Multi-scene self-adaptive power electronic hydrogen production power supply system and control method
By developing a multi-scenario adaptive power electronic hydrogen production power supply system and control method, the problem of insufficient adaptability of existing hydrogen production power supply systems to multiple operating conditions has been solved, achieving efficient, safe, and stable operation of the electrolyzer and improving electrolysis efficiency and safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TONGJI UNIV
- Filing Date
- 2026-01-08
- Publication Date
- 2026-04-17
AI Technical Summary
Existing hydrogen production power systems lack adaptability to multiple operating conditions, have low accuracy in operating condition identification, unreasonable control architecture, lack of targeted functional design, and poor adaptive capabilities, resulting in slow start-up of electrolyzers, high energy consumption, decreased efficiency, and insufficient safety.
A multi-scenario adaptive power electronic hydrogen production power supply system and control method are adopted, including a main circuit module, a status detection module, a hierarchical control unit and a function execution module. Multi-modal operating condition identification and parameter optimization are achieved through Kalman filtering, improved fuzzy neural network and particle swarm algorithm. Combined with harmonic excitation characteristics to activate the function, adaptive start-up and shutdown and performance testing of the electrolyzer are realized.
It improves electrolysis efficiency, shortens cold start-up time, enhances safety, and enables efficient, safe, and stable operation of the electrolytic cell, reducing the need for manual intervention.
Smart Images
Figure CN121879134A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hydrogen production power technology, and in particular to a multi-scenario adaptive power electronic hydrogen production power system and control method. Background Technology
[0002] Hydrogen energy, as an important carrier of clean energy, has seen its large-scale application in hydrogen production technology become a hot topic in the industry. Among these technologies, water electrolysis for hydrogen production is widely used due to its strong environmental friendliness and high purity. The power electronic hydrogen production power supply is a core component of the water electrolysis hydrogen production system, and its performance directly determines the hydrogen production efficiency, electrolyzer lifespan, and operational safety.
[0003] Existing hydrogen production power sources have the following technical drawbacks: (1) Insufficient adaptability to multiple operating conditions: When the cold electrolytic cell is started, the temperature is low and the internal resistance is high. The traditional power supply adopts a constant boost strategy, which leads to slow start-up and high energy consumption. When running at low power, the current is unstable, which causes the electrolytic cell to fail polarization. Moreover, the operating condition identification only relies on a single dimension parameter, without considering the electrolytic cell decay trend and parameter mutation characteristics. It lacks clear algorithm support, which leads to the lag in mode switching. (2) Lack of targeted functional design: There is no automated module for testing the performance of the electrolyzer, so it is impossible to monitor the degradation status of the electrolyzer in real time; the activation requirements of the catalytic activity of the electrolyzer are not considered, and the electrolysis efficiency is prone to decline after long-term operation. (3) Unreasonable control architecture: Most adopt a single control layer design, which makes it difficult to take into account both real-time drive and global parameter optimization. The parameter optimization has no clear algorithm calculation formula, resulting in low matching degree of operating parameters. Especially for electrolytic cells, it is impossible to make full use of their personalized characteristics to maximize efficiency. (4) Lack of adaptive capability: Existing power supplies are mostly designed for various electrolytic cells, which are highly versatile but lack precision. The operating condition identification algorithm does not combine the personalized data of the electrolytic cell, resulting in low identification accuracy and inability to quickly adapt to complex operating conditions such as the transition from cold start to normal operation and low power fluctuations. Summary of the Invention
[0004] The purpose of this invention is to provide a multi-scenario adaptive power electronic hydrogen production power supply system and control method, which solves the problems of poor adaptability to multiple working conditions, low accuracy of working condition identification, single function, and insufficient optimization of control architecture in the prior art. By clarifying the algorithm calculation formula, the feasibility of the technical solution is realized, and the high-efficiency, safe and stable operation of the electrolyzer is achieved.
[0005] To achieve the above objectives, the present invention provides a multi-scenario adaptive power electronic hydrogen production power supply system and control method, including: a main circuit module for converting input electrical energy into DC power adapted to the electrolyzer, including a rectifier unit, a filter unit and an inverter unit; The status monitoring module is used to collect electrolytic cell operating parameters, power output parameters, and environmental parameters; The hierarchical control unit includes a hardware driver layer and a functional layer. The hardware driver layer is used for driving power devices, real-time protection, and basic signal conversion, while the functional layer is used for operating condition identification, operating mode selection, and overall optimization of operating parameters. The functional execution module includes an automatic start-stop unit, a performance testing unit, and a harmonic excitation unit, which respectively realize the adaptive start-stop control, performance parameter detection, and characteristic activation functions of the electrolytic cell.
[0006] Preferably, the harmonic excitation unit generates harmonic signals within a specific frequency range through the PWM modulation strategy of the inverter unit. The amplitude of the harmonic signal is 5%-15% of the rated operating voltage of the electrolytic cell, and the waveform is a superposition of sine waves or square waves. The expression of the harmonic signal is as follows: ; in, For harmonic amplitude, For harmonic frequencies, The initial phase is 0-π / 2.
