Adaptive suppression method of current ripple of electrolytic cell driven by multiphase interleaved chopper

Abstract

This invention discloses an adaptive current ripple suppression method for an electrolytic cell driven by a multiphase interleaved chopper, relating to the field of power electronic conversion technology. The method includes: online identification of the frequency domain impedance characteristics of the electrolytic cell circuit at multiple harmonic frequency points; based on the frequency domain impedance characteristics, joint optimization of the interleaved phase shift and switching frequency of the multiphase interleaved chopper; obtaining optimal control parameters by minimizing a preset current ripple objective function; sending the optimal control parameters to the multiphase interleaved chopper for execution; and repeating the aforementioned online identification and joint optimization according to preset update conditions. This invention precisely targets current ripple from the traditional voltage ripple, achieving tightly coupled joint optimization of control parameters. It can dynamically adjust based on the real-time operating conditions of the electrolytic cell, achieving ripple suppression under all operating conditions.

Description

Technical Field

[0001] This invention belongs to the field of power electronic conversion technology, and in particular to an adaptive suppression method for current ripple in an electrolytic cell driven by a multiphase interleaved chopper. Background Technology

[0002] In large-scale electrolytic hydrogen production systems, high-quality direct current (DC) is required to supply the electrolyzer. Current ripple, as the alternating component of DC, not only fails to effectively produce hydrogen but also converts into heat, reducing the energy conversion efficiency of hydrogen production. Ripple of specific frequencies and amplitudes accelerates the degradation of electrode materials, shortening the electrolyzer's lifespan.

[0003] To achieve high-power DC supply and ripple suppression, existing technologies employ multi-phase interleaved parallel DC-DC converter topologies, such as multi-phase interleaved buck choppers. The basic working principle is to uniformly stagger the switching actions of multiple power units in time, so that the ripples generated by each unit cancel each other out at the output.

[0004] Applying existing ripple suppression methods directly to electrolytic cell loads leads to a series of problems. For example, due to the complex electrochemical characteristics and parasitic parameters of the electrolytic cell circuit, its impedance varies significantly with frequency, causing the optimal voltage ripple to no longer be equivalent to the optimal current ripple that affects performance. Therefore, further research and innovation are needed to address these issues in existing technologies. Summary of the Invention

[0005] Purpose of the invention: In view of the above-mentioned problems in the prior art, this application provides an adaptive suppression method for current ripple in a multiphase interleaved chopper driven electrolytic cell.

[0006] Technical solution: According to one aspect of this application, an adaptive suppression method for current ripple in a multiphase interleaved chopper-driven electrolytic cell is proposed, comprising:

[0007] Online identification of frequency domain impedance characteristics of electrolytic cell circuits at multiple harmonic frequency points;

[0008] Based on the frequency domain impedance characteristics, the interleaved phase shift and switching frequency of the multiphase interleaved chopper are jointly optimized, and the optimal control parameters are obtained by minimizing the preset current ripple objective function.

[0009] The optimal control parameters are sent to the multiphase interleaved chopper for execution, and the aforementioned online identification and joint optimization are repeated according to the preset update conditions.

[0010] Beneficial effects: This invention precisely targets current ripple from voltage ripple, enables tightly coupled joint optimization of control parameters, and allows dynamic adjustment based on the real-time operating conditions of the electrolytic cell, achieving ripple suppression under all operating conditions. Attached Figure Description

[0011] Figure 1 This is a flowchart of the adaptive suppression method for current ripple in a multiphase interleaved chopper driven electrolytic cell according to this application.

[0012] Figure 2 This is a flowchart illustrating the online identification of frequency domain impedance characteristics in this application.

[0013] Figure 3 This is a flowchart of the joint optimization for this application.

[0014] Figure 4 This is a flowchart of the trigger condition judgment process based on performance feedback in this application.

[0015] Figure 5 This is a flowchart of the timer-based trigger condition determination process in this application. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Through research and analysis, the applicant discovered that existing methods aim to minimize voltage ripple, but this requires the load impedance to remain essentially constant at each harmonic frequency. The applicant found that due to the electrochemical characteristics and parasitic parameters of the electrolytic cell circuit, the impedance varies with frequency, making optimal voltage ripple and optimal current ripple not equivalent.

[0019] Current methods treat staggered phase shift optimization and switching frequency adjustment as two independent, decoupled control problems, with switching frequency adjustment used to avoid resonance. In actual operation, changes in the switching frequency directly affect the impedance values ​​corresponding to each harmonic, altering the optimal phase shift value; conversely, phase shift adjustment redistributes harmonic energy distribution, affecting the selection of the optimal switching frequency. This artificial decoupling makes it difficult for the control strategy to converge to the global optimum, limiting ripple suppression performance.

[0020] To solve these problems, combined with Figures 1 to 5 The present invention will be specifically described through the following embodiments.

[0021] In this embodiment, an adaptive current ripple suppression method for a multiphase interleaved chopper-driven electrolytic cell is provided to address the problem that the interleaved parallel control strategy does not consider the dynamic changes in the frequency domain impedance of the electrolytic cell circuit. The method involves solving the objective function based on the identified frequency domain impedance characteristics, using the total energy of the output current ripple as the optimization objective and the inherent impedance of the circuit as the weighting coefficient. The interleaved phase shift and switching frequency of the chopper are used as joint decision variables to obtain the optimal control parameters. The optimal control parameters are then sent to the chopper for execution, and the process is repeated according to preset update conditions to form an adaptive closed loop.

[0022] Based on this technical method, a simulation system incorporating a multiphase interleaved chopper and an electrolytic cell load can be constructed. The electrolytic cell is modeled using an equivalent circuit model, such as the Landles model, to simulate its electrochemical impedance characteristics at different frequencies. Furthermore, this embodiment can directly identify the actual circuit impedance online without relying on a specific or preset electrolytic cell model.

[0023] The adaptive current ripple suppression method proposed in this application can be applied to an electrolytic hydrogen production power supply system composed of N-phase interleaved parallel buck choppers. This system mainly includes a DC input source, N parallel power switch arms, corresponding drive circuits, an output filter inductor, and a DC bus. The DC bus is used to connect the electrolyzer load, and voltage and current sensors are installed on the DC bus to collect the output waveform during system operation. The method is executed by a central controller, such as a digital signal processor or a field-programmable gate array (FPGA).

[0024] In this embodiment of the application, the adaptive current ripple suppression method includes:

[0025] Step 101: Identify the frequency domain impedance characteristics of the electrolytic cell circuit at multiple harmonic frequency points online.

[0026] In this system, the output of the multiphase interleaved chopper is connected to the electrolytic cell via a filter inductor to form the electrolytic cell circuit. In actual operation, the circuit impedance of the electrolytic cell is a complex impedance, the amplitude of which varies with frequency and is not purely resistive or a constant value. Its frequency dependence stems from the electrochemical impedance of the electrolytic cell, as well as parasitic inductances and capacitances from cables, busbars, etc. These parasitic parameters have a more pronounced impact at high frequencies. Therefore, to suppress current ripple, it is necessary to accurately obtain the impedance information of the electrolytic cell circuit at each harmonic frequency point of the switching ripple.

[0027] In this embodiment, the phase frequency domain impedance characteristics are represented as a discrete data sequence or lookup table, which stores multiple harmonic frequency points and their corresponding impedance amplitude mapping relationships.

[0028] Step 102: Based on the frequency domain impedance characteristics, jointly optimize the interleaved phase shift and switching frequency of the multiphase interleaved chopper, and obtain the optimal control parameters by minimizing the preset current ripple objective function.

