A high-power hydrogen production inverter predictive control method

Abstract

This invention provides a predictive control method for a high-power hydrogen production converter, belonging to the interdisciplinary technical field of high-power power electronic conversion technology, transient control of renewable energy microgrids, and advanced stochastic predictive control algorithms. The method includes state space dimensionality reduction and construction of the transition probability matrix; and extreme speed based on absorption state reconstruction. FPP Early warning simulation to predict the future N Step warning signal FPP Based on early warning signals FPP The system dynamically softens the tracking target and uses probabilistic active zeroing of the error tracking power of the high-power hydrogen production converter system; it uses probabilistic rise-fall prediction to solve for the optimal duty cycle to withstand shocks; and it applies physical safety constraints to the duty cycle change rate for asymmetric limiting and issuing control commands. This invention overcomes the lag blind spot of existing control strategies that are "based on deterministic feedback and reactive compensation".

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of high-power power electronic conversion technology, transient control of renewable energy microgrids and advanced stochastic predictive control algorithms, and particularly relates to a predictive control method for a high-power hydrogen production converter. Background Technology

[0002] With the advancement of global "dual-carbon" goals, large-scale renewable energy-based green hydrogen production technology has become the core of energy transition. In megawatt-scale hydrogen production plants, due to the inherent randomness and volatility of wind and solar microgrids, hydrogen production converter systems must frequently cope with transient voltage drops under weak grid conditions. The alkaline electrolyzer (ALK) involved in this invention is currently the mainstream load for megawatt-scale hydrogen production, and its extremely fragile electrocatalytic active membrane is highly sensitive to the rate of current change. In actual operation, any microsecond-level transient current jump can lead to polarization voltage oscillations, membrane physical fatigue tearing, and even serious safety accidents caused by hydrogen-oxygen mixing. The transient stability of the converter under grid disturbances has become a core technical challenge in this field.

[0003] To address this challenge, the industry has developed a topology evolution path primarily focused on "hardware upgrades," such as employing expensive fully controlled IGBT rectifiers or thyristor hybrid topologies. While these advanced IGBT-based topologies theoretically possess strong resistance to voltage drops, they face obstacles in commercial engineering practice, including extremely high initial investment costs, severe circulating currents from multiple parallel units, and significant high-frequency losses, leading to a substantial increase in the cost per kilowatt-hour. Based on a deep consideration of system economy and reliability, this invention resolutely adopts the classic two-stage topology of "12-pulse open-loop rectification + multi-phase interleaved parallel buck converter." This topology boasts extremely low cost, robust and durable device physical structures, and extremely high full-load efficiency, perfectly meeting the continuous DC power supply requirements of ALK electrolyzers, which often reach thousands of amperes. However, the inherent flaw of this classic topology lies in the "open-loop" characteristic of the front-end rectifier: when the grid experiences a voltage transient drop, its DC bus will inevitably collapse proportionally, and the existing deterministic lag control algorithm will therefore generate a violent transient current jump, directly hitting the fatal flaw of the electrolytic cell diaphragm. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a predictive control method for high-power hydrogen production converters, overcoming the lag blind spot inherent in existing control strategies that rely on deterministic feedback and reactive compensation.

[0005] To achieve the above objectives, the technical solution adopted by this invention is as follows: a predictive control method for a high-power hydrogen production converter, wherein the predictive control method includes the following steps in each discrete control cycle: S1. Construct the state space dimension reduction and state transition probability matrix for the high-power hydrogen production converter input to the microgrid; S2. The one-step state transition probability matrix obtained based on the dimensionality reduction of the state space and the extreme speed of absorption state reconstruction. FPP Early warning simulation to predict the future of high-power hydrogen production converters N Step warning signal FPP ; S3, based on early warning signals FPP The tracking target is dynamically softened, and the error tracking power of the high-power hydrogen production converter system is actively zeroed out by probability to generate a brand-new tracking error. S4, Based on early warning signals FPP The expected voltage virtual feedforward and duty cycle optimization are combined with the tracking error to obtain the optimal duty cycle for the high-power hydrogen production converter to resist impact; whereby the optimal duty cycle serves as the benchmark for the single-step change rate constraint of the duty cycle. S5. Asymmetric limiting and physical safety constraints are applied to the single-step change rate of the duty cycle to generate the actual duty cycle. The actual duty cycle is then sent as a control command to the underlying actuator of the high-power hydrogen converter to complete predictive control.

[0006] The beneficial effects of this invention are as follows: This invention relates to a low-level drive control method for high-power hydrogen production converter systems under transient drop conditions in weak power grids or high-proportion new energy microgrids. This invention breaks through the technical biases of "relying on post-event error feedback" and "MPC strongly relying on online iterative optimization" in the traditional power electronics control field, through the deep integration of pure mathematical analysis and discrete probability models.

[0007] Further, S1 includes: The prediction model is simplified using Markov properties; Based on the simplified processing results, the real-time input per-unit value of the microgrid is discretized into multiple states, resulting in a state transition probability matrix:

[0008] in, Represents the state transition probability matrix. Indicates the system from its historical state i Transition to the target state j The probability of.

[0009] The beneficial effects of the aforementioned further solutions are: eliminating the curse of dimensionality and constructing a data foundation for microsecond-level prediction: Traditional microgrid transient analysis is directly based on complex differential equations in the continuous time domain. If future states are to be extrapolated in the microsecond-level control chips of the converter, the "curse of dimensionality" is highly likely to occur, leading to computing power collapse and memory overflow. This invention utilizes the Markov property of no aftereffect to perform "state space dimensionality reduction," cleverly discretizing and condensing the infinitely continuous microgrid fluctuation waveform into a finite number of state nodes, and then transforming it into an extremely lightweight state transition probability matrix. This process significantly reduces the dimensionality of the system state space from the source, completely eliminating the storage burden of massive historical state data, and laying an extremely efficient underlying mathematical foundation for subsequent ultra-fast probabilistic extrapolation.