[0007] Preferably, the functional layer has a built-in personalized parameter database for the electrolyzer, which includes the electrolyzer's rated parameters, cold start characteristic curve, efficiency optimization range, safety constraint threshold, and historical decay data.
[0008] This invention also provides a control method for a multi-scenario adaptive power electronic hydrogen production power supply system, comprising the following steps: S1. After the system is powered on, it reads the personalized parameter database of the electrolytic cell, completes the communication adaptation between the hardware and the functional layer, and initializes the Kalman filter parameters and the weight matrix of the improved fuzzy neural network. S2, the status detection module collects data in real time, and the functional layer executes the multimodal fusion working condition identification process, which successively goes through Kalman filter preprocessing, multi-dimensional feature extraction, dynamic weight allocation and improved fuzzy neural network inference to complete working condition identification and validity confirmation; S3. Select the corresponding working mode based on the working condition identification results, and generate parameter optimization instructions through the algorithm; S4. The hardware driver layer executes optimization instructions to control the main circuit to output adapted power and synchronously trigger automatic start / stop, electrolytic cell performance testing, and harmonic excitation characteristic activation functions. S5. Real-time monitoring of operating parameters. When an anomaly is detected, the hardware driver layer will trigger the protection mechanism first, and the functional layer will perform emergency shutdown or mode switching based on the operating condition identification results.
[0009] Preferably, in S2, the Kalman filter preprocessing specifically involves using the Kalman filter algorithm to suppress noise in the acquired voltage and current signals. The core calculation formula is as follows: Equations of state: ,in, For voltage / current state vectors, Here is the state transition matrix. To control the input matrix, This is process noise; Observation equation: ,in These are sensor observations. For the observation matrix, To observe noise; Filter gain update: ,in express k Kalman filter gain at time t. express k The prior error covariance matrix at time t. Represents the observation matrix. Represents the observation matrix The transpose of the matrix, Represents the observation noise variance matrix; State estimation update: ,in express k The posterior state estimate at time t. express k The prior state estimate at time t. express k The vector of sensor observations at time 10:00. This represents the predicted observations based on prior state estimates; Adaptive adjustment of filter coefficients: ,in, This represents the adaptive filtering coefficients of the Kalman filter. This is the rated voltage of the electrolytic cell, ranging from 60V to 100V.
[0010] Preferably, multi-dimensional feature extraction adds a decay trend predictor and parameter change rate feature to the traditional features; Decline trend predictor: Calculated using a weighted linear regression model, the formula is as follows: ; in, For electrolysis efficiency, The slope of the weighted linear regression model. The intercept of the weighted linear regression model; This represents the cumulative operating time of the electrolytic cell. For the first i The weights of each historical data sample, The total number of historical data samples used in the regression calculation. For the first i The operating time of the electrolytic cell corresponding to each historical sample It is a weighted average of the runtime of all historical samples. For the first i Electrolysis efficiency corresponding to each historical sample This is the weighted average of the electrolysis efficiencies of all historical samples. Weighting: ; in, The current moment; Attenuation rate: That is, the decay trend predictor factor; Parameter change rate characteristics: Calculated using a 50ms sliding window, the formula is as follows: Voltage change rate: ; Rate of change of current: ; in, The rate of change of voltage. For the first k Filtered voltage value at each sampling time. For the first k Filtered voltage values at -50 sampling times. The sampling period is The rate of change of current, For the first k The filtered current value at each sampling time. For the first k -Filtered current values at 50 sampling times.
[0011] Preferably, dynamic weight allocation: a dynamic weight matrix for working condition adaptation is introduced, calculated as follows: Weight matrix definition: ,satisfy ; in, It is a dynamic weight matrix. The weights for the temperature characteristics of the electrolytic cell. The weights for the output power characteristics, The weights for the voltage fluctuation coefficient characteristics. The weights of the runtime feature. The weights of the features that predict the decay trend. The weights for the parameter change rate feature; Working condition adaptive weights: Cold start condition, : ; Low power operation : ; Under normal operating conditions, : ; Weight iteration update: ;in, This is the dynamic weight matrix for the next iteration. This is the dynamic weight matrix at the current moment. For the first i The weight update amount of each feature. For the first i The working condition distinguishability of each feature For the first i Average working condition discrimination of each feature.
