] The first term penalizes power deviation, and the second term suppresses frequent switching. is the coefficient; Critic loss: ; Actor Loss: .

[0050] Training process: Digital twin environment: Build a joint simulation platform for electrolyzers and external power packs; Offline pre-training: Gradient updates are used until the rewards converge; Online fine-tuning: After deployment, the network is incrementally updated using real-time operating data to maintain adaptability to operating condition drift.

[0051] Control execution: Model inference output per cycle (1–5ms) , The control execution module will Translate into switch drive signal to realize capacitor combination switching; at the same time The data is sent to the inverter interface to control battery charging and discharging, thereby achieving cross-time domain power balance at the millisecond level.

[0052] In this embodiment, the process of obtaining the optimal switch closing state combination based on the current, voltage and state vector by the deep learning model is as follows, including the feature fusion stage, dynamic relationship modeling and optimization decision output.

[0053] Feature fusion stage: The original current / voltage values ​​are extracted into 32-dimensional features through the fully connected layer, the switch state vector is converted into 16-dimensional features through the embedding layer, and the historical data is generated into 20-dimensional time series features through the LSTM network. The above features are spliced ​​together to form a 128-dimensional comprehensive feature vector.

[0054] Dynamic Relationship Modeling: Capacitor control strategy generation: The feature vector passes through a three-layer neural network (64→32→8 neurons). Each layer uses the ReLU activation function to introduce nonlinearity. The 8 nodes in the output layer use the Sigmoid function. An output value of 0.3 indicates a 30% probability of closing the capacitor. Battery decision generation: A parallel branching network (64→16→3 neurons) outputs three numerical values ​​corresponding to the propensity scores of [charge / hold / discharge], which are converted to a probability distribution (e.g., [0.1, 0.7, 0.2]) through Softmax.

[0055] Optimize decision output: Capacitor switching decision: Threshold judgment is performed on eight probability values ​​(action is taken if > 0.5), and a switch action penalty mechanism is added. Switching is only executed when the profit exceeds the historical average by 10%; Battery command generation: Select the command with the highest probability ("Hold" in the above example) and combine it with the current SOC status. If the battery level is less than 20%, force a charge.

[0056] Design dynamic adjustment mechanisms, including short-term responses and medium- and long-term adjustments.

[0057] Short-term response (millisecond level): When a sudden current change >5% is detected, small-capacity capacitors are switched first (response speed <2ms) and the action threshold is dynamically adjusted according to the dI / dt value.

[0058] Medium- and long-term regulation (seconds): When the current is high for 10 seconds, gradually increase the number of large-capacity capacitors to trigger the battery discharge mode and adjust the inverter output phase.

[0059] The control execution module controls the controllable ports of the capacitor group to perform corresponding on-off or off operation according to the optimal on-off state combination instruction calculated by the algorithm model module, so as to adjust the power of the system, offset the influence of current fluctuation, and stabilize the power in a reasonable range.

[0060] The system has a current-temperature-resistance multi-parameter sensing unit, a reconfigurable capacitor array and its intelligent switch control circuit, a multi-modal data fusion controller based on deep reinforcement learning, and a time-space collaborative execution mechanism.

[0061] The capacitor group has a plurality of capacitors forming a capacitor array, and each capacitor corresponds to a separate charge control switch and a discharge control switch. The program issues control information according to the threshold value, for example, the charge control switch is closed at the same time as the discharge control switch is opened, to ensure that the corresponding capacitor is in a charging state; conversely, the charge control switch is opened at the same time as the discharge control switch is closed, to ensure that the corresponding capacitor is in a discharging state.

[0062] In this embodiment, the capacitor group adopts a modular parallel structure, and each capacitor unit (such as aluminum electrolytic, thin film or super capacitor) has an independent electronic switch interface (such as MOSFET, IGBT) to achieve high-frequency fast switching. It is recommended to select an industrial-grade capacitor with a unit capacity of 1-10 mF, and the voltage level is selected according to the system voltage configuration (such as 450V, 630V level).