[0029] In this embodiment, joint optimization refers to combining the interleaved phase shift vector θ with the switching frequency f. _sw As common decision variables, a unified objective function is solved. This objective function aims to minimize the output current ripple, rather than the voltage ripple, and outputs optimal control parameters. For example, this parameter set may include the optimal switching frequency value f. _sw * And the optimal N-dimensional staggered phase shift vector θ * N represents the number of parallel phases of the chopper. That is, by taking the receiver's frequency domain impedance characteristics as input, the previously independent problems of interleaved phase shift optimization and switching frequency selection are unified into a mathematical framework for joint optimization.

[0030] Step 103: Send the optimal control parameters to the multiphase interleaved chopper for execution, and repeat the online identification and joint optimization according to the update conditions.

[0031] Specifically, the optimal control parameters are sent to the pulse width modulation (PWM) module of the power supply to update the switching frequency and initial phase of each phase carrier, so that the chopper operates in a new and better control state. Since the operating state of the electrolyzer changes dynamically over time, such as temperature, pressure, current density, or electrode aging, its frequency domain impedance characteristics will change, so a single optimization is not enough.

[0032] Accordingly, by establishing a loop mechanism based on preset update conditions, the online identification and joint optimization are re-executed periodically or when necessary, so that the control parameters can track the changes in the electrolyzer's operating conditions in real time and suppress current ripple.

[0033] As a feasible solution, this embodiment provides an online identification method for the frequency domain impedance of an electrolytic cell circuit based on phase-shift perturbation. Utilizing the degrees of freedom controlled by a multi-phase interleaved chopper, i.e., the phase of each phase carrier, a small perturbation is applied to actively generate a controllable excitation signal. This method requires no additional hardware injection circuitry or incurs additional energy loss, and can identify the frequency domain impedance characteristics of the electrolytic cell circuit online.

[0034] For example, the method of injecting high-frequency sinusoidal or broadband perturbation signals into the system using an external signal generator and coupling circuit to obtain load impedance characteristics encounters several technical problems when applied to high-power electrolyzer scenarios, such as those operating at hundreds to thousands of amperes. Another example is that, to avoid affecting the normal electrolytic hydrogen production process, the injected perturbation signal amplitude must be very small. This requirement results in a low signal-to-noise ratio in the response signal, which is then overwhelmed by the switching ripple noise of the converter. Furthermore, injecting additional signals leads to non-DC energy loss, which is converted into heat, reducing the overall efficiency of the system.

[0035] Furthermore, the frequency domain impedance characteristics of the electrolytic cell circuit at multiple harmonic frequency points are identified online, specifically including:

[0036] Step 201: Apply a controlled micro-perturbation to the carrier phase shift of at least one phase in the multiphase interleaved chopper to actively generate measurable voltage and current responses at preset harmonic frequency points; wherein the micro-perturbation is a perturbation of preset amplitude.

[0037] Frequency domain impedance characteristics are obtained by frequency-by-frequency calculation based on voltage and current responses.

[0038] Specifically, for the identified phase shift cancellation harmonic frequency points, the controller selects one phase, for example, using phase 1 as the reference phase, and adjusts its carrier phase shift θ. _1 A controlled, small disturbance δ of preset amplitude is applied, meaning the amplitude of the disturbance does not exceed a preset proportion of the standard phase shift interval. For example, in an N-phase system, the phase shift interval is 2π / N, and the small disturbance δ is set to 2% to 5% of this interval. This small disturbance can break the original ideal cancellation conditions, causing voltage and current ripple components to reappear at the phase shift cancellation harmonic frequency points. These voltage and current ripple components are measured by sensors, i.e., the voltage response and current response. This process does not interfere with the macroscopic DC operating state of the electrolyzer.

[0039] Accordingly, after applying a phase shift perturbation, the system operates stably for a preset time window, such as 30 to 50 switching cycles, during which voltage and current waveform data at the output terminal are simultaneously acquired. Spectral analysis is then performed on the acquired voltage and current response waveforms. For each target harmonic frequency point f to be identified... _kThe complex amplitude V of the voltage harmonic at that frequency point is calculated using either the Discrete Fourier Transform or the more computationally efficient Goertzel algorithm. _k and the complex amplitude of current harmonics I _k , where f _k =k×f _sw The impedance amplitude at a given frequency can be calculated frequency-by-frequency using Ohm's law. The formula is as follows:

[0040] |Z _k ∣=∣V _k ∣ / ∣I _k |;

[0041] Where |…| represents the modulus or magnitude of the complex number, and |Z _k | is the impedance amplitude at the k-th harmonic frequency, |V _k | represents the voltage amplitude of the k-th harmonic extracted from the voltage response, |I _k | represents the current amplitude of the k-th harmonic extracted from the current response. Using steady-state identification and perturbation identification, the impedance amplitudes at all frequency points are obtained, merged, and formed a complete frequency domain impedance characteristic sequence covering the first harmonic to the preset highest harmonic K.

[0042] Step 202: Distinguish between steady-state residual harmonic frequency points and phase-shift canceled harmonic frequency points among multiple harmonic frequency points; wherein, for residual harmonic frequency points, extract voltage and current harmonic components from the output waveform of the multiphase interleaved chopper in steady-state operation to calculate impedance;

[0043] For phase shift cancellation harmonic frequency points, impedance is identified by applying a small perturbation.

[0044] Specifically, the harmonic frequency points that need to be identified are classified. When the N-phase interleaved chopper uses a standard equal-spaced phase shift, that is, the phase shift θ of the j-th phase... _j When set to 2π(j-1) / N, the voltage ripple spectrum of its superimposed output exhibits sparse characteristics. This sparseness only occurs at the switching frequency f. _sw At the Nth harmonic and its integer multiples, i.e., k=m×N, where m is a positive integer, the ripple will not be canceled, and this frequency point is the steady-state residual harmonic frequency point.

[0045] For such frequency points, when the system is working normally, it will continuously generate excitation signals. The steady-state waveform can be collected by the voltage and current sensors at this time, and the voltage and current harmonic amplitudes at the corresponding frequency can be extracted by spectrum analysis to calculate the impedance.

[0046] Correspondingly, for other harmonic frequencies other than the Nth harmonic, such as those from the 1st to the (N-1)th harmonic, the theoretical output ripple amplitude is zero under equal-spaced phase shifts. These frequencies are the phase-shift cancelling harmonic frequencies. At these frequencies, due to the lack of an excitation signal, it is difficult to directly identify the impedance; therefore, an active method is used to generate the excitation. This classification method introduces active excitation only when necessary to improve identification efficiency.

[0047] In another possible implementation, online identification further includes a credibility verification of the identification results, which includes:

[0048] Calculate in advance the theoretical voltage response amplitude induced by a small disturbance at the harmonic frequency point; compare the theoretical voltage response amplitude with the measured amplitude of the measurable voltage response;

[0049] When the deviation between the two exceeds the threshold, the identification result of the corresponding harmonic frequency point is marked as low confidence, and its weight is reduced in the joint optimization.

[0050] In real-world physical systems, the measurement process is affected by noise, sensor nonlinearity errors, and interference from unmodeled dynamics, causing deviations in the identification results for some frequency points. To identify unreliable data points, a reliability check is introduced.

[0051] Specifically, before applying a small disturbance δ, the controller determines the system parameters, including the input voltage V. _in The duty cycle D and the disturbance amount δ are used to pre-calculate the theoretical voltage response amplitude V corresponding to the disturbance at the k-th harmonic frequency. _theory,_k This theoretical value is pre-obtained based on the Fourier analysis model of the chopper. The measured voltage response amplitude V is then obtained. _meas,_k Then, the relative deviation between the two is calculated.