[0012] Preferred, improved fuzzy neural network inference: A five-level neural network structure is used, with the following calculation formulas for each layer: Input layer: ;in, For the first i The weighted value of each input feature. The standardized eigenvalues, i range from 1 to 6; Fuzzification layer: using Gaussian membership function. ;in, For the first i The input feature belongs to the first... j The membership degree of a fuzzy subset. As cluster center, Standard deviation j Take 1-3; and All were generated based on 1000h trial operation data of the electrolytic cell; Rule-based reasoning layer: Employs Mamdani fuzzy logic; rule strength: ;in, For the first l The rule strength of a fuzzy rule l Take 1-108; Deblurring layer: Center of gravity method: ;in, This is the clear output value after deblurring. Output values for the rules; Hidden layer: Hybrid activation function: ;in, For the first m The output values of each hidden layer node and These are the first and second parts of the Sigmoid function. m Each node's weight and bias. and These are the first two parts of the ReLU function. m Each node's weight and bias. m Take 1-12; Output layer: Operating condition category coding Confidence level and operating condition mutation markers ; in, The final output value of the output layer, cold start. O =1, running normally =2, low power operation =3, Performance Test =4; For the first m The connection weights from hidden layer nodes to the output layer For the first m The output values of each hidden layer node For output layer bias; To determine the confidence level for the operating condition, Marker for sudden changes in operating conditions; Operating condition verification: When the confidence level of operating condition identification is... And the operating condition change marker When the current operating condition is deemed valid, a switch to normal mode is triggered; when the operating condition changes abruptly, a flag is displayed. At this time, a fast identification branch is initiated, reducing the inference cycle from 1 second to 200 ms. Simultaneously, the functional layer sends pre-adjustment instructions to the hardware driver layer, including: pre-adjustment voltage instructions. Pre-adjustment current command: .
[0013] Preferably, in S3, the operating modes include: Accelerated startup of cold electrolyzer mode: The gradient current boost formula increases in 5-second increments, with the initial current being 30%-50% of the rated current. The boost rate is adjusted based on temperature feedback to avoid large current surges. The harmonic excitation superposition formula clearly defines the synthesis logic of the fundamental voltage and harmonic signals. The fundamental voltage is calculated based on the electrolyzer's equivalent resistance and output current, ensuring the excitation signal is adapted to the electrolyzer's characteristics. The specific calculation formula is as follows: Gradient current boost formula: ;in, for t The power supply outputs current commands to the electrolytic cell at all times. This is the rated current of the electrolytic cell, and , Startup time; Temperature feedback adjustment: When When, adjust to: ; Harmonic excitation superposition: , ;in, This is the equivalent resistance of the electrolytic cell; Efficiency-first mode: The particle swarm optimization algorithm is used to optimize the PWM duty cycle and switching frequency. The fitness function aims to maximize electrolysis efficiency. The fitness function calculation formula is: ;in, For hydrogen energy production power, For electrolysis efficiency, This refers to the power supply output power. Particle position update: ;in, ;in, For PWM duty cycle, The switching frequency; Particle velocity update: ;in, The inertia weight is set to 0.7-0.9. and All are acceleration factors. , ; Constraints: , , ; Normal operating mode: Outputs according to rated parameters, output voltage command. Current command Fluctuation adjustment is achieved through duty cycle fine-tuning with an adjustment step size of 0.001. The fluctuation adjustment calculation formula is: when hour, , This is the adjusted PWM duty cycle. This represents the PWM duty cycle at the current moment. Low-power safety constraint mode: Current lower limit constraint formula: ,when hour, ;in, This refers to the rated power of the electrolytic cell; Switching frequency adjustment formula: .
[0014] Preferably, in S4, the automatic start-stop function is as follows: when starting, the liquid level and insulation status of the electrolytic cell are detected first, and if they are qualified, the preset voltage boosting process is executed; when stopping, a gradual voltage reduction + delayed power-off strategy is adopted, with a voltage reduction rate of 5V / s and a main circuit cut off after a delay of 30s. Electrolyzer performance testing function: periodically triggered, test parameters include electrolysis efficiency, gas production rate, voltage and current response speed, test results are compared with a personalized parameter database, and performance degradation warning is generated. Feature activation function: Triggers harmonic excitation after cold start and during performance testing. The excitation duration is 10 minutes, and the excitation frequency is switched in steps of 50Hz, 200Hz, 500Hz, and 1kHz. The harmonic amplitude is adaptively adjusted based on the gas production rate changes fed back by the status detection module.
[0015] Therefore, the beneficial effects of the above-mentioned multi-scenario adaptive power electronic hydrogen production power supply system and control method of the present invention are as follows: (1) Efficiency improvement: This invention maintains electrolysis efficiency above 90% by accurately adapting to the multimodal working condition identification, optimizing layered parameters and dynamically compensating for attenuation trends. (2) Start-up acceleration: The cold start time is significantly shortened. The combination of gradient current boost formula and temperature feedback adjustment algorithm further reduces energy consumption during the start-up phase. (3) Enhanced safety: The response time for identifying sudden changes in operating conditions is shortened by 50%. The quantification algorithm for low-power safety constraints, fast protection response, and automated start-stop process effectively avoid risks such as overvoltage, overcurrent, and polarization failure, and extend the service life of the electrolytic cell. (4) High level of intelligence: automated performance testing, feature activation and multi-modal working condition adaptive switching, algorithm calculation ensures automated execution of the process, reduces human intervention, and adapts to the unattended requirements of large-scale hydrogen production scenarios.