[0010] Furthermore, S2 includes: Forcibly applying the state transition probability matrix The fatal fall state in the data is reconstructed into an absorbing state, yielding the absorbing state transition matrix. ; Based on absorption state transition matrix Using the initial probability vector π(0) generated from the current sampling, the future probability vector is determined by the CK equation. N After one control cycle, the system's state distribution vector π( N ); Extract the state distribution vector π( N The numerical value of the corresponding absorption state in the ) yields the future N Warning signal that the country has fallen into a state of disaster for the first time. FPP .

[0011] The beneficial effects of the above-mentioned further solutions are: extreme dimensionality reduction computing power, completely breaking through the underlying execution barriers of complex prediction algorithms: traditional high-end MPC and Markov multi-step prediction algorithms are prone to computational overflow, simulation algebra loop errors, or stuttering in microsecond-level discrete digital control cycles. This invention, through "absorption state reconstruction," cleverly simplifies the complex multi-step first-in-time integration of a stochastic process into an extremely lightweight dimensionality-reduced matrix multiplication. This process removes the underlying computing power obstacles for porting advanced prediction algorithms to low-cost digital processors, breaks down the computing power barriers of microsecond-level control, and provides an extremely fast disaster early warning and simulation mechanism for high-power hydrogen production converter systems.

[0012] Furthermore, the warning signal FPP The expression is as follows:

[0013]

[0014]

[0015] in, Describing the absorption state transition matrix of N Power of 1.

[0016] Furthermore, S3 includes: Define target tracking weights that decay exponentially with the probability of first passage. Simultaneously define the dynamic damping coefficient that increases exponentially with the probability of first passage. l 3; Based on dynamic damping coefficient l 3. The improved MPC value function of the 6-phase interleaved Buck converter system was calculated. :

[0017]

[0018] in, This represents the steady-state damping reference constant of a high-power hydrogen production converter system during steady-state operation. This represents the damping surge gain constant. This represents the tracking error weighting coefficient. This represents the true expected target current reconstructed using probabilistic softening. Indicates the first k The parallel branch in the next n Predicted inductor current at +1 step. Indicates the first k The parallel branch is currently the first n The theoretical duty cycle to be solved within the frame. Indicates the first k The parallel branch is currently the first n The theoretical duty cycle to be solved within -1 beat; Using target tracking weights Reconstruct the true expected target current of the current shot. I aim And calculate the new tracking error e new This completes the zeroing process for error tracking power.

[0019] The beneficial effects of the above-mentioned further solutions are: softening of dynamic targets and draining of transient impact kinetic energy: traditional anti-transient solutions suffer from system physical hysteresis, causing feedback errors to accumulate and eventually trigger huge di / dt peak currents. This invention creatively exploits the "physical hysteresis window" brought about by capacitor energy storage (such as... Figure 6 As shown), and within that window, the first-pass probability ( FPPAn advanced damping softening mechanism was constructed. This mechanism forcibly softens the controller's tracking target to the measured current level before the busbar experiences substantial collapse, ensuring that the tracking error is strictly reduced to zero. This invention abandons the traditional approach of "relying on a large inductor to withstand the impact" and directly removes the kinetic energy that causes drastic changes in the duty cycle from the analytical equations, eliminating the source of tearing the electrolytic cell diaphragm from the mathematical foundation.

[0020] Furthermore, the tracking error e new The expression is as follows:

[0021]

[0022]

[0023] in, Indicates the original target reference phase current. This represents the actual current feedback value currently sampled. This represents the sensitivity adjustment constant for adjusting the trigger steepness of a high-power hydrogen production converter system to enter a transient defense state.

[0024] Furthermore, S4 includes: Utilizing early warning signals FPP Pre-construct the expected feedforward voltage V in_expected ; Value function for improved MPC Perform a discrete expansion and set the partial derivatives... This yields a purely algebraic equation; where, Indicates the first k The parallel branch is currently the first n The theoretical duty cycle to be solved within the frame; Extracting the optimal duty cycle for high-power hydrogen production converters to withstand impact using pure algebraic equations D calc ; Optimal duty cycle D calc Latch and set the optimal duty cycle D calc This serves as a physical benchmark for constraining the single-step change rate of duty cycle in high-power hydrogen converters.

[0025] The beneficial effects of the above-mentioned further solution are: extremely fast algebraic optimization and software-compensated physical support: the MPC algorithm of traditional high-power multiphase interleaved converters requires traversing a massive number of switch combinations for online rolling iteration, which consumes a great deal of chip computing power and results in serious physical delays in the issuance of control commands; at the same time, in order to resist power imbalances caused by grid transient drops, the industry is usually forced to stack expensive and massive DC bus support capacitors for hydrogen production systems. This invention, by directly taking the partial derivative of the value function, completely reduces the time-consuming online iterative optimization to a time complexity of only O ( 1 This invention utilizes pure algebraic equations to achieve microsecond-level ultra-fast extraction of the optimal duty cycle; more importantly, it leverages early warning signals. FPP By constructing a virtual feedforward voltage in advance, and using pure algebraic equations to calculate the advanced amplified defensive duty cycle before the physical bus experiences substantial collapse, this mechanism cleverly utilizes the extremely fast "software computing power" of the underlying algorithm to perfectly compensate for the extremely expensive "hardware capacitor" support space in high-power hydrogen production converters. This significantly reduces the cost of heavy equipment while providing the system with a robust defense against sudden voltage drops in the microgrid.

[0026] Furthermore, the optimal duty cycle D calc The expression is as follows:

[0027]

[0028] in, This represents the polarization voltage of the electrolytic cell. Indicates the filter inductance. This indicates the underlying discrete control cycle. This represents the true expected target current reconstructed using probabilistic softening. This represents the actual current feedback value currently sampled. This represents the actual physical bus voltage. This represents the estimated maximum busbar drop amplitude of a high-power hydrogen production converter system under extreme operating conditions.

