Dual-high-frequency resonance suppression cooperative control method

Abstract

The invention discloses a dual-high-frequency resonance suppression cooperative control method, which comprises the steps of positioning a potential resonance path and a target frequency band based on topology and parameters, acquiring full-bandwidth original time domain data of the target frequency band, and preprocessing the full-bandwidth original time domain data to obtain target frequency band time domain data; carrying out resonance characteristic detection and historical consistency verification on the target frequency band time domain data through time-frequency coupling modal analysis, and outputting effective resonance characteristic parameters; calculating and optimizing inhibition bandwidth, damping and gain through effective resonance characteristic parameters, and combining phase compensation and collaborative alignment to obtain a dynamic phase compensation signal and an alignment signal; the method comprises the following steps: by taking suppression bandwidth, damping ratio and gain thereof, real-time observation data and prior history as input, firstly performing parameter decoupling recursion identification, and then outputting a globally optimal high-credibility primary and secondary modal parameter set through three-layer optimization adaptive optimization; a driving signal is generated and acts on the power electronic switching device, and real-time dynamic suppression of the target resonance frequency point is achieved.

Description

Technical Field

[0001] This invention relates to the field of high-frequency resonance suppression technology, and in particular to a dual high-frequency resonance suppression collaborative control method. Background Technology

[0002] High-frequency resonance refers to a power system phenomenon in new power distribution networks where the high-frequency switching behavior (such as PWM control) of power electronic equipment (inverters, converters) creates a resonant circuit with the parasitic inductance and capacitance of the lines, resulting in abnormal amplification of voltage or current at a specific frequency, which threatens equipment safety, system stability and power quality.

[0003] High-frequency resonance is a complex physical process influenced by multiple factors. When new energy sources are integrated into the distribution network, the extensive use of power electronic devices alters the electrical characteristics of the traditional power grid. For example, the switching devices and control strategies of modular multilevel converters (MMCs) introduce impedance characteristics into the grid. When the impedance of power electronic devices interacts with that of the grid, negative damping characteristics emerge. In oscillating systems, mechanisms that should provide damping fail, instead exacerbating the oscillations. Furthermore, impedance mismatch is a significant concern. While the impedance characteristics of the power grid vary with operating conditions, the impedance characteristics of power electronic devices remain relatively constant; therefore, impedance mismatch leads to high-frequency resonance.

[0004] However, the following problems still exist when suppressing high-frequency resonance: Under the coupling effect of two high-frequency resonant circuits, the primary and secondary resonant frequencies exhibit nonlinear motion, with characteristic frequency spacing showing alternating contraction and even overlap. When the frequency spacing is smaller than the system's physical identification, classical frequency domain tracking algorithms cannot correctly identify the resonant state. Complex dual-mode energy spectra show time-varying phenomena where the proportion of primary and secondary mode energies is greater than the ratio of primary and secondary mode energies but less than the ratio of primary and secondary mode energies, posing a significant challenge to real-time modal energy identification.

[0005] When the spacing between two high-frequency resonant points is within the corresponding impedance adjustment bandwidth, the suppression resonant points within the overlapping band can exhibit excessive gain in traditional multiresonant suppression filter designs. This impedance response characteristic, contradicting the design of suppression resonant filters, is caused by the interaction of heterogeneous impedance networks in complex and unknown power grids. In practical cases, the energy transfer path between multiple resonant points is disrupted, rendering the linear superposition principle of traditional multiresonant suppression filter design theory inapplicable. Furthermore, the recombination of harmonic energy in the aliasing band can also cause inappropriate resonant point excitation.

[0006] When designing damping circuits for two resonant modes, a strong dynamic interaction exists during the adjustment of the gains of the primary and secondary circuits. The primary circuit can improve the suppression effect of the corresponding resonant mode in the short term, but simultaneously injects energy into the secondary resonant mode through the interaction channel. If the gains are distributed equally, the overall suppression effect is poor. Due to the "interactive antagonism" of the gains of the primary and secondary circuits, multi-objective optimization often lingers at local stability points deviating from the Pareto effective front, making it difficult to achieve the optimal balance of suppression effects.

[0007] The delay differences in dual-loop control lead to asymmetry in phase lag response. The superposition of phase lag vectors between two or more loops may result in a positive feedback loop at a certain frequency. Phase compensation cannot achieve accurate phase matching for conventional dual-frequency dynamic migration, while predictive phase compensation, although improving phase matching speed, also increases control difficulty and exacerbates noise amplification to some extent.