[0063] To improve the ability to suppress ripples of different frequency bands, the capacitor group can be configured with "large capacitor + small capacitor": Large capacitors (>1 mF) are used to filter low-frequency large-amplitude fluctuations; Small capacitors (<100 µF) are used to filter high-frequency rapid disturbances (>1 kHz); Physically arranged in parallel zones, each zone can be independently connected to the main bus.

[0064] The battery part is preferably a high-rate lithium iron phosphate (LFP) or super capacitor module, which supports 0.5-1 C rate charging and discharging, and is connected to the main power link through a bidirectional DC-AC inverter. The inverter has a voltage synchronization function and can seamlessly switch between different power modes.

[0065] This structure supports a dual-mode absorption mechanism of millisecond-level capacitor switching and second-level battery regulation, realizing cross-spectrum power stability control.

[0066] The topology reconstruction of the capacitor array is based on real-time optimization of load impedance, and a genetic algorithm is used to realize the optimal configuration of the capacitor combination.

[0067] The deep learning controller includes a Transformer encoder and a graph neural network for modeling the coupling effect between modules.

[0068] Wherein, the actuator adopts EtherCAT bus to realize μs-level synchronous control, and the switch driving circuit has the rapid rising edge characteristic of dV / dt≥50kV / μs.

[0069] Wherein, the feature fusion algorithm contains frequency domain analysis (1-4kHz bandwidth) and time domain differential processing, and generates a 1024-dimensional feature vector.

[0070] The scheme designs a capacitor-battery hybrid buffer system, which absorbs the second-level power fluctuation of wind power / photovoltaic through multi-port controllable capacitor group fast switching equivalent capacitance value (millisecond response). For example: when the current increases suddenly, increase the access capacitance value to suppress the voltage peak; when the current decreases suddenly, reduce the capacitance value to maintain the discharge persistence, and realize dynamic capacitance tuning. Through the cooperation of battery-inverter, the battery acts as an energy pool to cope with minute-level fluctuations, and the inverter converts alternating current (such as grid supplement) into direct current matching the impedance of the electrolytic cell, realizing seamless switching of multiple energy sources.

[0071] The parallel combination of multiple capacitors can form a filter network of different frequency bands (such as large-capacitance filtering low frequency and small-capacitance filtering high frequency), which can specifically eliminate current ripple. Experimental data show that reasonable configuration of the capacitor group can reduce the ripple coefficient from ±15% to below ±2%, realizing high-frequency ripple suppression.

[0072] In view of the problem of impedance mismatch, the deep learning algorithm model analyzes the relevance of historical current data and capacitor switch state through hidden layers (such as LSTM network), and predicts the impedance change trend of the electrolytic cell in real time. For example: when it is detected that the temperature rise causes the resistance to decrease, the amount of capacitor access is automatically reduced to maintain constant power. The "optimal switch combination" generated by the output layer not only considers the current state, but also optimizes the long-term adjustment sequence through reinforcement learning (such as DQN), forming a multi-modal adjustment strategy to avoid frequent switching actions.

[0073] By outputting phase-synchronous direct current through the inverter, it ensures that the charge release rates of the two poles are consistent (error <0.5%), preventing local impedance mutations caused by charge accumulation, and realizing electrode charge balance control.

[0074] In combination with Figure 2 The system working principle shown in the figure further introduces the scheme, including system initialization, real-time monitoring and data acquisition, data vectorization and transmission, algorithm model processing, power regulation operation, and continuous monitoring and feedback regulation.

[0075] System initialization: Before the electrolytic cell starts, the entire controllable mode-based steady-state power regulation system is initialized. The power threshold of the deep learning algorithm model in the algorithm model module is set within the stable interval, the initial on-off state of the capacitor bank is set (for example, all off or part of the capacitors are connected according to the preset initial state), and the current monitoring module is calibrated to ensure accurate monitoring of the current value, ensuring that the microcontroller and drive circuit in the control execution module can communicate normally and receive instructions.