[0029] Furthermore, S5 includes: Calculated target tracking weights Maximum permissible duty cycle single-step change rate for real-time dynamic linkage ; Optimal duty cycle D calc Using this as a constraint benchmark, and combining it with the actual duty cycle of the previous control cycle, the theoretical single-step change expected by the MPC algorithm in the current control cycle is calculated. ; Based on the single-step change rate of the maximum permissible duty cycle For theoretical single-step change Perform asymmetric hard constraint trimming with forced engineering orientation ; Asymmetric hard constraint clipping Calculate the final actual duty cycle It then issues these commands as control instructions to the underlying execution mechanisms.

[0030] The beneficial effects of the above-mentioned further solutions are: Strengthening the asymmetric physical defense line and significantly extending the lifespan of hydrogen production equipment: The diaphragm of an alkaline (ALK) electrolyzer is extremely vulnerable to high-frequency transient currents, and abnormal impacts can easily trigger safety accidents involving hydrogen-oxygen mixing. This invention introduces an "asymmetric hard constraint tailoring" defense line at the end of the algorithm execution, tailored to the physical characteristics of the electrolyzer. It strictly limits the duty cycle loading rate in the forward direction, completely eliminating the risk of diaphragm rupture caused by electromagnetic stress; and moderately relaxes the unloading rate in the reverse direction, giving the system the ability to rapidly detach under extreme fault conditions. This asymmetric defense barrier, constructed using pure software computing power, significantly extends the fault-free operating life of megawatt-level green electricity hydrogen production equipment under weak power grids without increasing any hardware capital expenditure.

[0031] Furthermore, the actual duty cycle The expression is as follows:

[0032]

[0033]

[0034] in, This indicates the duty cycle that was latched and actually issued in the previous control cycle. This indicates the optimal duty cycle.

[0035] The beneficial effects of this invention: This invention abandons the superficial optimization of traditional power electronics control, which involves "patching and fixing," and instead delves into the core of control science's underlying mathematical analysis and physical timing, proposing a disruptive innovation with at least the following beneficial effects: 1) The refinement of control philosophy: from "passive feedback after the fact" to "probabilistic prediction before the fact," nipping overshoot shocks in the bud at their mathematical roots. The fatal flaw of previous technologies: existing PI control or conventional MPC algorithms are essentially passive control "based on error feedback." That is, the controller only activates when a microgrid voltage drop causes a bus voltage collapse, leading to a substantial deviation in the electrolyzer current. At this point, the error integral has already saturated, and in an attempt to forcibly catch up to the target, the controller violently increases the duty cycle, triggering devastating current reignition and high-frequency... di / dt peak.

[0036] The profound innovation of this invention: This invention utilizes the inherent "physical hysteresis window" of DC support capacitor discharge to preemptively address the issue before a substantial collapse in bus voltage, leveraging the Markov first-pass probability (...). FPP The engine extracts the "probability of a future fall." When the probability of danger approaches 1, the damping coefficient... α The system instantly resets to zero, forcing the controller's mathematical tracking target to be compromised to the current measured current.

[0037] The beneficial effects achieved: This invention abandons the traditional approach of passively buffering the impact by relying on a large filter inductor. Instead, it directly eliminates the driving force that causes drastic changes in duty cycle at the root in the pure mathematical analytical equations (actively bringing the tracking error to zero). This enables the electrolyzer current to smoothly and gently land regardless of the severity of the grid voltage drop, completely overcoming the world-class engineering challenge of transient impact tearing of the electrolyzer diaphragm.

[0038] 2) Limiting the dimensionality reduction of underlying computing power: Breaking down the barriers to implementing complex stochastic prediction algorithms in microsecond-level control. The computational disaster of past technologies: In discrete simulation and digital control, predicting the state of multiple future steps requires extremely large matrix powers or even integral operations in traditional Markov chains; while traditional MPC relies on online iterative optimization solvers such as quadratic programming (QP). In the low-level control of converters with step sizes of only tens of microseconds, this inevitably leads to computational overflow, algebraic loop deadlock, or extreme simulation lag, making it impossible to generate executable low-level C code.

[0039] The profound innovation of this invention lies in its two instances of "extreme computational dimensionality reduction." First, it creatively introduces an "absorption state," simplifying the complex multi-step recursion in the first half of the process into a simple 3×3 constant matrix multiplication. Second, it reduces the time-consuming quadratic programming optimization process offline by calculating partial derivatives, directly reducing the dimensionality to a time complexity of only [missing information]. O(1 The pure algebraic assignment formula for ).

[0040] The beneficial effects achieved include: removing the computational power barrier for the deployment of high-end algorithms. This allows complex stochastic prediction and optimization models, which previously could only run on supercomputers, to run in a lightweight and smooth manner on conventional discrete simulation engines, reducing single-step execution time by at least three orders of magnitude and greatly enhancing the industrial applicability of the algorithms.

[0041] 3) Reconstruction of the underlying hardware defenses: an "asymmetric" physical safety net that perfectly matches the physical vulnerability of hydrogen electrolyzers. The rigid constraints of previous technologies: Traditional converter duty cycle limiting is often fixed and symmetrical (for example, the upper and lower limits are fixed in a certain dead zone). This rigid mathematical constraint is completely divorced from the physical and chemical characteristics of alkaline (ALK) electrolyzers.

[0042] The profound innovation of this invention: It innovatively introduces a factor related to the damping coefficient. α A real-time dynamic linkage "asymmetric hard constraint tailoring" mechanism. This invention deeply understands that electrolytic cells are susceptible to "violent loading" (which can cause severe electromagnetic forces to tear the diaphragm), but require "rapid unloading" (power must be quickly withdrawn to avoid danger in case of failure). Therefore, this invention strictly limits the forward loading rate (e.g., 1.0 times the limit) while moderately relaxing the reverse unloading rate (e.g., -1.5 times the limit).

[0043] The beneficial effects achieved are as follows: Even under extreme conditions where strong electromagnetic noise interference causes the early warning algorithm to fail temporarily, this purely software-constructed asymmetric physical defense can still firmly lock down sudden output changes. This underlying protection algorithm significantly extends the fault-free continuous operation life of expensive megawatt-level green electricity hydrogen production equipment under weak power grids.