[0008] Under high-frequency and dual-frequency resonance conditions, the system parameter space rapidly grows into a heterogeneous, time-varying high-dimensional space, influenced by variables such as resonant frequency, modal damping, and coupling coefficients. Low-dimensional identification algorithms face a severe curse of dimensionality; strong correlations between variables increase the identifiability of single variables, significantly worsening the statistical variance of the estimation results. In particular, as the dual-frequency points dynamically approach each other, the correlations between variables become even stronger, making parameter identification physically impossible.

[0009] Suppression systems need to balance high-dimensional characteristics such as resonance depth suppression, response time guarantee, phase margin guarantee, and impedance matching optimization. Traditional optimization algorithms often face a conflict between the complexity of engineering models and real-time computing power when solving these multi-objective and multi-dimensional engineering problems. Establishing a reasonable dimensionality reduction and hierarchical optimization mechanism is the foundation for improving the efficiency of multi-modal collaborative suppression.

[0010] The purpose of this invention is to design a dual high-frequency resonance suppression and coordinated control method to address the problems existing in the prior art. Summary of the Invention

[0011] In view of this, the purpose of this invention is to propose a dual high-frequency resonance suppression and coordinated control method that can solve the above-mentioned problems.

[0012] This invention provides a dual high-frequency resonance suppression and coordinated control method, comprising: Based on topology and parameters, potential resonant paths and target frequency bands are located. The full bandwidth of the target frequency band raw time domain data is collected, and the full bandwidth raw time domain data is preprocessed to obtain the target frequency band time domain data. Time-frequency coupled modal analysis is used to detect resonance characteristics and verify historical consistency of time-domain data in the target frequency band, and output effective resonance characteristic parameters. By calculating and optimizing the suppression bandwidth, damping and its gain through effective resonance characteristic parameters, and combining phase compensation and cooperative alignment, dynamic phase compensation signal and aligned signal are obtained; Taking the suppression bandwidth, damping ratio and its gain, real-time observation data and prior history as input, the parameter decoupling recursive identification is first performed, and then the three-layer optimization adaptive optimization is used to output the globally optimal high-confidence primary and secondary modal parameter set. The globally optimal high-confidence primary and secondary mode parameter set is used to generate a driving signal and apply it to power electronic switching devices to achieve real-time dynamic suppression of the target resonant frequency.

[0013] The beneficial effects of this invention are: First, it enables high-precision perception and real-time acquisition of complex system data, providing accurate foundational data for subsequent optimization and avoiding optimization deviations caused by errors in the original input. Primary and secondary modal identification can accurately distinguish and extract core and secondary modalities affecting system performance, achieving information dimensionality reduction and feature extraction, effectively improving the accuracy and robustness of parameter identification. It provides downstream algorithms with a highly reliable initial parameter set, laying a solid and reliable foundation for global system optimization.

[0014] Secondly, by fusing primary and secondary modes, the model's representational ability and overall consistency are further enhanced, improving its adaptability to various operating conditions. This effectively suppresses noise and redundant interference, achieving efficient simplification and refinement of parameter vectors, thus reducing computational overhead for subsequent optimization steps. Based on information from identification confidence levels or case libraries, the initial parameters of the optimization algorithm are intelligently selected, improving the global convergence and search efficiency of the optimization algorithm and reducing the risk of getting trapped in local optima.

[0015] Third, a particle swarm optimization algorithm based on quantum tunneling probability is adopted, which significantly improves the global search capability and convergence speed for parameter optimization, resulting in a better set of system performance parameters. In a multi-objective space, fuzzy logic and adaptive weight allocation are used to achieve coordination and optimal trade-offs among the objectives, ensuring optimal overall system performance under multiple constraints.

[0016] Fourth, by constructing a historical operating condition-parameter case library, the system fully leverages its past optimization experience to achieve intelligent matching and rapid parameter reuse with the current operating condition, improving the scientific rigor and efficiency of the initial optimization point. Using efficient correlation metrics such as cosine similarity, dynamic and intelligent optimization initialization switching is achieved, effectively addressing parameter inheritance barriers or initialization blind spots under new operating conditions. This significantly reduces the waste of computational resources in global random exploration, improves the real-time performance, adaptability, and robustness of parameter optimization in practical applications, and enhances the algorithm's generalization and scalability. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings required in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the method in this embodiment.

[0019] Figure 2 This is the equivalent circuit diagram of this embodiment.

[0020] Figure 3 This is a real-time tracking diagram of the dual high-frequency resonant frequencies in this embodiment.

[0021] Figure 4 This is a graph showing the real-time response time of this embodiment.

[0022] Figure 5 This is a real-time tracking speed curve diagram of this embodiment.