[0076] Real-time monitoring and data acquisition: After the electrolytic cell starts, the current monitoring module begins to monitor the current value of the electrolytic cell in real time, and the data acquisition and conversion module simultaneously acquires the on-off state of each controllable port of the capacitor bank. For example, the current value and capacitor bank state data are collected once every certain time interval (such as 10 milliseconds).

[0077] Data vectorization and transmission: The data acquisition and conversion module converts the collected current value into a vector form, such as converting the real-time monitored current value I into a vector [I_x, I_y, I_z] according to certain quantization rules (the quantization rules can be determined according to actual system requirements and data processing algorithms), and also converts the on-off state of the capacitor bank into a vector form (such as [0, 1, 0, 1] indicating that the 2nd and 4th capacitors are connected and the 1st and 3rd capacitors are disconnected), and then transmits the two vectors to the algorithm model module.

[0078] Algorithm model processing: After receiving the vectorized data, the algorithm model module performs feature extraction and nonlinear transformation on the input data through a pre-set neural network algorithm (such as a multi-layer perceptron algorithm) in the hidden layer, analyzes the relationship between the current value and the on-off state of the capacitor bank and the system power. According to the pre-trained deep learning algorithm model, the parameters are fine-tuned internally, and the optimal on-off state combination that can maintain the system power within the stable interval threshold is calculated. For example, the optimal on-off state combination of the capacitor bank at this time is calculated as [1, 0, 1, 0] (indicating that the 1st and 3rd capacitors are connected and the 2nd and 4th capacitors are disconnected).

[0079] Power regulation operation: After receiving the optimal on-off state combination instruction output by the algorithm model module, the microcontroller in the control execution module performs logical processing, and then provides sufficient driving signals for the controllable ports of the capacitor bank through the drive circuit, adjusts the on-off state of the capacitor bank to [1, 0, 1, 0], changes the connection state of the capacitor bank, and adjusts the power of the system to make it stable within a reasonable interval.

[0080] Continuous monitoring and feedback regulation: During the entire operation of the electrolytic cell, the current monitoring module continuously monitors the current value, the data acquisition and conversion module repeatedly acquires and transmits data, the algorithm model module continuously adjusts the parameters, and the control execution module continuously controls the capacitor group according to the adjusted instructions, forming a closed-loop control system to ensure that the power of the electrolytic cell is always stable in a reasonable range. Even if the current fluctuates due to wind and light fluctuations and the factors of conductive materials, it can quickly and effectively adjust.

[0081] The scheme uses a hybrid buffer system formed by capacitors and batteries to achieve millisecond-level response by quickly switching equivalent capacitance values through a multi-port controllable capacitor group to absorb second-level power fluctuations of wind power and photovoltaic power.

[0082] Through the cooperation of the battery-inverter, the battery is used as an energy pool to deal with minute-level fluctuations, and the inverter converts alternating current (such as grid power) into direct current that matches the impedance of the electrolytic cell, realizing seamless switching of multiple energy sources.

[0083] The parallel combination of multiple capacitors can form a filter network of different frequency bands to specifically eliminate current ripple and achieve high-frequency ripple suppression.

[0084] To solve the problem of impedance mismatch, a deep learning algorithm model analyzes the correlation between historical current data and capacitor switch states through hidden layers (such as LSTM networks) to predict the trend of electrolytic cell impedance changes in real time. The "optimal switch combination" generated by the output layer not only considers the current state but also optimizes long-term adjustment sequences through reinforcement learning, forming a multi-modal adjustment strategy to avoid frequent switching actions.

[0085] By outputting phase-synchronous direct current through the inverter, the release rate of the two-pole charge is ensured to be consistent, preventing local impedance mutations caused by charge accumulation and achieving electrode charge balance control.