[0052] For example, the deviation rate is calculated as Deviation. _k =abs(V _meas,_k -V _theory,_k ) / V _theory,_k The deviation rate is compared with a preset threshold, which is set based on experience or system requirements, for example, 20%. When Deviation... _k If the percentage exceeds 20%, the identification result at that frequency point is deemed unreliable and marked as low confidence. For frequency points marked as low confidence, the impedance weighting coefficient is multiplied by a penalty factor less than 1, such as 0.5, to reduce the negative impact of the impedance identification value on the optimization results.

[0053] As an alternative implementation, a method for constructing an impedance-weighted current ripple objective function is also provided.

[0054] Common multiphase interleaved choppers optimize by minimizing output voltage ripple, suitable for scenarios where the load is an ideal resistor or the impedance is essentially constant within the harmonic frequency range. However, the impedance of the electrolytic cell circuit is frequency-dependent. For example, a small voltage ripple may occur at a frequency that corresponds to a low impedance point in the circuit, such as the LC resonant point, still resulting in a large current ripple. The actual current ripple flowing into the electrolytic cell, not the voltage ripple at the power supply port, affects electrolysis efficiency and electrode life. Therefore, an optimization model is constructed with minimizing current ripple as the objective.

[0055] Specifically, for an N-phase interleaved chopper, given the interleaved phase shift vector θ and the switching frequency f _sw The amplitude of the k-th harmonic ripple of the output voltage is obtained through Fourier analysis. This voltage ripple passes through an impedance of Z. _k After the electrolytic cell circuit is completed, the current ripple amplitude is the quotient of the voltage ripple amplitude and the impedance amplitude, and its calculation formula is:

[0056] I _ripple,_k =V _ripple,_k / ∣Z _k |;

[0057] Among them, I _ripple,_k V represents the current ripple amplitude of the k-th harmonic. _ripple,_k Voltage ripple of the kth harmonic

[0058] Amplitude, |Z _k | represents the impedance amplitude of the circuit at the kth harmonic frequency.

[0059] To minimize the total energy of the current ripple generated by all harmonic components, which is proportional to the sum of the squares of the amplitudes of all current ripple harmonics, the joint optimization includes:

[0060] Step 301: Based on the frequency domain impedance characteristics, calculate the corresponding impedance weighting coefficient for each frequency point in the harmonic frequency range; establish a current ripple objective function with the impedance weighting coefficient as the weight and minimizing the total energy of the output current ripple.

[0061] Accordingly, in order to minimize the total energy of the current ripple, i.e. minimize Σ∣I _ripple,_k | 2 Based on the calculation formula for current ripple amplitude, the optimization objective is to minimize Σ∣V _ripple,_k | 2 / ∣Z _k | 2Therefore, the contribution of each harmonic to the total current ripple depends not only on its voltage ripple amplitude but also on the weighted effect of the square of the impedance amplitude at the corresponding frequency. That is, the lower the impedance at a frequency, the greater its weight and the higher its optimization priority. Therefore, the impedance weighting coefficient introduced in this embodiment can be used to quantify the importance of each harmonic. The current ripple objective function is no longer a mathematical indicator but a physical function directly related to the total energy of the current ripple.

[0062] Step 302, the impedance weighting coefficient is calculated as follows:

[0063] Based on the input voltage and duty cycle of the multiphase interleaved chopper, the inherent voltage ripple amplitude at each harmonic frequency is obtained.

[0064] This makes the impedance weighting coefficient proportional to the square of the inherent voltage ripple amplitude at the corresponding harmonic frequency, and inversely proportional to the square of the impedance amplitude of the electrolytic cell circuit at that harmonic frequency.

[0065] In some alternative implementations, the impedance weighting factor w is invoked. _k The calculation formula is:

[0066] w _k =(V _in 2 ×sin 2 (k×π×D)) / (k 2 ×π 2 ×∣Z _k | 2 ); where w _k V is the impedance weighting factor for the k-th harmonic. _in The input DC voltage of the chopper, sin 2 (•)for

[0067] The square of the sine function, k is the harmonic order, D is the duty cycle of the chopper, |Z _k | represents the impedance amplitude of the electrolytic cell circuit at the kth harmonic frequency, obtained through identification.

[0068] Specifically, the numerator (excluding the constant) of this impedance weighting formula is proportional to the square of the inherent voltage ripple amplitude generated by the chopper at the k-th harmonic, and can represent the source strength of the voltage excitation at that frequency. _k | 2 This indicates the circuit's ability to suppress the excitation. Therefore, w _k The larger the value, the stronger the voltage excitation at that frequency point, or the weaker the circuit's suppression capability, or both, resulting in a greater contribution to the total current ripple.

[0069] Accordingly, after calculating all target harmonic orders, for example, the impedance weighting coefficient sequence from k=1 to K,

[0070] A current ripple objective function J can be constructed, which takes the staggered phase shift vector θ and the switching frequency f_sw as variables and is expressed as the sum of the weighted contributions of all harmonic components. The formula for the objective function J is:

[0071] J(θ,f _sw )=Σ _{k=1 to K} w _k ×∣S(θ,k)∣ 2 ;

[0072] Where J(θ,f) _sw Let Σ be the objective function to be minimized, and let Σ represent the summation over all harmonic orders k from 1 to K.

[0073] w _k The calculated value is related to the switching frequency f. _sw The relevant impedance weighting coefficients, S(θ,k) are phase shift superposition factors, defined as S(θ,k)=Σ _{j=1 to N} e (i×k×θ_j) Where N is the number of phases, θ _j Let be the phase shift of the j-th phase, i be the imaginary unit, and e be the base of the natural logarithm. The square of the modulus of the phase shift superposition factor S(θ,k) |S(θ,k)| 2 It characterizes the effect of N-phase interleaved superposition on voltage ripple at the kth harmonic frequency.

[0074] In this embodiment, the objective function J shifts the optimization objective from voltage ripple to current ripple through impedance weighting coefficients, and the switching frequency f... _sw Through the impedance term |Z _k | Endogenously coupled into the optimization model, it can serve as the basis for joint optimization.

[0075] As an optional implementation, this embodiment also provides a hierarchical solution algorithm for solving non-convex optimization problems based on interleaved phase shift and switching frequency as joint decision variables.

[0076] Specifically, because the phase shift vector θ is a continuous multidimensional variable, for the objective function J(θ,f) _sw A direct global brute-force search is quite difficult. Furthermore, the switching frequency f... _sw Through impedance Z _k The objective function is strongly coupled, and simple alternating optimization cannot converge to the global optimum. Therefore, this embodiment designs a hierarchical solution strategy, decomposing the complex problem into multiple easily manageable subproblems, specifically including:

[0077] Step 401: Joint optimization is performed using a hierarchical strategy that combines outer layer switching frequency discrete search with inner layer phase shift optimization.

[0078] Specifically, the outer loop processes the switching frequency f. _sw Due to hardware and electromagnetic design constraints, the switching frequencies exhibit a discrete finite set; therefore, an traversal search is employed. The inner loop processes the interleaved phase shift vector θ for each fixed switching frequency f. _sw The dimensionality of the phase-shift optimization problem is reduced, allowing it to be solved using continuous optimization algorithms. In this embodiment, two different types of decision variables are decoupled, and a nested hierarchical strategy is used to achieve the solution, transforming the complex joint optimization problem into a phase-shift optimization subproblem that can be solved in a finite time.

[0079] Step 402, the outer layer switching frequency discrete search specifically includes: traversing each candidate switching frequency in the candidate switching frequency set; for each candidate switching frequency, based on the frequency domain impedance characteristics, calculating the impedance value corresponding to each harmonic frequency point under the candidate switching frequency through interpolation, generating the corresponding impedance weighting coefficient, which is used for inner layer phase shift optimization.