[0023] Figure 6 This is a real-time network bandwidth usage curve for this embodiment. Detailed Implementation

[0024] To facilitate understanding by those skilled in the art, the structure of the present invention will now be described in further detail with reference to the accompanying drawings. It should be understood that, unless otherwise specified, the order of the steps mentioned in this embodiment can be adjusted according to actual needs, and they can even be executed simultaneously or partially simultaneously.

[0025] like Figure 1 As shown, this embodiment of the invention provides a dual high-frequency resonance suppression and coordinated control method, including: S1 locates potential resonant paths and target frequency bands based on topology and parameters, collects full-bandwidth raw time-domain data of the target frequency band, and preprocesses the full-bandwidth raw time-domain data to obtain the target frequency band time-domain data; S101 establishes an equivalent circuit based on topology and component parameters, calculates series / parallel resonant frequencies, and determines potential resonant paths and target frequency bands; S102 collects the corresponding full-bandwidth raw time-domain voltage and current data for the target frequency band and performs time synchronization and alignment processing. S103 sequentially performs DC removal, normalization, and bandpass filtering on the synchronized time-domain voltage and current data to obtain the target frequency band time-domain data.

[0026] In this step, an equivalent circuit is established based on the topology and component parameters, such as... Figure 2As shown, the main circuit topology comprises four main parts: the DC-side circuit, the inverter circuit, the output filter of the LCL parallel LC branch, and the power grid. The DC-side simulated solar photovoltaic array provides DC energy supply, and the inverter circuit converts the DC energy into AC energy. Through primary and secondary mode identification and hierarchical extraction of characteristic parameters, high-precision frequency point discrimination is achieved, improving the identifiability of primary and secondary modes and effectively solving the technical problems of frequency overlap, primary and secondary mode energy aliasing, and the inability to identify due to reduced characteristic frequency spacing.

[0027] The specific formula for calculating the series / parallel resonant frequency is as follows: Introducing new physical quantities =2 , =54 .

[0028] (1) and Series resonance occurs:

[0029] Solving for:

[0030] Substitute parameters:

[0031] (2) and Parallel resonance occurs:

[0032] Solving for:

[0033] Substitute parameters:

[0034] S2 performs resonance characteristic detection and historical consistency verification on the time-domain data of the target frequency band through time-frequency coupled modal analysis, and outputs effective resonance characteristic parameters. S201 performs a Hanning window short-time Fourier transform on the time-domain data based on the target frequency band to obtain the time spectrum. The calculation formula is as follows: , in, Represents the time spectrum. Indicates time displacement. Represents frequency domain variables, Represents time-domain data, Let N represent the Hanning window function, N represent the window length for each frame, and n represent the global temporal sample index. Represents the imaginary unit; S202 performs a local peak search on the amplitude spectrum of the target frequency band in each time frame, and refines the frequency and amplitude by interpolation in the peak neighborhood using cubic splines to generate a set of candidate peaks; S203 calculates the historical mean and standard deviation of each candidate peak, eliminates false peaks and transient interference, and outputs the effective peaks and their characteristics.

[0035] In this step, a hybrid time-frequency method is used to separate overlapping resonant modes and identify their characteristics. Time-frequency analysis is then performed using the Hanning window short-time Fourier transform technique. The Hanning window function is used to suppress spectral leakage. For weak resonance characteristics, the algorithm optimizes frequency resolution using cubic spline interpolation: after detecting a candidate resonance peak, four frequency points before and after it are selected to construct a local spectral curve, generating 4000 interpolation points, successfully improving the frequency resolution from 17Hz to 0.05Hz. Simultaneously, a dynamic history verification mechanism eliminates transient interference by calculating the mean and standard deviation (σhistory < 0.06μhistory) of the most recent seven detection results. This method maintains a detection accuracy of over 95% even at a low signal-to-noise ratio of -30dB, making it particularly suitable for complex scenarios where the spacing between two frequency points approaches the Nyquist limit, such as resolving two frequency points with a spacing within 8.75kHz at a sampling rate of 35kHz.

[0036] By integrating parameters and intelligent initial values, the robustness of the optimization model is improved through strong correlation between primary and secondary loops, dimensionality reduction, redundancy removal, and anti-interference, thus solving the problems of strong coupling between primary and secondary resonant modes, high variable correlation, and large initialization bias and identification variance.