[0044] 4) System-level economic disruption: Using "high-frequency software computing power" to offset "heavy asset hardware investment" to achieve significant cost reduction. The previous capital-intensive approach to technology: For the bus collapse caused by the "commutation dead zone" of 12-pulse thyristors, which lasts for several milliseconds when the power grid drops, the traditional industry could only adopt two extremely expensive methods: one is to stack a massive number of thin-film capacitors to "harden" the energy loss; the other is to abandon the cheap thyristors and use the extremely expensive fully controlled IGBT rectifier bridge.

[0045] The profound innovation of this invention lies in its proposal of a cost-reduction concept: "trading software prediction time for hardware support space." Through... FPP Probabilistic prediction increases the "expected voltage" in the denominator of the control equation in advance, so that the system has completed the advance amplification compensation of the duty cycle before the physical power failure.

[0046] Beneficial effects achieved: This invention endows inexpensive hardware with the survivability of a high-end, fully controlled power supply. This allows megawatt-scale hydrogen production projects to continue confidently employing the extremely low-cost, robust 12-pulse thyristor topology, while reducing the size of the massive DC capacitors previously used to support dead zones by more than 30%. This significantly reduces capital expenditure on heavy power electronic equipment while improving the system's extreme shock resistance, providing a highly competitive low-cost solution for the commercialization of hydrogen production. Attached Figure Description

[0047] Figure 1 This is a topology diagram of a high-power hydrogen production converter system with a 12-pulse open-loop rectifier and a 6-phase interleaved Buck converter.

[0048] Figure 2 This is the overall block diagram of a control system based on Markov first-pass probability.

[0049] Figure 3 This is a flowchart of the single-step control algorithm execution based on a discrete simulation engine.

[0050] Figure 4 This is a schematic diagram of the state transitions of an absorbing Markov chain.

[0051] Figure 5 This is a graph showing the mapping between the dynamic damping coefficient and the penalty weight as the first-pass probability changes.

[0052] Figure 6 This is a comparison of the time-series response of grid-side voltage, early warning probability, and DC bus voltage under transient voltage drop conditions in a microgrid. Detailed Implementation

[0053] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0054] Example Before describing the present invention, the following terms shall be explained: FPP (First Passage Probability): The first pass probability refers to the cumulative probability that the system will fall into a specific dangerous state for the first time within the prediction time domain.

[0055] MPC (Model Predictive Control): An advanced algorithm based on a discrete model of a system to predict future responses and continuously optimize control commands.

[0056] CCM (Continuous Conduction Mode): This refers to the operating state where the converter inductor current is always greater than zero throughout a switching cycle.

[0057] ALK (Alkaline Water Electrolysis): Alkaline water electrolyzer, the high-power hydrogen production load of this invention.

[0058] Addressing the contradictions inherent in the background technology, this invention refuses to compromise on expensive hardware upgrades. Instead, it innovatively introduces a microsecond-level Markov predictive control algorithm, achieving a groundbreaking first-ever probabilistic early warning system. This invention compensates for hardware deficiencies through a purely software-based reconstruction of underlying computing power. Milliseconds before the actual physical bus falls, it uses probabilistic radar to preemptively modify the underlying model's predictive control equations. This invention relates to proactive defensive predictive control that deeply couples stochastic probability with electromagnetic transient equations. It endows traditional, inexpensive open-loop systems with forward-looking, proactive transient defense capabilities, perfectly eliminating the risk of current surges in electrolyzers without any additional hardware costs, pushing the economy and safety of high-power hydrogen production power supplies to their limits.

[0059] like Figure 1 and Figure 2 As shown, the hydrogen production power system provided by the present invention mainly includes four physical parts: a front-end rectifier module, a DC bus support module, a back-end step-down module, and a processing unit based on a discrete domain state-space prediction model.

[0060] The pre-amplifier module uses a 12-pulse rectifier connected to the AC microgrid via a phase-shifting transformer, eliminating specific subharmonics, but its DC-side output voltage... (i.e., the input voltage of the subsequent Buck-buck converter) It exists in an open-loop, uncontrollable state. The mathematical expression for its output voltage is:

[0061] in, Indicates the output DC bus voltage. Indicates the effective value of the grid-side line voltage. Indicates the trigger angle.

[0062] The above formula shows that the output DC bus voltage RMS value of grid-side line voltage They exhibit a strictly linear proportional relationship. As semi-controlled devices, thyristors have a natural commutation dead zone. When a transient voltage drop occurs in the microgrid, not only does the effective value of the grid-side line voltage decrease... When the voltage drops, the thyristor may also "physically lock up" due to the reverse voltage of the power grid, resulting in a decrease in the DC bus voltage. A proportional or even more severe cliff-like collapse will inevitably occur.

[0063] The subsequent step-down module: A 6-phase interleaved parallel Buck converter is responsible for stepping down the unstable bus voltage and outputting a constant high DC current to the electrolytic cell. The continuous-time domain electromagnetic equation of a single-phase Buck circuit in continuous conduction mode (CCM) is:

[0064] in, This indicates the value of the filter inductance in the subsequent step-down module. Indicates the first k Parallel branches in continuous time t The inductor current under the current, Indicates the first k Parallel branches in continuous time t The drive duty cycle below, Indicates the continuous time of the pre-stage rectification. t The DC bus voltage output at all times. This indicates that the alkaline electrolyzer connected to the output terminal operates continuously for a period of time. t Real-time polarization voltage at any given moment.

[0065] Discretize it using the forward Euler method to obtain the prediction model of conventional model predictive control (MPC):

[0066] in, Indicates the first k The parallel branch is currently the first n The measured inductor current of the sampled image. Indicates the first k The parallel branch in the next n Predicted inductor current at +1 step. This indicates the underlying discrete control cycle. Indicates the first k The parallel branch is currently the first n The theoretical duty cycle to be solved within the frame. Indicates the continuous time of the pre-stage rectification. t The DC bus voltage output at all times. This indicates that the alkaline electrolyzer connected to the output terminal operates continuously for a period of time. t Real-time polarization voltage at any given moment.