[0080] In this embodiment, during interpolation estimation, it is necessary to ensure that the physical frequencies corresponding to each harmonic fall within the frequency coverage range of the identified impedance sequence at each candidate switching frequency to make the interpolation effective. If some harmonic frequencies of a candidate frequency exceed the identified range, the upper and lower limits of the impedance identification frequency can be extended accordingly, or a monotonic extrapolation estimation of the impedance amplitude can be used for the frequency points exceeding the range, and a low confidence mark can be added to the estimation result. For example, an appropriate extrapolation method can be selected based on the frequency variation trend of the actual impedance curve.

[0081] Specifically, a set of candidate switching frequencies is predefined. For example, based on the switching loss characteristics of the system's power devices and the design margin of the magnetic components, the switching frequency range is 50kHz to 70kHz, and this set is set to include a set of 5 discrete frequency values ​​{50kHz, 55kHz, 60kHz, 65kHz, 70kHz}.

[0082] For each switching frequency f in this set _sw,_i Its harmonic frequencies k×f _sw,_i The reference frequency k×f used for impedance identification _sw,_0 The impedances are not completely overlapped, therefore the original frequency domain impedance characteristics data cannot be used. Based on this, this embodiment uses interpolation to estimate the impedance value at the new frequency point; for example, the output discrete impedance sequence (f...) is used to... _k ,∣Z _k Using known data points, linear interpolation or spline interpolation is employed to estimate the value of k×f at the new harmonic frequency point. _sw,_i The impedance amplitude at that point.

[0083] After obtaining the current candidate frequency f _sw,_i The corresponding complete impedance spectrum is used to calculate and generate the impedance weighting coefficient w specific to this candidate frequency. _k This process ensures that the phase shift optimization of the inner layer is based on the actual physical characteristics at the current frequency.

[0084] Step 403, inner layer phase shift optimization, is specifically an iterative process, which includes:

[0085] Start the iteration using the initial phase shift vector as the current phase shift vector;

[0086] In each iteration, the gradient of the current ripple objective function with respect to each adjustable phase shift component is analytically calculated based on the dedicated impedance weighting coefficients and the current phase shift vector.

[0087] Based on the calculated gradient, determine the update direction and step size of the phase shift vector, and update the current phase shift vector.

[0088] Repeatedly analyze the gradient and phase shift vector updates until the change in the current ripple objective function value is less than the preset convergence threshold or the maximum number of iterations is reached. The converged phase shift vector is then used as the output of this optimization.

[0089] In the N-phase system, the first phase is selected as the reference phase, and its phase shift is θ. _1 It is fixed at zero and does not participate in the optimization iteration. The adjustable phase shift component is the θ corresponding to the 2nd to Nth phase. _2 θ _3 , …, θ _N There are N-1 possible phase shifts. The optimal phase shift vector θ is output. ∗ It is an N-dimensional vector, with the first component always being 0, and the remaining N-1 components being the optimization results.

[0090] Specifically, this iterative process is fully executed for each candidate frequency and its corresponding weighting coefficient passed in from the outer loop. One or more initial phase shift vectors can be set. During the phase shift vector update process, the steepest descent method can be used for phase shift updating, i.e., θ _m (new) =θ _m (old) -η×(ΦJ / Φθ _m ), where η is a preset small positive number, called the learning rate or step size, and Φ is the partial derivative. Optimization algorithms such as the conjugate gradient method and the quasi-Newton method can also be used to accelerate the convergence speed.

[0091] To avoid the gradient magnitude affecting convergence due to large changes in system parameters, the gradient descent method can be implemented by using a normalized gradient step size, i.e., normalizing the gradient vector before each iteration; or by using an automatic step size search, which automatically determines the step size that satisfies the conditions of sufficient descent and curvature by performing a line search along the negative gradient direction based on the Wolf condition.

[0092] In some scenarios, the initial value of the learning rate η is set between 0.01 and 0.1, and is adaptively adjusted during iterations. For example, if the objective function value decreases in two consecutive iterations, η is increased to 1.2 times the original value to accelerate convergence; when the objective function value increases, η is decreased to 0.5 times the original value to improve stability.

[0093] As an alternative, the conjugate gradient method or the quasi-Newton method can be used instead of the steepest descent method to improve convergence speed and numerical stability. In this case, the step size can be automatically determined by the Wolfe condition. When repeatedly updating the analytical gradient and phase shift vector, the convergence threshold can be set to a small positive number, such as 10. -6 The maximum number of iterations can be set to 100 or 200 to prevent the algorithm from getting stuck in an infinite loop.

[0094] Step 404, the analytical gradient is calculated as follows: for any adjustable phase shift component, its gradient is calculated as the sum of all harmonic orders; where the contribution of each harmonic order is proportional to the impedance weighting coefficient of that order, the order value, and the product of the sine function of the phase shift difference between the adjustable phase shift component and all other phases at the corresponding order multiple.

[0095] Specifically, for the objective function J(θ,f) _sw The phase shift component θ of the m-th phase _m Taking the partial derivative, we obtain its analytical gradient expression:

[0096] ΦJ / Φθ _m =-2×Σ _(k=1 to K) k×w _k ×Σ _(j≠m，j=1 to N) sin(k×(θ _m -θ _j ));

[0097] Wherein, ΦJ / Φθ _m For the objective function J, the phase shift component θ _m The partial derivatives, i.e., the gradient components, Σ _(k=1 to K) express

[0098] Summing over all harmonic orders k from 1 to the preset highest order K, w _k Σ represents the k-th order impedance weighting coefficient at the current candidate frequency, where N is the total number of phases. _(j≠m，j=1toN)This represents the summation over all phases j except the m-th phase, where sin is the sine function, (θ) _m -θ _j Let be the phase shift difference between phase m and phase j. This expression can directly calculate the gradient, which is faster than the numerical difference method for approximate estimation.

[0099] Step 405: The iterative process of inner phase shift optimization is executed multiple times, each time using a different initial phase shift vector selected from a preset set of initial phase shift vectors;

[0100] Furthermore, after all execution iterations have converged, the current ripple objective function values ​​corresponding to the optimization output results of each iteration are compared, and the output result that minimizes the objective function value is selected as the optimal phase shift vector for the current candidate switching frequency.

[0101] The objective function J is a complex trigonometric polynomial with respect to phase shift θ, exhibiting multiple local minima. Gradient descent from a single initial point tends to converge to a local optimum. Therefore, this embodiment pre-defines multiple initial phase shift vectors to increase the probability of finding the global optimum. A pre-defined set of initial phase shift vectors is included, containing standard, equally spaced phase shift vectors, typically representing the optimal solution for voltage ripple and serving as a good benchmark for optimization. A small random perturbation is applied to the equally spaced phase shifts to generate 2 to 3 sets of candidate initial vectors. In this embodiment, a multi-starting-point optimization strategy is introduced to address the non-convexity of the objective function.

[0102] Furthermore, the iterative optimization algorithm starts from each of the pre-defined initial phase shift vectors and independently performs the optimization process to obtain multiple local optima. The solution that minimizes the current ripple objective function value is selected as the optimal phase shift vector. In other words, the inner-layer optimization starts from each initial vector in the set and completes the iterative optimization process separately, obtaining multiple local optima. These local optima are then substituted into the objective function J to calculate the corresponding objective function value. The solution that minimizes the objective function J is selected as the optimal phase shift vector at the current candidate switching frequency.

[0103] This embodiment, through the synergy of outer-layer frequency search and inner-layer multi-starting-point gradient optimization, can find the globally optimal solution (f) that minimizes the actual current ripple among all candidate control parameters. _sw * θ * ).

[0104] Alternatively, by exhaustively searching for candidate frequencies in the outer layer and optimizing phase shifts at multiple starting points in the inner layer, the combination that minimizes the objective function value is selected from the phase shift optimization results corresponding to all candidate switching frequencies, and the result is output. Since the inner objective function is a non-convex periodic trigonometric polynomial, the multi-starting-point strategy can increase the probability of finding the global optimum. In application scenarios with a low number of phases, such as N=3 to 6, this strategy can approximate or reach the global optimum. In this embodiment, the number of multiple starting points can also be increased according to actual needs.