[0037] S3 calculates and optimizes the suppression bandwidth, damping and its gain through effective resonance characteristic parameters, and obtains dynamic phase compensation signal and alignment signal by combining phase compensation and cooperative alignment. Based on real-time effective resonant characteristic parameters, S301 dynamically sets the suppression bandwidth and damping ratio of the dual-tuned filter, and dynamically generates the dual-tuned filter accordingly. The suppression bandwidth and damping ratio are optimized through a lead-delay complementary network, and the output suppression bandwidth, damping ratio and gain that meet the conditions are output. S3011 sets its dynamic suppression center frequency based on the effective peak and its characteristics. Variable suppression bandwidth ; S3012 based on dynamic suppression center frequency Variable suppression bandwidth Calculate the damping ratio Generate the dynamic transfer function of the dual-tuned filter. The formula for calculating the dynamic transfer function is as follows: , in, Let k denote the Laplace operator, and k denote the k-th tuning channel. This represents the center angular frequency of the k-th channel; S3013 will have variable suppression bandwidth Damping ratio By applying physical matching constraints and suppressing the phase through a lead-delay complementary network, the suppression bandwidth, damping ratio, and gain that meet the conditions are obtained.

[0038] In this step, resonance is suppressed by dynamically varying the stopband parameters of the dual-tuned filter. Its transfer function is designed as follows: , Among them, damping ratio stopband bandwidth Real-time variable (1.2-1.5 octaves). To eliminate phase deviation, the algorithm employs a lead-delay complementary network to suppress phase: the lead network ( = Compensation for high-frequency phase delay, hysteresis network This approach suppresses low-frequency phase abrupt changes, ultimately ensuring that the phase error in the dual high-frequency resonant band (2400-2500 Hz) is less than 2°. A smooth update variable (weighted at 0.65 historical values ​​+ 0.35 new values) is used to avoid step abrupt changes.

[0039] S302 dynamically optimizes and adjusts the damping ratio and gain of each mode by coupling the state-space model and extended Hamiltonian optimization, combined with weight allocation and dynamic Nash equilibrium solution. For the energy feedback problem of the primary and secondary resonant circuits, S3021 establishes a coupled state-space model, and the calculation formula is as follows: , in, This represents the time derivative of the main loop. This represents the time derivative of the secondary loop. Represents the equivalent circuit resistance. Indicates the state of the main circuit. Indicates the state of the secondary loop. Indicates the main circuit inductance. Indicates the DC bus voltage. This indicates the control input, where v represents the grid connection point voltage. Indicates the resistance of the damping branch. Indicates the capacitance of the damping branch. and Indicates the energy coupling coefficient; In this step, =0.18 and =0.22.

[0040] S3022 establishes the extended Hamiltonian function H, setting the state vector, control variables, weights, and constraints for energy optimization. The calculation formula is as follows:

[0041] in, This indicates the transpose of the main loop state vector. This represents the state weighting matrix of the main loop. This represents the transpose of the secondary loop state vector. This represents the state weighting matrix of the secondary loop. R represents the transpose of the control matrix, R represents the control weight matrix, and u represents the control input vector. This represents the transpose of the constraint matrix. Indicates the state term. Indicates the input action item; S3023 uses gradient projection-Nash equalization to solve for the optimal damping ratio and gain of the system in the current state in real time, with respect to the extended Hamiltonian function H. The optimal gradient term is: , in, R represents the gradient of Hamiltonian with respect to the control input u, R represents the control weight matrix, and u represents the control input vector. This represents the transpose of the input matrix.

[0042] In this step, a coupled state-space model is established to address the energy feedback problem between the primary and secondary resonant circuits: , Among them, the energy coupling coefficients are respectively =0.18 and =0.22. By extending the Hamiltonian function:

[0043] The gradient projection method is used to solve the dynamic Nash equilibrium, and the gradient term... Implicit Calculation in progress. The weight of the primary loop is 1.8 times that of the secondary loop, and the suppression ratio of the primary resonance peak reaches 23. The secondary resonance peak is not lower than 18. When the modal energy exceeds the threshold, the level tuning mechanism will increase the corresponding loop gain by 30%, achieving millisecond-level dynamic reconfiguration.

[0044] The S303 takes the resonant frequency, fractional-order parameters, and real-time frequency shift rate as inputs, and combines them with the optimized damping ratio and its gain. Through fractional-order phase compensation, adaptive step size adjustment, and high-frequency noise suppression, it obtains dynamic phase compensation signals and alignment signals.

[0045] S3031 constructs a phase-shift synthesizer based on the target resonant frequency and fractional-order parameters, obtaining the phase-shift synthesizer parameters and transfer function. transfer function The formula is as follows: , in, Represents the proportionality coefficient. Represents the Laplace operator. Indicates the cutoff frequency. Indicates the adjustable differential order; The S3032 adjusts the adjustable differential order via real-time frequency shift rate. The adjustment step size is used to adaptively update the phase shift synthesizer parameters; The S3033 takes the phase compensation signal obtained from the phase shift synthesizer, passes it through an IIR low-pass filter to suppress high-frequency noise, and outputs a dynamic phase compensation signal and an alignment signal.