[0086] Embodiment two: The steady-state electrolytic cell power regulation method based on controllable modalities includes the following steps: Obtain the input current signal of the electrolytic cell and the state information of the capacitor group, and obtain the current vector and switch state vector through vectorization processing; Through a pre-constructed deep learning model, the system power is maintained within a preset threshold range, and the optimal capacitor switch combination is determined based on the current vector and switch state vector; According to the obtained optimal switch combination, drive the capacitor group switch action, adjust the equivalent capacitance value, and realize power dynamic balance.

[0087] This embodiment solves the problem of power instability caused by current fluctuations during the operation of the electrolytic cell through the system of embodiment one, combining Figure 3The control logic of the present scheme is introduced, including system initialization, real-time monitoring and data acquisition, data vectorization and transmission, algorithm model processing, power regulation operation, and continuous monitoring and feedback regulation.

[0088] During the operation of the electrolytic cell, factors such as ion migration rate, electrode reaction rate, and bulk mass transfer process will cause the ion concentration (C) and pH value in the electrolyte to fluctuate under different operating conditions. C _ion The present invention builds a lightweight internal state prediction module for the electrolytic cell by sensor feedback, equivalent admittance model fitting, and historical operation data, to estimate the trend of pH value change and ion concentration gradient under the current operating point.

[0089] The above-mentioned internal state variables are not directly used as the output target of the controller, but are introduced into the controller structure through the construction of a "Physical Constraint Layer" to filter and correct the control results.

[0090] This layer can take one of the following two forms: As part of the deep neural network, it is embedded in the output layer of the Actor network, and the feasibility reconstruction is performed before the output action; Or, as part of the reward function, behaviors that violate the physical safety threshold are punished in the form of negative feedback: ; Where, is the pH fluctuation in the current period; is the ion concentration fluctuation; λ 2 ,λ 3is the physical stability constraint penalty factor.

[0091] Through the above-mentioned internal state perception mechanism and physical constraint embedding method, the present invention realizes the joint regulation transformation from "observable signal control" to "mechanism prediction constraint", effectively improving the operating stability and electrochemical safety margin of the system under extreme disturbance conditions.

[0092] 1. System initialization: Set the power threshold of the deep learning algorithm model in the algorithm model module; Set the initial on-off state of the capacitor bank; Calibrate the current monitoring module to ensure accurate monitoring of current values; Ensure that the microcontroller and drive circuit in the control execution module can communicate normally and receive instructions.

[0093] 2. Real-time monitoring and data acquisition: The current monitoring module monitors the current value of the electrolytic cell in real time; The data acquisition and conversion module simultaneously obtains the on-off state of each controllable port of the capacitor bank.

[0094] 3. Data vectorization and transmission: The collected current value and capacitor bank on-off state are converted into vector form; The two vectors are transmitted together to the algorithm model module.

[0095] 4. Algorithm model processing: The algorithm model module performs feature extraction and nonlinear transformation on the input data through a preset neural network algorithm in the hidden layer; According to the pre-trained deep learning algorithm model, the parameter fine-tuning is performed internally, and the optimal on-off state combination that can maintain the system power within the stable interval is calculated.

[0096] 5. Power regulation operation: After the control execution module receives the optimal on-off state combination instruction, logical processing is performed; The driving circuit provides a driving signal for the controllable port of the capacitor bank, adjusts the on-off state of the capacitor bank, and thus adjusts the power of the system.

[0097] 6. Continuous monitoring and feedback regulation: The current monitoring module continuously monitors the current value; The data acquisition and conversion module continuously repeats the data acquisition and transmission operation; The algorithm model module continuously performs parameter fine-tuning.

[0098] Through the battery storage, dynamic continuous discharge is realized by the capacitor during discharge and the voltage is adjusted. The power fluctuates, but the AC power can be converted to DC current through the transformer to ensure that the two electrodes always release the same charge. The control execution module continuously controls the capacitor bank according to the adjusted instruction to form a closed-loop control system.