[0067] Cost function of traditional MPC Seeking excellence:

[0068] in, This represents the tracking error weighting coefficient. This represents the damping weighting coefficient. This represents the original target reference phase current.

[0069] Tracking error weighting coefficient l 1 determines the system's tolerance to "the predicted state variable (actual current) deviating from the given reference trajectory (target current)," and essentially characterizes the controller's closed-loop response bandwidth and forward drive gain.

[0070] When the tracking error weighting coefficient l When phase 1 is dominant, the system exhibits extremely strong "rigid tracking" characteristics. In order to forcibly eliminate steady-state errors within a single cycle or a finite number of cycles, the controller will output extreme control actions (such as sudden changes in duty cycle saturation). Although this can shorten the dynamic response time, under large disturbances such as grid transient drops, excessive error driving force can easily lead to insufficient system phase margin, thereby causing severe overshoot and high-frequency current reignition.

[0071] Damping weight coefficient l 2 determines the system's suppression strength against the "rate of change of switching state or duty cycle within adjacent control cycles," essentially introducing virtual inertia and damping characteristics into the system. Damping weighting coefficient. l 2 represents the system's "control cost". Increasing the damping weight coefficient... l 2 is equivalent to inserting a low-pass filter or adding physical viscous friction into the control loop. It effectively limits the severity of the switching action and smooths the transient transition process. However, in traditional control, if the damping weighting coefficient is... l Setting it too high will severely sacrifice the system's command following speed and resistance to load disturbances in steady state.

[0072] Therefore, the fatal limitation of traditional MPC lies in: the tracking error weighting coefficient. l 1 and damping weight coefficient l Both are global static constants, forming an irreconcilable contradiction. When facing extreme nonlinear disturbances such as deep grid sags, the statically fixed forward drive gain (tracking error weighting coefficient) becomes ineffective. l 1) With weak system damping (damping weight coefficient) l 2) The inability to adaptively reconstruct based on transient risks will inevitably lead to the optimization algorithm falling into "forced tracking deadlock", ultimately causing system instability and collapse.

[0073] If the deadbeat control concept is adopted (the tracking error term is forced to be 0, i.e.) i k (n+1)= I ref_ph The traditional output duty cycle is... for:

[0074] When the power grid drops When the current decreases, the denominator in the traditional formula becomes smaller, and this is to catch up with the fixed reference phase current. The numerator error term generates a huge positive thrust. This allows the controller to forcibly calculate an extremely wide duty cycle at the instant the bus drops. This kind of "passive and intense compensation after the fact" will cause the total output current to rise at a high frequency. di / dtSudden spikes in current. The diaphragm of the ALK electrolyzer is extremely fragile. High-frequency sudden current spikes not only cause violent oscillations in the polarization voltage, but can also directly tear the diaphragm, leading to hydrogen-oxygen mixing and causing serious safety accidents.

[0075] like Figure 3 As shown, this invention provides a predictive control method for a high-power hydrogen production converter, in each discrete control cycle. Within, perform the following steps in chronological order: S1. Construct the state-space dimensionality reduction and state transition probability matrix for the high-power hydrogen production converter input to the microgrid. The implementation method is as follows: The prediction model is simplified using Markov properties; Based on the simplified processing results, the real-time input per-unit value of the microgrid is discretized into multiple states, resulting in a state transition probability matrix.

[0076] In this embodiment, the state transition probability matrix is ​​constructed as follows: The probability prediction of this invention is based on the three cornerstone formulas of Markov chains.

[0077] Definition of no aftereffect: The memory and computational power required to simplify the prediction model using Markov properties. Let... For the system at time The state space is The expression for the prediction model is as follows:

[0078] in, Describes the partition of the finite state space into the first... n Discrete physical state nodes Represents conditional probability. This indicates the current state of the high-power hydrogen production converter system. t Discretized state, This indicates the high-power hydrogen production converter system at the next moment. t Discretized state of +1, This represents the high-power hydrogen production converter system from the initial time 0 to the previous time. t Historical state during the period -1 i This indicates the current state of the high-power hydrogen production converter system. j This indicates the target state that the high-power hydrogen production converter system will reach in the next moment. The equation represents the historical state values ​​experienced by the high-power hydrogen production converter system in the past; the left side of the equation represents the state transition of the system to in the next moment, given the complete historical trajectory of the high-power hydrogen production converter system from the initial moment to the current moment. j The probability; the right side of the equation indicates that only when the system is currently in a historical state.i Under the premise that the system will transition to the target state in the next moment. j The probability of; That is, the system will reach the target state in the next moment. The probability depends only on the current historical state. .

[0079] State transition probability and matrix transformation: Let the transition probability from historical state be... Transition to the target state The probability is Discretize the real-time input per-unit value of the microgrid into a healthy state ( ), early warning status ( ), Fatal fall state ( )wait n There are several states. Construct the state transition probability matrix. P :

[0080] And it satisfies the core constraint of row normalization: .

[0081] in, Represents the state transition probability matrix. Elements in the matrix (Include (etc.) indicates that the system has transitioned from its historical state. i Transition to the target state j The probability of.

[0082] S2. The one-step state transition probability matrix obtained based on the dimensionality reduction of the state space and the extreme speed of absorption state reconstruction. FPP Early warning simulation to predict the future of high-power hydrogen production converters N Step warning signal FPP The implementation method is as follows: Forcibly applying the state transition probability matrix The fatal fall state in the data is reconstructed into an absorbing state, yielding the absorbing state transition matrix. ; Based on absorption state transition matrix Using the initial probability vector π(0) generated from the current sampling, the future probability vector is determined by the CK equation. N After one control cycle, the system's state distribution vector π( N ); Extract the state distribution vector π( N The numerical value of the corresponding absorption state in the ) yields the future N Warning signal that the country has fallen into a state of disaster for the first time. FPP .