[0046] In this step, conventional integer-order phase shift networks, due to their order limitations, cannot simultaneously satisfy the nonlinear phase shift compensation problem over a wide frequency range. We construct a continuously adjustable phase shift synthesizer based on the fractional-order differential operator (0.5-1.5), whose transfer function is: , Where α is the adjustable differential order. The cutoff frequency is used. In the dual high-frequency resonance suppression, the look-ahead differential order α = 0.8 at the main resonant point (2400Hz) increases the high-frequency phase margin compensation for the resonance peak; the hysteresis differential order α = 1.2 at the secondary resonant point (2500Hz) reduces the impact of low-frequency phase jumps on resonance and stability. These two compensation strategies form a "phase offset" mechanism, limiting the phase error across the entire frequency band to within 1.5°. To achieve dynamic adjustment, the algorithm updates the value in real time based on the frequency shift rate: when a frequency drift exceeding 200Hz / s is detected, the α adjustment step size is increased from 0.05 to 0.1, accelerating phase tracking. To suppress high-frequency noise amplification during the compensation process, a cascaded infinite impulse response (IIR) low-pass filter is used. , It also provides -20dB attenuation at 5kHz while maintaining phase-matching accuracy. In experiments, it increased the stability margin from 32° (conventional approach) to 48° and reduced transient overshoot by 60%.

[0047] The hierarchical, multi-objective quantum particle swarm optimization algorithm, in conjunction with fuzzy weights, takes into account the global optimum of multiple performance indicators, improves optimization efficiency and compatibility, and solves the technical problems of multi-objective performance conflicts, impedance matching, response speed, resonance suppression, and high-dimensional model optimization convergence difficulties.

[0048] S4 takes suppression bandwidth, damping ratio and its gain, real-time observation data and prior history as input, first performs parameter decoupling recursive identification, and then performs three-layer optimization adaptive optimization to output a globally optimal high-confidence set of primary and secondary modal parameters.

[0049] S401 takes the collected observation data and initial parameters as input, and outputs a set of principal and secondary mode parameters with high reliability and smooth fusion through parameter decoupling, recursive estimation of primary and secondary modes, robust discrimination and fusion processing. S4011 constructs the parameter sensitivity matrix Φ and calculates its Gram matrix. The Jacobi iterative method is used to solve for the eigenvalues. To obtain all eigenvalues With eigenvectors; S4012 maps the parameters corresponding to the eigenvectors whose eigenvalues ​​are greater than the eigenvalue threshold to independent subspaces to obtain the decoupled primary and secondary mode parameter vectors; S4013 employs a short-window strong tracking recursive algorithm to recursively update the principal mode parameter vector. The update rate is dynamically adjusted using Kalman gain to obtain the estimated results of the current principal mode's 4D parameters. Its state update equation is as follows: , in, This represents the parameter vector estimate at the current time k. This represents the parameter estimate from the previous time step. Represents the gain matrix. This represents the actual observed output at time k. This represents the model output mapping; S4014 uses a long time window for robust estimation of submodal parameter vectors and employs the Huber loss function. To suppress outlier interference, the estimation results of the 8-dimensional parameters of the current submodal are obtained, and the Huber loss function is described. The expression is as follows: , in, This represents the difference between observations and model predictions. Indicates the switching threshold; S4015 calculates the Manhattan distance from the vector estimation results of the primary and secondary modal parameters. ,when When the confidence level is >2.5, the low-confidence parameter updates are frozen to obtain a set of valid updated confidence parameters; S4016 performs a weighted smooth update on the effectively updated set of trusted parameters to obtain a smoothly fused set of high-confidence parameters for the primary and secondary modes.

[0050] In this step, the controller involves 12 strongly coupled parameters, including gain coefficient, damping ratio, and bandwidth weight. Traditional identification methods often get trapped in local optima. This technique achieves orthogonality of the parameter space through Gram matrix eigenvalue decomposition: first, the parameter sensitivity matrix Φ is constructed, and its Gram matrix is ​​calculated. Then, the Jacobi iterative method is used to solve for the eigenvalues. . For feature values ​​greater than the threshold ( The parameters corresponding to the eigenvectors of the model are mapped to independent subspaces, thereby decoupling strongly correlated parameters. A differentiated identification strategy is designed to address the dynamic differences between primary and secondary resonant modes: The principal modal parameters (4-dimensional): using a short-window (50 ms) strong tracking recursive algorithm, the state update equation is as follows: , Kalman gain The update rate is dynamically adjusted, and when the prediction error exceeds 5%, the learning rate is increased to 3 times the base value.

[0051] Submodal parameters (8 dimensions): Robust estimation is performed using a long time window (300ms), and the Huber loss function is used to suppress outliers.