[0105] According to one aspect of this application, the triggering and execution of an adaptive closed loop are also provided, illustrating the transformation from static optimal parameter settings to an adaptive control system that can respond to changes in operating conditions in real time.

[0106] Specifically, it outputs the globally optimal control parameters, i.e., the optimal switching frequency f. _sw * and the optimal phase shift vector θ * The controller writes this information into the pulse width modulation (PWM) generation module. The frequency of each phase carrier is updated to f. _sw * Furthermore, the initial count value of each phase carrier, or the phase register is set according to the corresponding component of θ, to realize the control of the chopper switching timing at the hardware level.

[0107] To avoid current or voltage surges caused by sudden changes in control parameters, a smooth transition strategy is adopted for parameter switching. For example, the phase shift vector does not jump directly from the old value to the new value, but gradually transitions to the optimal value θ over several or dozens of switching cycles through linear interpolation. * The switching frequency is also synchronized and coordinated to ensure stable system operation.

[0108] After the parameters are updated, the system immediately enters the closed-loop monitoring and update trigger judgment phase. Based on this, this embodiment also provides two parallel triggering mechanisms, specifically including:

[0109] Step 501, the preset update conditions include performance feedback-based trigger conditions, and the determination process for these trigger conditions includes:

[0110] After the optimal control parameters are issued and executed, the real-time output current waveform of the chopper is continuously collected; based on the real-time output current waveform, the measured current ripple index is calculated; the measured current ripple index is compared with the preset deterioration threshold, and when the index exceeds the threshold, repeated execution is triggered.

[0111] In some scenarios, the system needs to rapidly adjust the output current density, or when the temperature and pressure of the electrolytic cell change abruptly, its impedance characteristics will change accordingly, rendering the original optimal parameters inapplicable. Therefore, during continuous operation of the controller, a current sensor is installed on the DC bus to collect the current waveform data stream in real time.

[0112] The controller's monitoring module analyzes the data at a relatively high frequency to calculate the measured current ripple index, for example, every few hundred switching cycles. This index can be the peak value of the current ripple or the RMS value of the current after removing the DC component; the RMS value better reflects the energy of the ripple.

[0113] Correspondingly, the preset degradation threshold can also take various forms. For example, the threshold can be a fixed absolute value, set according to the upper limit of the electrolyzer's tolerance to ripple. The threshold can also be a relative value. After each optimization, the system records the measured current ripple index under the current optimal state as a baseline value, and sets the degradation threshold to 130% of this baseline value. When the subsequently monitored index exceeds this threshold, it indicates that the system performance has deteriorated. At this time, the controller needs to be immediately triggered to re-execute and calculate the optimal parameters adapted to the current new operating conditions.

[0114] Step 502, the preset update conditions include timer-based trigger conditions, and the determination process for these trigger conditions includes:

[0115] Monitor the time interval since the last online identification and joint optimization was completed;

[0116] When the time interval reaches the preset periodic update time limit, the process is triggered to repeat.

[0117] This process employs a time-driven, periodic update mechanism, enabling the system to track and adapt to the slow, gradual changes in the electrolyzer's operating conditions. These changes arise from factors such as the slow drift in electrolyte temperature and electrode aging. Although the impedance characteristics caused by these factors change slowly, over time, the accumulated effects can cause the control parameters to gradually deviate from their optimal values. Therefore, periodic proactive optimization is necessary to maintain the system operating in its optimal state.

[0118] Specifically, the controller integrates a timer or counter, which resets and restarts after each identification and optimization process is completed. The controller continuously monitors the duration recorded by the timer in the background. The preset periodic update time limit can be configured based on experience or the characteristics of the electrolyzer, with the time limit set between 1 and 10 seconds. This time limit is faster than the typical time constant of changes in operating conditions such as electrolyzer temperature, allowing for proactive parameter re-optimization before operating conditions drift. When the preset time limit is reached, the controller forcibly executes a new round of identification and optimization processes, regardless of whether performance degradation conditions are triggered.

[0119] In other words, the two triggering conditions take effect in parallel within the system. The event-triggered mechanism based on performance feedback can quickly respond to sudden changes in system operating conditions, while the periodic triggering mechanism based on timers can continuously track the slow drift of gradual operating conditions. The two complement each other and can suppress current ripple to an optimal level in various complex and dynamic industrial environments.

[0120] Furthermore, this embodiment also includes anomaly handling strategies. For example, during online identification, if the frequency points marked as low confidence exceed a preset proportion, such as exceeding 50% of the total number of identified frequency points, the system determines that the current identification result is unreliable, abandons the current identification result, and re-identifies in the next update cycle. For example, if the inner phase shift optimization reaches the maximum number of iterations at all multiple starting points but still does not meet the convergence condition, the system selects the set with the smallest objective function value as the approximate optimal solution, and shortens the time interval of the next periodic update to allow for re-optimization. As another example, when a voltage or current sensor malfunctions, causing abnormal sampling data, such as a reading that is always zero or exceeds the limit, the controller freezes the current control parameters and issues a fault alarm signal.

[0121] In some scenarios, a five-phase interleaved parallel electrolytic hydrogen production power supply is set up, currently operating stably at a switching frequency of 60 kHz and a standard equal-spaced phase shift, with each phase spaced 72° apart, supplying 1000 A of DC power to the electrolyzer. Based on this, the processing flow using the adaptive suppression method proposed in this application includes:

[0122] Initially, the system only knows the impedance at the steady-state residual harmonic point of 300kHz (5×60kHz). At this point, the controller decides to perform an identification.

[0123] Applying a small, known perturbation to the carrier phase shift of phase 1, such as increasing it by 3.6° (5% of the 72° phase shift interval), instantly breaks the ideal harmonic cancellation conditions. This causes measurable, weak voltage and current ripples to appear at previously canceled frequency points such as 60kHz, 120kHz, 180kHz, and 240kHz.

[0124] Accordingly, a high-speed sensor is mounted on the output busbar to acquire the waveform of the new ripple component. The controller executes the Goertzel algorithm on the acquired data to extract the complex amplitudes of voltage and current at frequency points, and calculates the impedance amplitude frequency by frequency, for example, |Z. _60kHz ∣=∣V _60kHz ∣ / ∣I _60kHz |, where the Gosser algorithm only calculates the spectrum at specific frequency points.

[0125] Next, the controller performs a reliability check. The voltage ripple corresponding to a 3.6° phase shift disturbance is theoretically calculated based on the system parameters and compared with the measured value. If the measured value at 180kHz deviates from the theoretical value by more than 20%, Z_ 180kHz Data points are marked as low confidence. The output is a lookup table-like data structure containing at least (60kHz, 0.5Ω, high confidence), (120kHz, 1.2Ω, high confidence), (180kHz, 0.8Ω, low confidence), etc.

[0126] Furthermore, the controller acquires the real-time impedance spectrum and proceeds to the decision-making stage. Its objective is to minimize the actual current ripple, not the voltage ripple.

[0127] The controller has a preset set of candidate switching frequencies, such as {55kHz, 60kHz, 65kHz}. Each candidate frequency is evaluated sequentially. Taking 65kHz as an example, the controller uses an impedance lookup table and linear interpolation to estimate the impedance values ​​at new harmonic frequency points such as 65kHz, 130kHz, and 195kHz.

[0128] Calculate the impedance weighting coefficient w corresponding to each harmonic frequency point. _k If the impedance estimate at a certain frequency (e.g., 195kHz) is particularly low, then the corresponding W at that point... _k It will increase.