[0083] In this embodiment, combined with Figure 4 The state transition model shown executes the following sub-steps: Absorption State Reconfiguration: Since hydrogen production equipment faces irreversible damage once subjected to a voltage drop, this invention does not focus on the steady-state probability of fault recovery, but only on the probability of the initial trigger. Therefore, the state transition probability matrix is ​​forcibly... P Fatal fall states (such as) The state is reconstructed into an "absorbing state," meaning that upon entering this state, the transition probability is 0 and the residence probability is 1, resulting in the absorbing state transition matrix. P abs :

[0084] Rapid solution of the CK equation: The CK equation explains the process of the system. n The probability matrix of the step transition P (n) = P n An initial probability vector π(0) is generated based on the current sampling (e.g., if the current state is alert, then π(0) = [0, 1, 0]). Future predictions are then performed. N After one control cycle, the system's state distribution vector π( N ):

[0085] in, Represents the absorption state transition matrix of N The power of 1 represents the high-power hydrogen production converter system after... N The multi-step state transition probability matrix after each control cycle transition.

[0086] Extract the state distribution vector π( N The value of the corresponding absorption state (third term) in the equation is used to obtain the future... N The cumulative probability of falling into a state of catastrophe for the first time within a step FPP : .

[0087] S3, based on early warning signals FPP The tracking target is dynamically softened by utilizing the probability-based active zeroing of the error tracking dynamics of the high-power hydrogen production converter system to generate entirely new tracking errors. The implementation method is as follows: Define target tracking weights that decay exponentially with the probability of first passage. Simultaneously define the dynamic damping coefficient that increases exponentially with the probability of first passage. l 3; Based on dynamic damping coefficient l3. The improved MPC value function of the 6-phase interleaved Buck converter system was calculated. ; Using target tracking weights Reconstruct the true expected target current of the current shot. I aim And calculate the new tracking error e new This completes the zeroing process for error tracking power.

[0088] In this embodiment, based on the early warning signal FPP The tracking target dynamically softens (molecular remodeling). The catastrophic mechanism of traditional errors: when a fall occurs, i curr The (currently sampled feedback value) subsequently decreases. The error kinetic energy of the traditional controller... e trad = I ref_ph - i curr It will surge instantly. The controller is trying to "forcibly catch up" with the fixed target. I ref_ph It will desperately output the maximum duty cycle, which can easily cause devastating current reignition and overshoot during voltage drops or recovery.

[0089] Molecular Reconstruction: To prevent forced pursuit of full-load current, this invention introduces a nonlinear dimensionality reduction-resistance enhancement synergistic softening mechanism. A target tracking weight is defined that decays exponentially with the probability of first passage. α :

[0090] in, c This represents the sensitivity adjustment constant for regulating the trigger steepness of a high-power hydrogen production converter system entering a transient defensive state. It is used to adjust the trigger steepness of the system entering the defensive state. When a deep dip occurs in the microgrid, the target tracking weight... α It will decay exponentially and rapidly, approaching 0, and is used to actively cut off the forced tracking drive force of full-load current in subsequent control.

[0091] To match target tracking weights α In conjunction with the attenuation characteristics, a dynamic damping coefficient is defined that increases exponentially with the probability of first passage. l 3:

[0092] in, l 0 indicates that the converter system is in steady-state operation. FPPThe steady-state damping reference constant when (=0) ensures the system's fast dynamic response bandwidth under normal operating conditions. B This represents the damping surge gain constant, used to adjust the nonlinear response rate at which the system transitions to a "high-viscosity" defensive state. The dynamic damping coefficient increases when the microgrid's transient sag risk intensifies. l The exponential amplification of the 3-axis is due to the core mechanism of applying extremely high penalty weights to the drastic changes in the duty cycle of the switching transistor in the cost function, thereby forcibly suppressing the sudden changes in the control action of the system, thus injecting a large amount of virtual damping into the converter and effectively avoiding secondary oscillation shocks.

[0093] Introducing target tracking weights α and dynamic damping coefficient l After step 3, the improved MPC value function equation for this 6-phase interleaved Buck converter system can be expressed as:

[0094] in, k This indicates the parallel branch number of the 6-phase interleaved Buck converter. This represents the tracking error weighting coefficient. This represents the true desired target phase current reconstructed using probabilistic softening. Indicates the first k The parallel branches in the next discrete sampling time n +1 predicts the inductor current. This indicates the current discrete sampling count of the underlying microprocessor executing the control algorithm. Indicates the first k The parallel branch is currently the first n The theoretical duty cycle to be solved within the frame. Indicates the first k The parallel branch was in the previous shot n -1 is the duty cycle that has been latched and actually issued. This represents the steady-state damping reference constant of a high-power hydrogen production converter system during steady-state operation. This represents the damping surge gain constant; Expected Reconstruction: Utilizing Target Tracking Weights α Reconstructing the true expected goal of the current shoot I aim And a brand new tracking error e new :

[0095]

[0096] Combination Figure 5As shown in the parameter mapping curve, when the system detects a risk of grid dip, the underlying parameters of the control system will undergo the following adaptive reconfiguration: like Figure 5 As shown by the descending curve, when danger approaches... FPP →1, α →0, at this point the actual expected target I aim Instantaneous feedback value to actual current i curr The compromise is that the controller's mathematical expectation perfectly matches the measured current. This mechanism proactively eliminates the physical driving force of current tracking from its purely mathematical roots. e new →0), completely cutting off the generation di / dt The source of the peak.

[0097] In target tracking weight α At the same time as attenuation, such as Figure 5 As shown by the rising curve in the figure, the dynamic damping coefficient in the cost function l 3. Increases rapidly. Dynamic damping coefficient l The surge in 3 significantly increases the penalty on the control increment (the amplitude of the switching action), forcing the system into an extremely conservative "high-impedance" state, which limits the intense energy exchange between the inductor and the filter capacitor.