[0052] , Where the threshold c = 2σ (σ is the standard deviation of measurement noise). The credibility assessment module calculates the Manhattan distance of the parameter vector: , in, Represents the current parameter vector. Indicates the reference mean. Represents the precision matrix, when At a confidence level of >2.5, low-confidence parameter updates are frozen, reducing the identification error of core parameters to less than 1%. This framework improves parameter convergence speed by 40% and parameter identification stability by 2.3 times.

[0053] S402 uses primary and secondary modal parameters as initial values, and integrates joint system observation data, optimization objectives, and historical cases as inputs. Through a three-layer hierarchical optimization architecture, it achieves adaptive and efficient parameter optimization, and outputs a globally optimal set of high-confidence primary and secondary modal parameters that satisfies multi-objective constraints.

[0054] S4021 uses the primary and secondary mode parameter sets as initial values ​​and employs an improved particle swarm optimization algorithm that incorporates quantum tunneling probability within a 4-dimensional gain parameter space to calculate the difference between the current fitness of a particle and the global optimum. According to the difference The quantum tunneling probability is calculated, and an optimization is performed based on the quantum tunneling probability to obtain the globally optimal parameter set. The formula for calculating the quantum tunneling probability is as follows: , in, Represents Boltzmann's constant. Indicates temperature parameter, Represents an exponential function; S4022 takes the globally optimal parameter set as input, applies fuzzy logic rules to combine weights in a 6-dimensional control target space, coordinates and schedules multiple targets, and constructs a membership function. The weights are adaptively adjusted according to the operating conditions, and a multi-objective constraint parameter set is output. The calculation formula for the membership function is as follows: , in, Indicates the expected value. Indicates the tolerable bandwidth. Indicates the steepness of the adjustment function. This represents the actual input variable to be assigned to; S4023 takes a set of multi-objective constraint parameters as input, constructs a decision case library of historical working conditions and optimal parameters, and uses cosine similarity to measure the similarity between the current working condition and historical cases. If the similarity is greater than the similarity threshold, the historical optimal parameters are directly selected as the initial solution for optimization. If the similarity is lower than the similarity threshold, it automatically switches to global randomization initialization.

[0055] In this step, a three-layer hierarchical optimization architecture is constructed for complex optimization problems with multiple objectives and constraints: The bottom layer (4-dimensional gain space) employs a quantum tunneling-modified particle swarm optimization algorithm. The quantum tunneling probability is introduced into the standard particle swarm optimization update formula: , Where ΔE is the fitness difference between the current particle and the global optimum, kB is the Boltzmann constant, and T is the temperature parameter. When a particle gets stuck in a local optimum (stagnates for more than 10 generations), it is probabilistically... The trigger position is randomly reset, breaking through local extrema. This improvement enables the algorithm to converge to the global optimal neighborhood within 30ms.

[0056] Mid-level (6-dimensional control target coordination): Constructing a fuzzy rule-based dynamic optimization mechanism. Quantifying target priority by defining membership functions: , in For the expected value, To tolerate bandwidth, Adjusting the steepness of functions, for example, the priority of harmonic distortion rate (THD) is set to the maximum when in steady state ( =3), the weight decreases during the transient process ( =1.5), to coordinate the overshoot suppression requirements.

[0057] Top-level (Historical Case Guidance): Build a decision knowledge base covering 500 sets of historical optimal solutions, and use cosine similarity retrieval to match cases.

[0058] , When the similarity reaches 0.85ms or higher, the previous best parameters are directly loaded as the initial solution; when the similarity is lower than 0.6, a global random search is started; through the bidirectional update mechanism of two parallel layers, the bottom elite solution set is initialized and the middle fuzzy rule is updated, and the top strategy is pushed down to the bottom initialization, which improves the global optimization efficiency by 40%, shortens the transient response time to 80ms, and reduces the optimization time of repeated jobs by 65%.

[0059] By using a historical case library and dynamic similarity matching, the system intelligently switches initial values, efficiently utilizes resources, and improves the adaptability to new operating conditions and global convergence guarantee. This solves the technical problems of low utilization of historical parameters, poor efficiency of random initialization under dynamic frequency point new operating conditions, and high computational complexity.

[0060] S5 generates a driving signal from the globally optimal high-confidence primary and secondary mode parameter set and applies it to the power electronic switching device to achieve real-time dynamic suppression of the target resonant frequency.

[0061] In this step, the obtained globally optimal high-confidence primary and secondary mode parameter set can be loaded into the resonance suppression controller to generate the PWM drive signal.