[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the composition structure of a multi-scenario adaptive power electronic hydrogen production power supply system according to the present invention. Figure 2 This is a schematic diagram of the control method steps for a multi-scenario adaptive power electronic hydrogen production power supply system according to the present invention. Detailed Implementation
[0018] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0019] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0020] Example 1: like Figure 1 As shown, this invention provides a multi-scenario adaptive power electronic hydrogen production power supply system, including: a main circuit module for converting input electrical energy into DC power adapted to the electrolyzer, comprising a rectifier unit, a filter unit, and an inverter unit. The rectifier unit uses a three-phase bridge rectifier circuit, the filter unit uses an LC filter (10mH inductor, 1000μF capacitor), and the inverter unit uses a full-bridge inverter topology with IGBT model FF450R12ME4.
[0021] The status detection module is used to collect electrolytic cell operating parameters, power output parameters and environmental parameters, including voltage sensor (LV28-P), current sensor (LA28-NP), temperature sensor (PT100), and liquid level sensor (ultrasonic liquid level gauge, data sampling frequency 1kHz).
[0022] The hierarchical control unit comprises a hardware driver layer and a functional layer. The hardware driver layer is used for driving power devices, real-time protection, and basic signal conversion. The functional layer is used for operating condition identification, operating mode selection, and overall optimization of operating parameters. The functional layer includes a personalized parameter database for the electrolyzer, containing its rated parameters, cold start characteristic curves, efficiency optimization range, safety constraint thresholds, and historical degradation data. The hardware driver layer uses a TMS320F28335 DSP chip, the functional layer uses an STM32H743 ARM chip, and the CAN bus communication module is a TJA1050 with a transmission rate of 1 Mbps.
[0023] The functional execution module includes an automatic start / stop unit, a performance testing unit, and a harmonic excitation unit, which respectively realize the adaptive start / stop control, performance parameter detection, and feature activation functions of the electrolytic cell. The automatic start / stop unit controls the main circuit's on / off state via relays. The performance testing unit integrates an SD card data storage module (storing 1000 hours of operating data). The harmonic excitation unit generates harmonic signals within a specific frequency range using the inverter unit's PWM modulation strategy. The amplitude of the harmonic signals is 5%-15% of the electrolytic cell's rated operating voltage, and the waveform is a superposition of sine waves or square waves. The expression for the harmonic signals is: in, For harmonic amplitude, For harmonic frequencies, The initial phase is 0-π / 2.
[0024] like Figure 2 As shown, the control method for a multi-scenario adaptive power electronic hydrogen production power supply system includes the following steps: S1. After the system is powered on, it reads the personalized parameter database of the electrolytic cell (including the rated voltage, rated current, cold start temperature threshold, efficiency optimization range, historical attenuation data, etc. of the electrolytic cell), completes the hardware and functional layer communication adaptation, and initializes the Kalman filter parameters (state transition matrix). A Process noise variance Q Observation noise variance R The improved fuzzy neural network weight matrix (initial weights are set based on factory test data) ensures data transmission synchronization and algorithm initialization.
[0025] S2, the status detection module collects data in real time, and the functional layer executes the multimodal fusion working condition identification process, which successively goes through Kalman filter preprocessing, multi-dimensional feature extraction, dynamic weight allocation and improved fuzzy neural network inference to complete working condition identification and validity confirmation; Kalman filtering preprocessing is performed as follows: Since the voltage and current signals acquired by the state detection module are susceptible to power grid fluctuations and sensor noise interference, a Kalman filtering algorithm is used for noise reduction. The filter coefficients are adaptively adjusted based on the rated voltage / current of the fixed electrolytic cell to ensure the accuracy of feature extraction. The core calculation formula is as follows: Equations of state: ,in, For voltage / current state vectors, Here is the state transition matrix. To control the input matrix, Process noise (variance) ).
[0026] Observation equation: ,in These are sensor observations. For the observation matrix, For observation noise (variance) ).
[0027] Filter gain update: ,in express k Kalman filter gain at time t. express k The prior error covariance matrix at time t. Represents the observation matrix. Represents the observation matrix The transpose of the matrix, Represents the observation noise variance matrix; State estimation update: ,in express k The posterior state estimate at time t. express k The prior state estimate at time t. express k The vector of sensor observations at time 10:00. This represents the predicted observations based on prior state estimates; Adaptive adjustment of filter coefficients: ,in, This represents the adaptive filtering coefficients of the Kalman filter. This is the rated voltage of the electrolytic cell, ranging from 60V to 100V.
[0028] Multi-dimensional feature extraction breaks through the limitations of traditional single-feature recognition. It adds two core features to the traditional features, extracting a total of six core input features to form a multi-modal feature vector, calculated as follows: Traditional characteristics: Electrolytic cell temperature: (Data collected directly from the sensor, unit: °C).
[0029] Output power: (Voltage and current product after Kalman filtering, unit: kW).
[0030] Voltage fluctuation coefficient: ,in, n =100 (100ms sliding window) This represents the average voltage within the window.
[0031] Runtime: (Unit: h).