[0098] S4, Based on early warning signals FPP The expected voltage virtual feedforward and duty cycle optimization are combined with the tracking error to obtain the optimal duty cycle for the high-power hydrogen production converter to withstand impact. The optimal duty cycle serves as the benchmark for the single-step change rate constraint of the duty cycle, and its implementation method is as follows: Utilizing early warning signals FPP Pre-construct the expected feedforward voltage V in_expected ; Value function for improved MPC Perform a discrete expansion and set the partial derivatives... This yields a purely algebraic equation; where, Indicates the first k The parallel branch is currently the first n The theoretical duty cycle to be solved within the frame; Extracting the optimal duty cycle for high-power hydrogen production converters to withstand impact using pure algebraic equations D calc ; Optimal duty cycle D calc Latch and set the optimal duty cycle D calcAs a physical benchmark for constraining the single-step change rate of duty cycle in high-power hydrogen converters, it defines the abrupt boundary for subsequent shock-resistant asymmetric loading and unloading.

[0099] In this embodiment, based on the early warning signal FPP The expected voltage virtual feedforward and duty cycle optimization (denominator reconstruction) are as follows: Combination Figure 6 As shown in the timing response characteristics, this invention cleverly utilizes the objectively existing "physical lag window" between grid-side dropout and DC bus collapse, enabling the algorithm to perform proactive defense actions: Virtual feedforward reconstruction: in physical bus voltage V in Before a substantial drop occurs due to capacitor discharge, utilize early warning signals. FPP Pre-construct the expected feedforward voltage V in_expected :

[0100] in, V drop_max This represents the maximum busbar drop amplitude estimated by the system under extremely severe operating conditions.

[0101] Improved Cost Function Derivative Optimization: To completely eliminate the underlying computational barrier caused by the reliance on quadratic programming (QP) for online optimization in traditional MPC, this invention improves the aforementioned cost function... J Perform offline expansion:

[0102] To completely eliminate the computational barrier between MATLAB discrete simulation and low-level microsecond-level computation, this invention expands the computation and directly sets the partial derivatives in the algorithm kernel. This eliminates the time-consuming online iterative solution and yields a pure algebraic equation from which the optimal duty cycle is extracted. D calc :

[0103] in, This represents the polarization voltage of the electrolytic cell. Indicates the filter inductance. This indicates the underlying discrete control cycle. This represents the true expected target current reconstructed using probabilistic softening. This represents the currently sampled current feedback value. This represents the actual physical bus voltage. This represents the estimated maximum busbar drop amplitude of a high-power hydrogen production converter system under extreme operating conditions.

[0104] When danger approaches, the expected feedforward voltage introduces a feedforward mechanism. V in_expected The forced reduction causes the expected voltage in the denominator of the control equation to drop earlier, thus affecting the calculated theoretical duty cycle. D calc The trend of expansion is being shown ahead of time. This "denominator reconstruction" mechanism provides an extremely valuable lead time compensation buffer for the impending actual voltage collapse.

[0105] S5. Asymmetric limiting and physical safety constraints are applied to the single-step change rate of the duty cycle to generate the actual duty cycle. The actual duty cycle is then used as a control command to be sent to the underlying actuator of the high-power hydrogen production converter to complete predictive control. The implementation method is as follows: Calculated target tracking weights Maximum permissible duty cycle single-step change rate for real-time dynamic linkage ; Optimal duty cycle D calc Using this as a constraint benchmark, and combining it with the actual duty cycle of the previous control cycle, the theoretical single-step change expected by the MPC algorithm in the current control cycle is calculated. ; Based on the single-step change rate of the maximum permissible duty cycle For theoretical single-step change Perform asymmetric hard constraint trimming with forced engineering orientation ; Asymmetric hard constraint clipping Calculate the final actual duty cycle It then issues these commands as control instructions to the underlying execution mechanisms.

[0106] In this embodiment, the duty cycle rate of change is probability-locked (asymmetric limiting as a fallback): Although the preceding steps have eliminated the driving force of overshoot at the mathematical analytical level, in order to prevent sudden output changes in the event of rare strong electromagnetic noise interference, sampling spikes, or severe model mismatch, this step establishes a final physical safety line before the underlying driver is issued: Dynamic redline tightening: Traditional control often uses a fixed dead zone for amplitude limiting, while this invention sets a limit that is related to the damping coefficient. α Maximum permissible duty cycle single-step change rate for real-time dynamic linkage dD max :

[0107] in, k 1 indicates the high rate of change allowed by the system under healthy steady state, in order to ensure dynamic tracking performance. k 2 indicates an extremely low rate of change during the transient defensive period.

[0108] Using the principle of interpolation smoothing, when the system is in absolute safety ( α When →1), grant the controller broad action permissions. k 1) When danger approaches ( α When →0), control is instantly stripped and tightened to the physical minimum ( ). k 2) A seamless and flexible transition from "high-sensitivity tracking" to "absolute locking" has been achieved.

[0109] Extracting theoretical changes: Calculate the theoretical single-step change Δ expected by the MPC algorithm in the current control cycle. D :

[0110] in, This indicates the duty cycle that was latched and actually issued in the previous control cycle. This represents the theoretically optimal duty cycle for resisting transient shocks, obtained from the aforementioned steps.

[0111] Asymmetric tailoring for the physical properties of electrolyzers: the theoretical variation Δ D Perform asymmetric hard constraint clipping with strong engineering orientation:

[0112] This "asymmetric" design (forward limiting 1.0, reverse limiting -1.5) is the core physical barrier protecting the ALK electrolyzer in this invention.

[0113] Strictly Constrained (Restricted Loading): The electrolyzer is extremely sensitive to drastic power loading. If the theoretical positive change Δ... D Too large, and it will cause huge di / dt The spikes, intense electromagnetic forces, and polarization overvoltages can instantly and physically tear apart the fragile internal diaphragm. Therefore, the forward voltage is strictly limited to 1.0. dDmax .

[0114] Reverse relaxation (allowing rapid unloading): When the system detects an irreversible severe voltage drop at the bus or an external short-circuit fault, the system needs to remove output power as quickly as possible to prevent upstream components from burning out. Therefore, the reverse relaxation is set to -1.5. dDmax This allows the system to "rapidly retreat" at a speed slightly greater than the load increase.

[0115] Safety-driven instruction generation and dead-time protection: Calculating the actual duty cycle ultimately sent to the underlying actuator. D out :

[0116] Meanwhile, to prevent shoot-through short circuits or bootstrap capacitor charging failures in semiconductor switching devices, an absolute duty cycle dead zone protection (setting a minimum duty cycle) is added. D min =0.02, maximum duty cycle D max =0.98): .