[0062] like Figure 2 As shown in the real-time resonant frequency graph, the system's resonant frequency changes from 0 to 0.8 seconds are recorded. Both resonant frequencies originate from their respective system master resonant frequencies. Resonant peak 1 fluctuates from 2500Hz, reaching a maximum of 2443.85Hz and a minimum of 2324.22Hz, with a fluctuation range of 119.63Hz. Resonant peak 2 fluctuates from 2400Hz, reaching a maximum of 2563.48Hz and a minimum of 2426.76Hz, with a fluctuation range of 236.62Hz. Overall, resonant peak 1 is more stable than resonant peak 2.

[0063] The following section provides an in-depth analysis based on the resonant frequency tracking performance analysis module: like Figure 3 As shown, the response time index reflects the system's efficiency in detecting and responding to frequency changes. The figure shows that the response start time of resonant peak 1 is approximately 0.11 s, and that of resonant peak 2 is approximately 0.17 s, both within the critical threshold of 0.3 s. This time difference may stem from the difference in the dynamic characteristics of the two resonant peaks. The signal energy in the frequency band where resonant peak 1 is located is stronger, or the system is more sensitive to it, while resonant peak 2 requires more complex spectral separation calculations due to band coupling effects. However, the overall response still meets the real-time control requirements, demonstrating the system's ability to quickly capture multi-band resonant abrupt changes.

[0064] like Figure 4 As shown, the tracking speed index reflects the dynamic performance of the system's frequency regulation. The tracking speed of resonant peak 1 reaches a peak of approximately 4965 Hz / s during certain periods, exhibiting strong transient response capability, which is attributed to the dynamic gain compensation and phase prediction mechanisms in the control algorithm. The tracking speed of resonant peak 2, on the other hand, remains within a relatively flat range, with fluctuations not exceeding 1200 Hz / s, indicating that the system employs a more robust tracking strategy for resonant peaks with smaller fluctuations to avoid overshoot. The difference in speed between the two reflects the differentiated processing logic of the control core for the magnitude of resonant peak fluctuations, which can both quickly suppress major harmonics and maintain the overall stability of the system.

[0065] like Figure 5 As shown, the data rate metric reveals the effectiveness of system resource scheduling. The data rate remained consistently at 136.719 KB / s after approximately 0.06 seconds, without any instantaneous data spikes or communication interruptions. This metric is closely related to the fixed detection interval of the FFT analysis module. The 0.05-second spectrum refresh cycle ensures uniform data throughput, meeting the computational demands of high-precision 2048-point spectrum analysis while reducing redundant data transmission through window function optimization. The stable data flow indicates that the system has achieved a balance between real-time performance and resource utilization, providing a reliable guarantee for continuous resonance monitoring.

[0066] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0067] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0068] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0069] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0070] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The words first, second, and third, etc., do not indicate any order. These words can be interpreted as names.

[0071] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0072] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0073] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0074] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

Claims

1. A dual high-frequency resonance suppression cooperative control method, characterized by, The method comprises the following steps: Based on topology and parameter positioning, potential resonance path and target frequency band are located, full-bandwidth original time-domain data of the target frequency band is collected, and target frequency band time-domain data is obtained by preprocessing the full-bandwidth original time-domain data; Through time-frequency coupling modal analysis, resonance characteristic detection and historical consistency verification are performed on the target frequency band time-domain data, and effective resonance characteristic parameters are output; The suppression bandwidth, damping ratio and gain are calculated and optimized through the effective resonance characteristic parameters, and the dynamic phase compensation signal and alignment signal are obtained by joint phase compensation and collaborative alignment; Taking the suppression bandwidth, damping ratio and gain, real-time observation data and prior history as inputs, the parameter decoupling recursive identification is performed first, and then the three-layer optimization adaptive optimization is performed to output the globally optimal high-confidence primary and secondary modal parameter set; The globally optimal high-confidence primary and secondary modal parameter set is used to generate a driving signal and act on the power electronic switching device to realize real-time dynamic suppression of the target resonance frequency point.

2. The dual high-frequency resonance suppression cooperative control method according to claim 1, characterized by, The method based on topology and parameter positioning, potential resonance path and target frequency band, collecting full-bandwidth original time-domain data of the target frequency band, and preprocessing the full-bandwidth original time-domain data to obtain target frequency band time-domain data comprises: Based on topology and element parameters, an equivalent circuit is established, series / parallel resonance frequency points are calculated, potential resonance paths and target frequency bands are determined; For the target frequency band, the corresponding full-bandwidth original time-domain voltage and current data are collected, and time synchronization alignment processing is performed; The time-domain voltage and current data after synchronization alignment are sequentially subjected to DC removal, normalization and band-pass filtering to obtain target frequency band time-domain data.