[0032] New features: Decline trend predictor: Calculated using a weighted linear regression model, the formula is as follows: in, For electrolysis efficiency, The slope of the weighted linear regression model. The intercept of the weighted linear regression model; This represents the cumulative operating time of the electrolytic cell. For the first i The weights of each historical data sample, The total number of historical data samples used in the regression calculation. For the first i The operating time of the electrolytic cell corresponding to each historical sample It is a weighted average of the runtime of all historical samples. For the first i Electrolysis efficiency corresponding to each historical sample This is the weighted average of the electrolysis efficiencies of all historical samples.
[0033] Weighting: in, This refers to the current moment.
[0034] Attenuation rate: That is, the decay trend predictor factor.
[0035] Parameter change rate characteristics: Calculated using a 50ms sliding window, the formula is as follows: Voltage change rate: Current change rate: ; in, The rate of change of voltage. For the first k Filtered voltage value at each sampling time. For the first k Filtered voltage values at -50 sampling times. The sampling period is The rate of change of current, For the first k The filtered current value at each sampling time. For the first k -Filtered current values at 50 sampling times.
[0036] Dynamic weight allocation: A dynamic weight matrix adapted to different operating conditions is introduced. The weight allocation for each operating condition is defined by a clear formula to meet the requirements of different feature importance under different operating conditions. The weight matrix is iteratively updated every 500ms based on feedback data from the hardware driver layer. The update amount is calculated through feature discrimination to ensure dynamic optimization of weight adaptation. The specific calculation formula is as follows: Weight matrix definition: ,satisfy ; in, It is a dynamic weight matrix. The weights for the temperature characteristics of the electrolytic cell. The weights for the output power characteristics, The weights for the voltage fluctuation coefficient characteristics. The weights of the runtime feature. The weights of the features that predict the decay trend. The weights are the feature of the rate of change of the parameters.
[0037] Working condition adaptive weights: Cold start condition, : ; Low power operation : ; Under normal operating conditions, : ; Weight iteration update: ;in, This is the dynamic weight matrix for the next iteration. This is the dynamic weight matrix at the current moment. For the first i The weight update amount of each feature. For the first i The working condition distinguishability of each feature For the first i Average working condition discrimination of each feature.
[0038] Improved fuzzy neural network inference: A five-level neural network structure is adopted, and the calculation formulas for each layer are as follows: Input layer: ;in, For the first i The weighted value of each input feature. The standardized eigenvalues, i range from 1 to 6; Fuzzification layer: using Gaussian membership function. ;in, For the firsti The input feature belongs to the first... j The membership degree of a fuzzy subset. As cluster center, Standard deviation j Take 1-3; and All were generated based on 1000h trial operation data of the electrolytic cell; Rule Inference Layer: The 108 rules in the rule inference layer are generated by combining three fuzzy subsets of six features, and effective rules are selected based on the personalized operating conditions of the fixed electrolytic cell. Mamdani fuzzy logic is used, with the rule strength as follows: ;in, For the first l The rule strength of a fuzzy rule l Take 1-108; Deblurring layer: Center of gravity method: ;in, This is the clear output value after deblurring. Output values for the rules; Hidden layer: The hybrid activation function combines the Sigmoid function (for gradient smoothing) and the ReLU function (to alleviate gradient vanishing). The output of the hidden layer nodes is obtained by weighted summation to ensure the accuracy of inference. Hybrid activation functions: ;in, For the first m The output values of each hidden layer node and These are the first and second parts of the Sigmoid function. m Each node's weight and bias. and These are the first two parts of the ReLU function. m Each node's weight and bias. m Take 1-12; Output layer: Simultaneously outputs the working condition category code, confidence level, and working condition change marker. The confidence level is calculated by normalizing the rule strength, and the working condition change marker is determined by the parameter change rate threshold. Operating condition category code Confidence level and operating condition mutation markers ;in, The final output value of the output layer, cold start. O =1, running normally =2, low power operation =3, Performance Test =4; For the first m The connection weights from hidden layer nodes to the output layer For the first m The output values of each hidden layer node For output layer bias; To determine the confidence level for the operating condition, This serves as a marker for sudden changes in operating conditions.
[0039] Operating condition verification: When the confidence level of operating condition identification is... And the operating condition change marker When the current operating condition is deemed valid, a switch to normal mode is triggered; when the operating condition changes abruptly, a flag is displayed. At this time, a fast identification branch is initiated, reducing the inference cycle from 1 second to 200ms. Simultaneously, the functional layer sends pre-adjustment instructions to the hardware driver layer to avoid voltage and current surges caused by sudden changes in operating conditions. Pre-adjustment instructions include: Pre-adjustment voltage instructions: Pre-adjustment current command: .
[0040] S3. The functional layer selects the corresponding working mode based on the working condition identification results, and generates parameter optimization instructions through the algorithm by combining the personalized parameters of the electrolytic cell, and sends them to the hardware driver layer.