[0129] The controller initiates the gradient descent algorithm. Starting from multiple initial phase shift vectors (including standard equally spaced phase shifts), it iterates efficiently based on the analytical gradient formula. This is due to the weight w at the 195kHz point. _195kHz With a large proportion, the optimization algorithm will automatically adjust the phase shift of each phase to suppress the ripple depth at 195kHz.

[0130] After iterative convergence, the optimal phase shift vector at a switching frequency of 65kHz is obtained. It is usually a non-equally spaced vector, such as {0°, 70°, 145°, 218°, 288°}.

[0131] The controller repeats the above optimization process for all candidate frequencies (55kHz, 60kHz) in the set. It compares the minimum objective function values ​​corresponding to all candidate frequencies and determines the combination of 65kHz and its corresponding non-equidistant phase shift to minimize the theoretical total current ripple energy.

[0132] Next, the controller outputs the optimal combination {f} _sw * =65kHz, θ *= (0°, 70°, 145°, 218°, 288°)} is the decision result, and the PWM module parameters are updated through a smooth transition. The system operates in the new state, and a 10-second periodic timer is started. After the timer expires, regardless of the system's operating state, the controller will automatically return to the first step and re-execute the complete online identification process to adapt to possible slow changes in the electrolytic cell, such as impedance drift caused by temperature rise.

[0133] As an alternative implementation, this application also provides another alternative solution, specifically including:

[0134] Using equal-spaced phase shifts and a preset switching frequency, the N-phase interleaved buck chopper is started to drive the electrolytic cell to achieve steady-state operation at the target current. The output voltage waveform and output current waveform under steady-state conditions are collected as the basis data for subsequent impedance identification.

[0135] Specifically, based on the rated operating conditions of the electrolyzer, the phase shifts of each phase carrier of the N-phase interleaved Buck chopper are configured to be equally spaced, that is, the initial phase shift of the j-th phase is taken as 2π(j-1) / N, where j=1,2,...,N, and the initial value of the switching frequency is set to f. _sw 0 The value range is determined based on the design parameters of the magnetic components and power devices. The duty cycle D is determined based on the ratio of the target output voltage to the input voltage, and the input voltage V is... _in As known system parameters, each phase chopper is driven to start sequentially and gradually increase its current until the output current reaches the target value and stabilizes. Based on this, the system uses an initial phase shift vector θ. 0 (Equal spacing) and initial switching frequency f _sw 0 It operates in a steady state.

[0136] During steady-state operation of the electrolytic cell, continuous M-series data are synchronously collected using voltage and Hall current sensors on the DC output bus at a sampling rate no less than 10 times the switching frequency. _0 The output voltage waveform v within one switching cycle _out (n) and current waveform i _out (n), where n is the discrete sampling point index. The acquisition window length M _0 No fewer than 50 switching cycles. After data acquisition is complete, verify the signal amplitude to confirm there is no truncation or saturation, and then buffer the set of waveform data.

[0137] In another embodiment, the sampling rate should be no less than twice the physical frequency corresponding to the set highest recognizable harmonic order K, i.e., f _s ≥2Kf _swThis satisfies the Nyquist sampling theorem and avoids spectral aliasing caused by higher harmonics. For configurations with K=2N to 3N (e.g., K=10 to 15 in a 5-phase system), the sampling rate needs to be no less than 20 to 30 times the switching frequency.

[0138] Furthermore, based on the current steady-state operation of the interleaved chopper, a controlled small perturbation is applied to the carrier phase shift of a specified phase. Utilizing the controllable change in the output ripple spectrum induced by this perturbation, amplitude information at each target harmonic frequency is extracted from the acquired voltage and current response signals. Frequency-by-frequency calculations are then performed to obtain the frequency-domain impedance amplitude sequence of the electrolytic cell circuit at discrete frequency points. Impedance identification is completed entirely using the chopper's own switching actions without any external excitation signal injection. Specifically, this includes:

[0139] Based on the harmonic coverage required for joint optimization, the highest identifiable harmonic order is set to K, typically ranging from 2N to 3N, where N is the number of interleaved phases. This determines the target frequency set for the impedance to be identified, which includes all harmonic frequencies from the 1st to the Kth order, i.e., fk. _sw 0 2f _sw 0 Until Kf _sw 0 .

[0140] The set is divided into two categories. The first category consists of steady-state residual harmonic frequencies, corresponding to orders that are integer multiples of N, such as N, 2N, etc. Under equal-interval phase shifts, these harmonics are not canceled out, and their amplitude information can be directly extracted from the steady-state waveform. The second category consists of phase-shift canceled harmonic frequencies, corresponding to orders that are not integer multiples of N, i.e., orders from 1 to K excluding integer multiples of N. Under equal-interval phase shifts, these harmonics are completely canceled out, and the output amplitude is zero. They need to be re-emerged through phase-shift perturbations to be identifiable.

[0141] To identify the second type of frequency point, a set of perturbation parameters is set, in which the first phase is selected as the object to be perturbed (with its phase shift as the reference). The phase shift perturbation amount δ is taken as 2% to 5% of the equal-spaced phase shift interval, that is, δ is taken as 0.02×(2π / N) to 0.05×(2π / N). This range allows the ripple caused by the perturbation to be distinguished by the sensor without interfering with the electrolysis process. The data acquisition window length M during the perturbation period is set to ≥30 switching cycles, of which the first part is used for transient establishment and the second part is used for steady-state signal extraction.

[0142] Next, read the output voltage waveform v of the buffer. _out (n) and current waveform i _out(n) Perform spectrum extraction on each of the first type of steady-state residual harmonic frequencies. Specifically, for each target frequency (corresponding to orders N, 2N, etc.), the Goertzel algorithm is used to extract the corresponding complex spectral coefficients from the voltage and current waveforms, and their magnitudes are taken to obtain the corresponding voltage harmonic amplitude and current harmonic amplitude. The ratio of the two is the loop impedance amplitude at that frequency. The calculation is completed for all first-type frequency points, forming a steady-state impedance subsequence that covers all target frequency points whose orders are integer multiples of N.

[0143] While maintaining the carrier phase shifts of phases 2 through N unchanged, the phase shift of phase 1 is increased by a perturbation δ from its original value. This phase shift adjustment is achieved by modifying the carrier comparison reference value of phase 1 in the pulse width modulation module, and switching at the carrier valley moment to avoid current jumps. After applying the perturbation, the system continues to run for M switching cycles under the new phase shift configuration. The first 10 cycles are discarded, and the perturbation voltage response waveform v is synchronously acquired in the remaining cycles. _pert (n) and the perturbation current response waveform i _pert (n). After the acquisition is completed, the carrier phase shift of the first phase is immediately restored to its original value, and the system returns to the state of equal-spaced phase shift operation.

[0144] Based on this, the collected perturbation voltage response waveform v _pert (n) and the perturbation current response waveform i _pert (n) Perform spectrum extraction at the second-type harmonic frequencies (non-integer multiples of N). For each target frequency, the Goertzel algorithm is used to calculate the complex spectral coefficients of the voltage and current waveforms, and the modulus values ​​are taken to obtain the corresponding perturbation voltage and current harmonic amplitudes. Calculations are performed on all second-type frequency points to form perturbation voltage amplitude sequences and perturbation current amplitude sequences, with the elements of the two sequences corresponding one-to-one according to harmonic order.

[0145] Based on the perturbation δ and system parameters, the theoretical voltage amplitude induced by the perturbation at the k-th harmonic is pre-calculated.

[0146] V _in ×|sin(k×π×D)|×δ / π;

[0147] The measured microvalues ​​are compared with the theoretical values. If the deviation at a certain frequency point exceeds 20%, it is marked as a low-confidence identification point, and its value is reduced when calculating the weighting coefficients.