[0099] The above is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A steady-state electrolyzer power regulation system based on controllable modes, characterized in that: include: The current monitoring module is used to collect the input current signal of the electrolytic cell and send it to the control unit; The voltage monitoring module is used to collect the input voltage signal of the electrolytic cell and send it to the control unit; An external power supply group includes a capacitor bank, a battery, and an inverter. The capacitor bank includes multiple capacitor units connected in parallel, and each capacitor unit adjusts the equivalent capacitance value of the connected system by turning on and off the corresponding switch. The battery is used for energy storage compensation. The inverter is used to perform bidirectional conversion between the AC power of the grid and the DC power of the electrolyzer. The control unit receives the current and voltage values ​​of the electrolytic cell and converts them into corresponding current vectors and voltage vectors. It also obtains the switch state code of the capacitor group and converts it into a switch state vector. Using a pre-built deep learning model, it determines the optimal capacitor switch combination based on the current vector and switch state vector, with the goal of maintaining the system power within a preset threshold range. This control instruction is then generated and sent to the execution module. The execution module drives the switch action of the capacitor group according to the received optimal capacitor switch combination, adjusts the equivalent capacitance value, and realizes dynamic power balance.

2. The controllable mode-based steady-state electrolyzer power regulation system according to claim 1, characterized in that: The control unit includes a data acquisition and conversion module and an algorithm model module. The data acquisition and conversion module obtains the current value, voltage value and switch state coding of the electrolytic cell and performs vector processing.

3. The controllable mode-based steady-state electrolyzer power regulation system according to claim 1, characterized in that: The algorithm model module uses a pre-built deep learning model to maintain the system power within a preset threshold range, obtains the optimal capacitor switch combination based on the current vector, voltage vector and switch state vector, and sends it to the execution module.

4. The controllable mode-based steady-state electrolyzer power regulation system according to claim 3, characterized in that: The deep learning model includes an input layer, a hidden layer and an output layer; The input layer is used to obtain the current vector and switch state vector; The hidden layer extracts features and performs nonlinear transformation on the acquired current vector and switch state vector, mining the complex relationship between the current value, the closed state of the capacitor bank switch, and the system power. The output layer determines the optimal switching state combination of the capacitor bank with the goal of maintaining the system power within a preset threshold range.

5. The controllable mode-based steady-state electrolyzer power regulation system according to claim 3, characterized in that: The deep learning model estimates the value function of a given state through a value function network, and implements training of the reward function and the objective function. After training, the policy network outputs the optimal capacitor switch combination and battery state switching instructions at the current moment.

6. The controllable mode-based steady-state electrolyzer power regulation system according to claim 1, characterized in that: The control instructions include an optimal capacitor switch combination and a battery state switching instruction, and the battery state switching instruction includes "switch to discharge state", "switch to storage state" and "maintain original state".

7. The controllable mode-based steady-state electrolyzer power regulation system according to claim 1, characterized in that: The deep learning model receives input current vector I, voltage vector V, capacitor bank state vector S through the state input layer. c and its historical window data .

8. The controllable mode-based steady-state electrolyzer power regulation system according to claim 1, characterized in that: The capacitor group includes a plurality of capacitor units connected in parallel, and the capacitor group is formed by combining different capacitor units with gradient differences in capacitance.

9. The controllable mode-based steady-state electrolyzer power regulation system according to claim 1, characterized in that: The execution module includes a microcontroller and a driving circuit. The microcontroller receives the optimal switch combination and uses the driving circuit to provide a driving signal to the switch of the capacitor group to control the switch to be closed or opened.

10. A method for implementing electrolytic cell power regulation based on the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Obtain the current signal, voltage signal and state information of the capacitor bank of the electrolytic cell, and obtain the current vector, voltage vector and switch state vector through vector quantization processing; Using a pre-built deep learning model, the optimal capacitor-switch combination is determined based on the current vector and switch state vector, with the goal of maintaining system power within a preset threshold. According to the obtained optimal capacitor switch combination, the capacitor group switch action is driven to adjust the equivalent capacitance value to achieve dynamic power balance.

Images