[0117] In summary, this invention overcomes the lag blind spot of existing control strategies that are "based on deterministic feedback and reactive compensation after the fact".

Claims

1. A predictive control method for a high-power hydrogen production converter, characterized in that, Within each discrete control cycle, the predictive control method includes the following steps: S1. Construct the state space dimension reduction and state transition probability matrix for the high-power hydrogen production converter input to the microgrid; S2. The one-step state transition probability matrix obtained based on the state space dimensionality reduction and the extreme speed of absorption state reconstruction. FPP Early warning simulation to predict the future of high-power hydrogen production converters N Step warning signal FPP ; S3, based on early warning signals FPP The tracking target is dynamically softened, and the error tracking power of the high-power hydrogen production converter system is actively zeroed out by probability to generate a brand-new tracking error. S4, Based on early warning signals FPP The expected voltage virtual feedforward and duty cycle optimization are combined with the tracking error to obtain the optimal duty cycle for the high-power hydrogen production converter to resist impact; whereby the optimal duty cycle serves as the benchmark for the single-step change rate constraint of the duty cycle. S5. Asymmetric limiting and physical safety constraints are applied to the single-step change rate of the duty cycle to generate the actual duty cycle. The actual duty cycle is then sent as a control command to the underlying actuator of the high-power hydrogen converter to complete predictive control.

2. The predictive control method for a high-power hydrogen production converter according to claim 1, characterized in that, S1 includes: The prediction model is simplified using Markov properties; Based on the simplified processing results, the real-time input per-unit value of the microgrid is discretized into multiple states, resulting in a state transition probability matrix: in, Represents the state transition probability matrix. Indicates the system from its historical state i Transition to the target state j The probability of.

3. The predictive control method for a high-power hydrogen production converter according to claim 1, characterized in that, S2 includes: Forcibly applying the state transition probability matrix The fatal fall state in the data is reconstructed into an absorbing state, yielding the absorbing state transition matrix. ; Based on absorption state transition matrix Using the initial probability vector π(0) generated from the current sampling, the future probability vector is determined by the CK equation. N After one control cycle, the system's state distribution vector π( N ); Extract the state distribution vector π( N The numerical value of the corresponding absorption state in the ) yields the future N Warning signal that the country has fallen into a state of disaster for the first time. FPP .

4. The predictive control method for a high-power hydrogen production converter according to claim 3, characterized in that, The warning signal FPP The expression is as follows: in, Describing the absorption state transition matrix of N Power of 1.

5. The predictive control method for a high-power hydrogen production converter according to claim 1, characterized in that, S3 includes: Define target tracking weights that decay exponentially with the probability of first passage. Simultaneously define the dynamic damping coefficient that increases exponentially with the probability of first passage. λ 3; Based on dynamic damping coefficient λ 3. The improved MPC value function of the 6-phase interleaved Buck converter system was calculated. : in, This represents the steady-state damping reference constant of a high-power hydrogen production converter system during steady-state operation. This represents the damping surge gain constant. This represents the tracking error weighting coefficient. This represents the true expected target current reconstructed using probabilistic softening. Indicates the first k The parallel branch in the next n Predicted inductor current at +1 step. Indicates the first k The parallel branch is currently the first n The theoretical duty cycle to be solved within the frame. Indicates the first k The parallel branch is currently the first n The theoretical duty cycle to be solved within -1 beat; Using target tracking weights Reconstruct the true expected target current of the current shot. I aim And calculate the new tracking error e new This completes the zeroing process for error tracking power.

6. The predictive control method for a high-power hydrogen production converter according to claim 5, characterized in that, The tracking error e new The expression is as follows: in, Indicates the original target reference phase current. This represents the actual current feedback value currently sampled. This represents the sensitivity adjustment constant for adjusting the trigger steepness of a high-power hydrogen production converter system to enter a transient defense state.

7. The predictive control method for a high-power hydrogen production converter according to claim 5, characterized in that, S4 includes: Utilizing early warning signals FPP Pre-construct the expected feedforward voltage V in_expected ; Value function for improved MPC Perform a discrete expansion and set the partial derivatives... This yields a purely algebraic equation; where, Indicates the first k The parallel branch is currently the first n The theoretical duty cycle to be solved within the frame; Extracting the optimal duty cycle for high-power hydrogen production converters to withstand impact using pure algebraic equations D calc ; Optimal duty cycle D calc Latch and set the optimal duty cycle D calc This serves as a physical benchmark for constraining the single-step change rate of duty cycle in high-power hydrogen converters.

8. The predictive control method for a high-power hydrogen production converter according to claim 7, characterized in that, The optimal duty cycle D calc The expression is as follows: in, This represents the polarization voltage of the electrolytic cell. Indicates the filter inductance. This indicates the underlying discrete control cycle. This represents the true expected target current reconstructed using probabilistic softening. This represents the actual current feedback value currently sampled. This represents the actual physical bus voltage. This represents the estimated maximum busbar drop amplitude of a high-power hydrogen production converter system under extreme operating conditions.

9. The predictive control method for a high-power hydrogen production converter according to claim 5, characterized in that, S5 includes: Calculated target tracking weights Maximum permissible duty cycle single-step change rate for real-time dynamic linkage ; Optimal duty cycle D calc Using this as a constraint benchmark, and combining it with the actual duty cycle of the previous control cycle, the theoretical single-step change expected by the MPC algorithm in the current control cycle is calculated. ; Based on the single-step change rate of the maximum permissible duty cycle For theoretical single-step change Perform asymmetric hard constraint trimming with forced engineering orientation ; Asymmetric hard constraint clipping Calculate the final actual duty cycle It then issues these commands as control instructions to the underlying execution mechanisms.

10. The predictive control method for a high-power hydrogen production converter according to claim 1, characterized in that, The actual duty cycle The expression is as follows: in, This indicates the duty cycle that was latched and actually issued in the previous control cycle. This indicates the optimal duty cycle.

Images