[0041] Accelerated startup of cold electrolyzer mode: The gradient current boost formula increases in 5-second increments, with the initial current being 30%-50% of the rated current. The boost rate is adjusted based on temperature feedback to avoid large current surges. The harmonic excitation superposition formula clearly defines the synthesis logic of the fundamental voltage and harmonic signals. The fundamental voltage is calculated based on the electrolyzer's equivalent resistance and output current, ensuring the excitation signal is adapted to the electrolyzer's characteristics. The specific calculation formula is as follows: Gradient current boost formula: ;in, for t The power supply outputs current commands to the electrolytic cell at all times. This is the rated current of the electrolytic cell, and , This is the startup duration.
[0042] Temperature feedback adjustment: When When, adjust to: .
[0043] Harmonic excitation superposition: , ;in, This is the equivalent resistance of the electrolytic cell.
[0044] Efficiency-first mode: The particle swarm optimization algorithm is used to optimize the PWM duty cycle and switching frequency. The fitness function aims to maximize electrolysis efficiency. The fitness function calculation formula is: ;in, For hydrogen energy production power, For electrolysis efficiency, This refers to the power output of the power supply.
[0045] Particle position update: ;in, ;in, For PWM duty cycle, This represents the switching frequency.
[0046] Particle velocity update: ;in, The inertia weight is set to 0.7-0.9. and All are acceleration factors. , ; Constraints: , , .
[0047] Normal operating mode: Outputs according to rated parameters, output voltage command. Current command Fluctuation adjustment is achieved through duty cycle fine-tuning with an adjustment step size of 0.001. The fluctuation adjustment calculation formula is: when hour, , This is the adjusted PWM duty cycle. This represents the PWM duty cycle at the current moment.
[0048] Low-power safety constraint mode: The current lower limit constraint formula ensures that the output current is not less than 10% of the rated current, and the switching frequency adjustment formula decreases linearly according to the output power to avoid polarization failure during low-power operation. Current lower limit constraint formula: ,when hour, ;in, This is the rated power of the electrolytic cell.
[0049] Switching frequency adjustment formula: .
[0050] S4. The hardware driver layer executes optimization instructions to control the main circuit to output adapted power, and synchronously triggers automatic start / stop, electrolytic cell performance testing, and harmonic excitation characteristic activation functions.
[0051] Automatic start / stop function: Upon startup, the electrolytic cell level and insulation status are checked first. If they pass the checks, a preset voltage boosting process is executed. Upon shutdown, a gradual voltage reduction + delayed power-off strategy is adopted, with a voltage reduction rate of 5V / s and a 30s delay before cutting off the main circuit. The status detection module feeds back operating data at a frequency of 1kHz. The functional layer updates the dynamic weight matrix and the improved fuzzy neural network weights based on the feedback data, iteratively optimizing parameters to form a closed-loop control.
[0052] Electrolyzer performance testing function: Triggered periodically, the test parameters include electrolysis efficiency, gas production rate, and voltage and current response speed. The test results are compared with a personalized parameter database to generate performance degradation warnings. After the electrolyzer performance testing function is executed, the test data is automatically updated to the personalized parameter database, providing the latest data support for the weighted linear regression calculation of degradation trend prediction factors.
[0053] Feature activation function: Triggers harmonic excitation after cold start and during performance testing. The excitation duration is 10 minutes, and the excitation frequency is switched in steps of 50Hz, 200Hz, 500Hz, and 1kHz. The harmonic amplitude is adaptively adjusted based on the gas production rate changes fed back by the status detection module.
[0054] S5. Real-time monitoring of operating parameters. When an anomaly is detected, the hardware driver layer will trigger the protection mechanism first, and the functional layer will perform emergency shutdown or mode switching based on the operating condition identification results.
[0055] When overvoltage (exceeding 120% of rated voltage), overcurrent (exceeding 150% of rated current), or overtemperature (electrolytic cell temperature exceeding 80℃) is detected, the hardware driver layer prioritizes triggering the protection mechanism (cutting off the IGBT drive signal) and simultaneously feeds back to the functional layer. The functional layer, based on the operating condition identification results, executes an emergency shutdown procedure (gradual voltage reduction + delayed power cut-off). If there is a sudden change in operating condition, the harmonic excitation output is cut off first to avoid the risk from escalating.
[0056] Therefore, the present invention adopts a multi-scenario adaptive power electronic hydrogen production power supply system and control method, which solves the problems of poor multi-condition adaptability, low accuracy of condition identification, single function, and insufficient optimization of control architecture in the prior art. By clarifying the algorithm calculation formula, the feasibility of the technical solution is realized, and the efficient, safe and stable operation of the electrolyzer is achieved.
[0057] 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 them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A multi-scenario adaptive power electronic hydrogen production power supply system, characterized in that, include: The main circuit module is used to convert the input electrical energy into DC power suitable for the electrolytic cell, and includes a rectifier unit, a filter unit and an inverter unit; The status monitoring module is used to collect electrolytic cell operating parameters, power output parameters, and environmental parameters; The hierarchical control unit includes a hardware driver layer and a functional layer. The hardware driver layer is used for driving power devices, real-time protection, and basic signal conversion, while the functional layer is used for operating condition identification, operating mode selection, and overall optimization of operating parameters. The functional execution module includes an automatic start-stop unit, a performance testing unit, and a harmonic excitation unit, which respectively realize the adaptive start-stop control, performance parameter detection, and characteristic activation functions of the electrolytic cell.