[0148] In another embodiment, the theoretical voltage response amplitude v should be determined in advance before performing the confidence check. _theory,k Whether it approaches zero, i.e., |sin(k×π×D)| is less than the preset minimum value ε _0For example, a value of 0.05. If this condition is met, it indicates that the excitation signal at this harmonic is too weak to be effectively identified under the current duty cycle. The corresponding frequency point is marked as low confidence and processed with the minimum weight in the joint optimization, and the deviation rate calculation is no longer performed. In this embodiment, the threshold can also be reasonably set according to the sensor noise level and signal resolution.

[0149] For each second-type frequency point, the amplitude of the perturbation voltage harmonic at that frequency is divided by the amplitude of the corresponding perturbation current harmonic to obtain the loop impedance amplitude at that frequency. That is, at frequency k×f... _sw 0 At this point, k is a non-integer multiple of N, and the impedance amplitude is equal to the ratio of the voltage harmonic amplitude to the current harmonic amplitude.

[0150] For low-confidence identification points, a confidence level label is added to their impedance values. After performing division calculations on all second-type frequency points one by one, a perturbation impedance subsequence covering non-integer multiples of N is formed.

[0151] The steady-state impedance subsequence (covering frequency points whose order is an integer multiple of N) and the perturbation impedance subsequence (covering frequency points whose order is not an integer multiple of N) are merged in ascending order of harmonic order to form a frequency domain impedance amplitude sequence {|Z} with continuous coverage from order 1 to order K. _1 |,|Z _2 |,...,|Z _K |}, where the subscript k represents the impedance amplitude corresponding to the kth harmonic, and the physical frequency is kf. _sw 0 This sequence fully characterizes the impedance characteristics of the electrolyzer circuit in the switching ripple correlation frequency band.

[0152] Furthermore, the frequency domain impedance amplitude sequence is read, and the impedance weighting coefficients for each harmonic order are constructed in combination with the current duty cycle and input voltage. Then, an optimization model is established with the output current ripple weighted effective value as the optimization objective, the phase shift vector and the switching frequency as joint decision variables. The model is solved by a hierarchical strategy that combines the outer layer switching frequency discrete search with the inner layer phase shift gradient optimization, and the optimal phase shift vector and the optimal switching frequency are output.

[0153] Specifically, the frequency domain impedance amplitude sequence {|Z} is read. _1 |,|Z _2 |,...,|Z _K |}, combined with input voltage V _in Given the duty cycle D, calculate the impedance weighting coefficients for each harmonic order k (k=1, 2, ..., K):

[0154] w _k =V _in 2 ×sin 2(k×π×D) / (k 2 ×π 2 ×|Z _k | 2 );

[0155] The numerator shows that the square of the inherent voltage ripple amplitude of the k-th harmonic is determined by the input voltage, duty cycle, and harmonic order, while the denominator is the square of the impedance amplitude of the circuit at that frequency. Therefore, a larger weighting coefficient indicates a stronger voltage excitation and lower circuit impedance at that harmonic frequency, resulting in a greater impact on current ripple, and thus a higher priority for suppression during optimization. For low-confidence identification points, multiplying their weighting coefficients by a reduction factor, such as 0.5, can reduce the impact of unreliable impedance estimation on the optimization results.

[0156] After the calculation is completed, the benchmark weighted coefficient sequence {w} is output. _1 w _2 , ..., w _K This sequence corresponds to the current switching frequency f. _sw 0 The weighted distribution under the given conditions.

[0157] The model is established using N-1 adjustable components in the phase shift vector (with the first phase fixed as the reference and the phase shift zero) and the switching frequency as joint decision variables, and the impedance-weighted sum of each harmonic current ripple as the optimization objective.

[0158] The objective function is the sum of weighted ripple contributions for all harmonic orders. For a given phase shift vector and switching frequency, the ripple intensity at the k-th harmonic after N-phase superposition is measured by the squared magnitude of the phase shift superposition factor. This squared magnitude can be expanded as follows: the number of phases N is a constant term, and the sum of the cosine functions of the phase differences between all phases is used. That is, for each pair of phase indices j and m (j is less than m), the cosine value of k times the phase difference is calculated, and then summed over all pairs.

[0159] Substituting the expanded result into the objective function and removing constant terms irrelevant to phase shift, the optimization problem is equivalent to minimizing a function relating each pair of phase shift differences. For each pair of phase shift differences, a weighted trigonometric sum is calculated, with each order weighting coefficient as a weight and the cosine of the phase shift difference as a basis function. Specifically, for a phase shift difference Δ, the corresponding weighted trigonometric sum is: g(Δ) = Σ _{k=1} K w _k ×cos(k×Δ). The overall objective function is the sum of all relative weighted triangular sums.

[0160] In this model, the phase shifts are arranged in ascending order and lie within an open interval of 0 to 2π; the switching frequency is selected from a pre-defined candidate set. This model unifies phase shift optimization and switching frequency in the same objective function. That is, the weighting coefficients are naturally coupled through the impedance term to the frequency selection and phase shift decision. Changing the switching frequency will shift the physical frequency position corresponding to each harmonic, change the impedance value and the distribution of the weighting coefficients, and affect the solution for the optimal phase shift.

[0161] Based on the rated operating range of the magnetic components and the switching loss constraints of the power devices, a set of candidate switching frequencies is established. This set contains L discrete selectable frequency values, for example, 5 candidate values, which are uniformly or non-uniformly distributed from the lowest to the highest within the design allowable range.

[0162] For each candidate switching frequency, the physical frequency corresponding to each harmonic is related to the current operating frequency f. _sw 0 The identified frequency points may not coincide. Based on a known frequency domain impedance amplitude sequence, its range is f. _sw 0 To Kf _sw 0 The impedance amplitude of each harmonic order at the physical frequency is estimated using linear interpolation. The interpolated impedance values ​​are then substituted into the weighting coefficient calculation formula to generate an independent candidate weighting coefficient vector for each candidate frequency. Each of the L candidate frequencies has weighting coefficients reflecting its harmonic impedance matching relationship.

[0163] For each candidate switching frequency and its corresponding candidate weighting coefficient vector, gradient iteration is performed independently. The phase shift vector is initialized with an equally spaced distribution as the starting search point. In each iteration, the gradient of the objective function with respect to each adjustable phase shift component is analytically calculated using the current phase shift vector. For the m-th adjustable phase shift component, its gradient is equal to -2 multiplied by the sum of all harmonic orders k. The contribution of each order is the product of the weighting coefficient of that order and the order value, multiplied by k times the sine of the phase shift difference summed over all remaining phase indices j (j ≠ m). The gradient expression is:

[0164] ΦJ / Φθ _m =-2×Σ _{k=1} K k×w _k ×Σ _{j≠m} sin(k×(θ _m -θ _j ));

[0165] Where Φ is the partial derivative, θ _m Let θ be the phase shift value before the m-th phase. _j w is the j-th equivalent pre-phase shift value. _k is the k-th order weighting coefficient corresponding to the current candidate frequency.

[0166] Update each adjustable phase shift component along the negative gradient direction using a step size η. After updating, check the phase sequence constraint. If a phase sequence out-of-bounds error occurs, project the corresponding component back to the feasible region boundary. Calculate the updated objective function value and determine if the absolute value of its difference from the previous iteration value is less than the preset convergence threshold ε. If the convergence condition is met or the maximum number of iterations is reached, terminate the iteration; otherwise, continue with the next round of gradient calculation and update.

[0167] Since the objective function is a non-convex periodic trigonometric polynomial, a single initial point may converge to a local optimum. Therefore, in addition to equally spaced initial points, 2 to 3 additional sets of initial phase shift vectors satisfying phase sequence constraints are randomly generated, and the iterative process is performed on each set. From the convergence results of all initial points, the one with the smallest objective function value is selected as the local optimum phase shift vector and its corresponding local minimum objective function value at that candidate frequency.