2. The multi-scenario adaptive power electronic hydrogen production power supply system according to claim 1, characterized in that: The harmonic excitation unit generates harmonic signals within a specific frequency range through the PWM modulation strategy of the inverter unit. The amplitude of the harmonic signals is 5%-15% of the rated operating voltage of the electrolytic cell, and the waveform is a superposition of sine waves or square waves. The expression for the harmonic signals is as follows: ； in, For harmonic amplitude, For harmonic frequencies, The initial phase is 0-π / 2.
3. The multi-scenario adaptive power electronic hydrogen production power supply system according to claim 1, characterized in that: The functional layer has a built-in personalized parameter database for the electrolyzer, which includes the electrolyzer's rated parameters, cold start characteristic curves, efficiency optimization range, safety constraint thresholds, and historical degradation data.
4. The control method for a multi-scenario adaptive power electronic hydrogen production power supply system as described in any one of claims 1-3, characterized in that, Includes the following steps: S1. After the system is powered on, it reads the personalized parameter database of the electrolytic cell, completes the communication adaptation between the hardware and the functional layer, and initializes the Kalman filter parameters and the weight matrix of the improved fuzzy neural network. S2, the status detection module collects data in real time, and the functional layer executes the multimodal fusion working condition identification process, which successively goes through Kalman filter preprocessing, multi-dimensional feature extraction, dynamic weight allocation and improved fuzzy neural network inference to complete working condition identification and validity confirmation; S3. Select the corresponding working mode based on the working condition identification results, and generate parameter optimization instructions through the algorithm; S4. The hardware driver layer executes optimization instructions to control the main circuit to output adapted power and synchronously trigger automatic start / stop, electrolytic cell performance testing, and harmonic excitation characteristic activation functions. S5. Real-time monitoring of operating parameters. When an anomaly is detected, the hardware driver layer will trigger the protection mechanism first, and the functional layer will perform emergency shutdown or mode switching based on the operating condition identification results.
5. The control method for a multi-scenario adaptive power electronic hydrogen production power supply system according to claim 4, characterized in that: In S2, the Kalman filter preprocessing specifically involves using the Kalman filter algorithm to suppress noise in the acquired voltage and current signals. The core calculation formula is as follows: Equations of state: ,in, For voltage / current state vectors, Here is the state transition matrix. To control the input matrix, This is process noise; Observation equation: ,in These are sensor observations. For the observation matrix, To observe noise; Filter gain update: ,in express k Kalman filter gain at time t. express k The prior error covariance matrix at time t. Represents the observation matrix. Represents the observation matrix The transpose of the matrix, Represents the observation noise variance matrix; State estimation update: ,in express k The posterior state estimate at time t. express k The prior state estimate at time t. express k The vector of sensor observations at time 10:
00. This represents the predicted observations based on prior state estimates; Adaptive adjustment of filter coefficients: ,in, This represents the adaptive filtering coefficients of the Kalman filter. This is the rated voltage of the electrolytic cell, ranging from 60V to 100V.
6. The control method for a multi-scenario adaptive power electronic hydrogen production power supply system according to claim 5, characterized in that: Multi-dimensional feature extraction adds decay trend prediction factors and parameter change rate features to the traditional features; Decline trend predictor: Calculated using a weighted linear regression model, the formula is as follows: ； in, For electrolysis efficiency, The slope of the weighted linear regression model. The intercept of the weighted linear regression model; This represents the cumulative operating time of the electrolytic cell. For the first i The weights of each historical data sample, The total number of historical data samples used in the regression calculation. For the first i The operating time of the electrolytic cell corresponding to each historical sample It is a weighted average of the runtime of all historical samples. For the first i Electrolysis efficiency corresponding to each historical sample This is the weighted average of the electrolysis efficiencies of all historical samples. Weighting: ; in, The current moment; Attenuation rate: That is, the decay trend predictor factor; Parameter change rate characteristics: Calculated using a 50ms sliding window, the formula is as follows: Voltage change rate: ; Rate of change of current: ; in, The rate of change of voltage. For the first k Filtered voltage value at each sampling time. For the first k Filtered voltage values at -50 sampling times. The sampling period is The rate of change of current, For the first k The filtered current value at each sampling time. For the first k -Filtered current values at 50 sampling times.
7. The control method for a multi-scenario adaptive power electronic hydrogen production power supply system according to claim 6, characterized in that: Dynamic weight allocation: A dynamic weight matrix for working condition adaptation is introduced, calculated as follows: Weight matrix definition: ,satisfy ; in, It is a dynamic weight matrix. The weights for the temperature characteristics of the electrolytic cell. The weights for the output power characteristics, The weights for the voltage fluctuation coefficient characteristics. The weights of the runtime feature. The weights of the features that predict the decay trend. The weights for the parameter change rate feature; Working condition adaptive weights: Cold start condition, : ； Low power operation : ； Under normal operating conditions, : ； Weight iteration update: ;in, This is the dynamic weight matrix for the next iteration. This is the dynamic weight matrix at the current moment. For the first i The weight update amount of each feature. For the first i The working condition distinguishability of each feature For the first i Average working condition discrimination of each feature.