[0168] Because the decision variables have a low dimensionality, for example, only 2 to 5 adjustable phase shift components for a typical 3 to 6 phase system, the total computational cost of gradient optimization with multiple initial points can be controlled within milliseconds.

[0169] After phase shift optimization is performed on all L candidate switching frequencies, the local minimum objective function values ​​of each candidate frequency are summarized, and the candidate frequency with the smallest value is selected. The candidate switching frequency corresponding to this minimum value is the optimal switching frequency f. _sw * The corresponding local optimal phase shift vector is the optimal phase shift vector θ. * .

[0170] This optimal result represents the interleaved control parameters that achieve the global minimum weighted effective value of current ripple impedance under the current electrolytic cell circuit impedance conditions. Compared with the traditional equal-spacing phase shift, this optimal solution may be a non-equal-spacing phase shift distribution. By appropriately reducing the cancellation effect of low-weighted harmonic orders, a deeper suppression of high-weighted harmonic orders (corresponding to the low-impedance frequency band of the circuit) is achieved, thus obtaining a better overall performance in terms of total current ripple.

[0171] Furthermore, the optimal switching frequency f _sw * and the optimal phase shift vector θ * The data is sent to the pulse width modulation module for execution, which continuously collects the output current waveform and calculates the measured current ripple index. When the index exceeds the preset threshold or reaches the periodic update time, it triggers the re-execution of online identification and joint optimization, and sends the updated optimization parameters down for execution again, forming an adaptive closed loop of perception, modeling, optimization, and execution.

[0172] Specifically, reading the optimal switching frequency f _sw * and the optimal phase shift vector θ *The phase shift values ​​of each carrier wave and the switching frequency are synchronously written into the control register of the pulse width modulation module. Phase shift switching employs a gradual, cycle-by-cycle approach, linearly transitioning from the current value to the target value over several switching cycles to avoid current surges caused by abrupt phase shift changes. Switching frequency switching is also performed at carrier trough times, coordinating and synchronizing with the phase shift transition. After parameter switching is complete, the system enters steady-state operation under the new control parameters.

[0173] During the continuous operation phase after the optimized parameters take effect, the output current waveform is periodically collected. The peak-to-peak value and RMS value of the current ripple are extracted from each collected waveform, and the measured current ripple index is calculated. This index serves two purposes: firstly, as a quantitative verification of the optimization effect, comparing it with the ripple level before optimization; and secondly, as a trigger criterion for subsequent adaptive updates and continuous monitoring.

[0174] The continuously calculated measured current ripple index is compared with a preset ripple degradation threshold, and it is simultaneously checked whether the time interval since the last impedance identification has exceeded the periodic update time limit, typically set to 1 to 10 seconds, which is much faster than the time constant of changes in electrolytic cell operating conditions. When the conditions are met, adaptive updates are triggered. For example, if the measured current ripple index exceeds the threshold, it indicates that the electrolytic cell impedance has shifted due to changes in operating conditions, i.e., changes in temperature, pressure, or current density, and the current optimized parameters are no longer optimal; if the periodic update time limit is reached, routine impedance tracking is performed.

[0175] After the update is triggered, use the current optimal phase shift vector θ. * and optimal switching frequency f _sw * Starting from the base operating state, impedance identification is re-executed. This time, the perturbation reference is the current optimal phase shift, not the equally spaced phase shift. The principle and operation procedure of the perturbation remain unchanged. Then, the new optimal parameters are solved, and the process is repeated. This continuous closed-loop system allows the interleaved control parameters to adaptively adjust in real time according to changes in the electrolyzer's operating state.

[0176] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for adaptive suppression of current ripple in an electrolytic cell driven by a multiphase interleaved chopper, wherein the multiphase interleaved chopper is connected to the electrolytic cell via a filter inductor to form an electrolytic cell circuit, characterized in that, include: Online identification of frequency domain impedance characteristics of electrolytic cell circuits at multiple harmonic frequency points; Based on the frequency domain impedance characteristics, the interleaved phase shift and switching frequency of the multiphase interleaved chopper are jointly optimized, and the optimal control parameters are obtained by minimizing the preset current ripple objective function. The optimal control parameters are sent to the multiphase interleaved chopper for execution, and the online identification and joint optimization are repeated according to the update conditions.

2. The method according to claim 1, characterized in that, Online identification of the frequency domain impedance characteristics of an electrolytic cell circuit at multiple harmonic frequency points, specifically including: A controlled micro-perturbation is applied to the carrier phase shift of at least one phase in a multiphase interleaved chopper to actively generate measurable voltage and current responses at a preset harmonic frequency; wherein the micro-perturbation is a perturbation of a preset amplitude. Frequency domain impedance characteristics are obtained by frequency-by-frequency calculation based on voltage and current responses.

3. The method according to claim 2, characterized in that, Online identification also includes: Distinguish between steady-state residual harmonic frequency points and phase-shift canceled harmonic frequency points among multiple harmonic frequency points; Among them, for the residual harmonic frequency point, the voltage and current harmonic components are extracted from the output waveform of the multiphase interleaved chopper in steady state operation to calculate the impedance; For phase shift cancellation harmonic frequency points, impedance is identified by applying a perturbation.

4. The method according to claim 1, characterized in that, Joint optimization specifically includes: Based on the frequency domain impedance characteristics, the corresponding impedance weighting coefficient is calculated for each frequency point in the harmonic frequency range. Establish a current ripple objective function with impedance weighting coefficients as weights to minimize the total energy of the output current ripple.

5. The method according to claim 4, characterized in that, The impedance weighting coefficient is calculated as follows: Based on the input voltage and duty cycle of the multiphase interleaved chopper, the inherent voltage ripple amplitude at each harmonic frequency is obtained. This makes the impedance weighting coefficient proportional to the square of the inherent voltage ripple amplitude at the corresponding harmonic frequency, and inversely proportional to the square of the impedance amplitude of the electrolytic cell circuit at that harmonic frequency.

6. The method according to claim 4, characterized in that, The joint optimization method utilizes a hierarchical strategy that combines outer-layer discrete search of switching frequency with inner-layer phase shift optimization to solve the problem.

7. The method according to claim 6, characterized in that, Discrete search of outer layer switching frequency, specifically including: Iterate through each candidate switching frequency in the candidate switching frequency set; For each candidate switching frequency, based on the frequency domain impedance characteristics, the impedance value corresponding to each harmonic frequency point at that candidate switching frequency is obtained by interpolation calculation, and the corresponding impedance weighting coefficient is generated for use in inner layer phase shift optimization.

8. The method according to claim 1, characterized in that, The update conditions include performance feedback-based triggering conditions, and the determination process for these triggering conditions includes: After the optimal control parameters are issued and executed, the real-time output current waveform of the chopper is continuously collected. The measured current ripple index is calculated based on the real-time output current waveform. The measured current ripple index is compared with a preset degradation threshold. When the index exceeds the threshold, repeated execution is triggered.

9. The method according to claim 1, characterized in that, The preset update conditions include timer-based trigger conditions, and the determination process for these trigger conditions includes: Monitor the time interval since the last online identification and joint optimization was completed; When the time interval reaches the preset periodic update time limit, the process is triggered to repeat.

10. The method according to claim 2, characterized in that, Online identification further includes a credibility verification of the identification results, which includes: The theoretical voltage response amplitude induced by a small disturbance at the harmonic frequency point is calculated in advance; Compare the theoretical voltage response amplitude with the measured amplitude of the measurable voltage response; When the deviation between the two exceeds the threshold, the identification result of the corresponding harmonic frequency point is marked as low confidence, and its weight is reduced in the joint optimization.

Images