8. The control method for a multi-scenario adaptive power electronic hydrogen production power supply system according to claim 7, characterized in that: Improved fuzzy neural network inference: A five-level neural network structure is adopted, and the calculation formulas for each layer are as follows: Input layer: ;in, For the first i The weighted value of each input feature. The standardized eigenvalues, i range from 1 to 6; Fuzzification layer: using Gaussian membership function. ;in, For the first i The input feature belongs to the th . j The membership degree of a fuzzy subset. As cluster center, Standard deviation, j Take 1-3; and All were generated based on 1000h trial operation data of the electrolytic cell; Rule-based reasoning layer: Employs Mamdani fuzzy logic; rule strength: ;in, For the first l The rule strength of a fuzzy rule l Take 1-108; Deblurring layer: Center of gravity method: ;in, This is the clear output value after deblurring. Output values for the rules; Hidden layer: Hybrid activation function: ;in, For the first m The output values of each hidden layer node and These are the first and second parts of the Sigmoid function. m Each node's weight and bias. and These are the first two parts of the ReLU function. m Each node's weight and bias. m Take 1-12; Output layer: Operating condition category coding Confidence level and operating condition mutation markers ; in, The final output value of the output layer, cold start. O =1, running normally =2, low power operation =3, Performance Test =4; For the first m The connection weights from each hidden layer node to the output layer. For the first m The output values of each hidden layer node For output layer bias; To determine the confidence level for the operating condition, Marker for sudden changes in operating conditions; Operating condition verification: When the confidence level of operating condition identification is... And the operating condition change marker When the current operating condition is deemed valid, a switch to normal mode is triggered; when the operating condition changes abruptly, a flag is displayed. At this time, a fast identification branch is initiated, reducing the inference cycle from 1 second to 200 ms. Simultaneously, the functional layer sends pre-adjustment instructions to the hardware driver layer, including: pre-adjustment voltage instructions. Pre-adjustment current command: .
9. The control method for a multi-scenario adaptive power electronic hydrogen production power supply system according to claim 8, characterized in that: In S3, the working modes include: Accelerated startup of cold electrolyzer mode: The gradient current boost formula increases in 5-second increments, with the initial current being 30%-50% of the rated current. The boost rate is adjusted based on temperature feedback to avoid large current surges. The harmonic excitation superposition formula clearly defines the synthesis logic of the fundamental voltage and harmonic signals. The fundamental voltage is calculated based on the electrolyzer's equivalent resistance and output current, ensuring the excitation signal is adapted to the electrolyzer's characteristics. The specific calculation formula is as follows: Gradient current boost formula: ;in, for t The power supply outputs current commands to the electrolytic cell at all times. This is the rated current of the electrolytic cell, and , Startup time; Temperature feedback adjustment: When When, adjust to: ; Harmonic excitation superposition: , ;in, This is the equivalent resistance of the electrolytic cell; Efficiency-first mode: The particle swarm optimization algorithm is used to optimize the PWM duty cycle and switching frequency. The fitness function aims to maximize electrolysis efficiency. The fitness function calculation formula is: ;in, For hydrogen energy production power, For electrolysis efficiency, This refers to the power supply output power. Particle position update: ;in, ;in, For PWM duty cycle, The switching frequency; Particle velocity update: ;in, The inertia weight is set to 0.7-0.
9. and All are acceleration factors. , ; Constraints: , , ; Normal operating mode: Outputs according to rated parameters, output voltage command. Current command Fluctuation adjustment is achieved through duty cycle fine-tuning with an adjustment step size of 0.
001. The fluctuation adjustment calculation formula is: when hour, , This is the adjusted PWM duty cycle. This represents the PWM duty cycle at the current moment. Low-power safety constraint mode: Current lower limit constraint formula: ,when hour, ;in, This refers to the rated power of the electrolytic cell; Switching frequency adjustment formula: .
10. The control method for a multi-scenario adaptive power electronic hydrogen production power supply system according to claim 4, characterized in that: In S4, the automatic start-stop function is as follows: when starting, the level and insulation status of the electrolytic cell are first detected, and if they are qualified, the preset voltage boosting process is executed; when stopping, a gradual voltage reduction + delayed power-off strategy is adopted, with a voltage reduction rate of 5V / s and a main circuit cut off after a delay of 30s. Electrolyzer performance testing function: periodically triggered, test parameters include electrolysis efficiency, gas production rate, voltage and current response speed, test results are compared with a personalized parameter database, and performance degradation warning is generated. Feature activation function: Triggers harmonic excitation after cold start and during performance testing. The excitation duration is 10 minutes, and the excitation frequency is switched in steps of 50Hz, 200Hz, 500Hz, and 1kHz. The harmonic amplitude is adaptively adjusted based on the gas production rate changes fed back by the status detection module.