3. The dual high-frequency resonance suppression cooperative control method according to claim 1, characterized by, The method of performing resonance characteristic detection and historical consistency verification on the target frequency band time-domain data through time-frequency coupling modal analysis, and outputting effective resonance characteristic parameters comprises: Performing Hanning window short-time Fourier transform on the time-domain data based on the target frequency band to obtain a time-frequency spectrum, and the calculation formula is as follows: ， wherein denotes the time-frequency spectrum, denotes the time shift, denotes the frequency domain variable, denotes the time domain data, denotes the Hanning window function, N denotes the window length of each frame, and n denotes the global time sample index, denotes the imaginary unit; Performing local peak value search on the amplitude spectrum of the target frequency band in each time frame, and using cubic spline interpolation to refine the frequency and amplitude in the peak neighborhood to generate a candidate peak set; Calculate the historical mean and standard deviation of each candidate peak, eliminate false peaks and transient interference, and output the effective peak and its characteristics.

4. The dual high-frequency resonance suppression cooperative control method according to claim 1, characterized by, The method of calculating and optimizing the suppression bandwidth, damping ratio and gain through the effective resonance characteristic parameters, jointly compensating the phase and aligning to obtain the dynamic phase compensation signal and alignment signal comprises: Based on real-time effective resonance characteristic parameters, the suppression bandwidth and damping ratio of the double-tuned filter are dynamically set, and the double-tuned filter is dynamically generated accordingly, the suppression bandwidth and damping ratio are optimized through the lead-time-lag complementary network, and the suppression bandwidth, damping ratio and gain that meet the conditions are output; Through coupling state space model and extended Hamilton optimization, combining weight distribution and dynamic Nash equilibrium solution, the damping ratio and gain of each mode are dynamically optimized and adjusted; Taking the resonance frequency, fractional order parameter and real-time frequency shift rate as inputs, combining the optimized damping ratio and gain, and through fractional order phase compensation, adaptive step adjustment and high-frequency noise suppression processing, the dynamic phase compensation signal and alignment signal are obtained.

5. The dual high-frequency resonance suppression cooperative control method according to claim 4, characterized by, The real-time effective resonance characteristic parameters are used to dynamically set the suppression bandwidth and damping ratio of the double-tuned filter, and the double-tuned filter is dynamically generated, the suppression bandwidth and damping ratio are optimized through the lead-time-lag complementary network, and the suppression bandwidth, damping ratio and gain meeting the conditions are output, including: According to the effective peak and its characteristic, the dynamic suppression center frequency is set , variable suppression bandwidth ; According to the dynamic suppression center frequency , variable suppression bandwidth Calculate the damping ratio , generate the dynamic transfer function of the double-tuned filter The calculation formula of the dynamic transfer function is as follows: ， wherein, represents the Laplacian operator, k represents the kth tuning channel, represents the center angular frequency of the kth channel; Variable suppression bandwidth and damping ratio The physical matching constraint and the phase suppression by the lead-lag complementary network are performed to obtain the suppression bandwidth, damping ratio and their gains satisfying the conditions.

6. The dual high-frequency resonance suppression cooperative control method according to claim 4, characterized by, The coupling state space model and the extended Hamilton optimization are combined to dynamically optimize and adjust the damping ratio and gain of each mode through weight distribution and dynamic Nash equilibrium solution, including: A coupling state space model is established to solve the energy mutual feedback problem of the primary and secondary resonance circuits, and the calculation formula is as follows: ， wherein denotes the time derivative of the primary circuit, denotes the time derivative of the secondary circuit, denotes the equivalent line resistance, denotes the primary circuit state, denotes the secondary circuit state, denotes the primary circuit inductance, denotes the DC bus voltage, denotes the control input, v denotes the grid point voltage, denotes the damping branch resistance, denotes the damping branch capacitance, and denotes the energy coupling coefficient; An extended Hamilton function H is established to set the state vector, control quantity, weight, and constraint for energy optimization, and the calculation formula is as follows: wherein denotes the main loop state vector transpose, denotes the state weighting matrix of the main loop, denotes the secondary loop state vector transpose, denotes the state weighting matrix of the secondary loop, denotes the control vector matrix transpose, R denotes the control weight matrix, u denotes the control input vector, denotes the transpose of the constraint matrix, denotes the state self term, denotes the input action term; Through gradient projection-Nash equilibrium, the optimal damping ratio and gain of the system under the current state are solved in real time for the extended Hamilton function H, and the optimal gradient term is as follows: ， wherein denotes the gradient of the Hamiltonian with respect to the control input u, R denotes a control weight matrix, u a control input vector, denotes the transpose of the input matrix.

7. The dual high-frequency resonance suppression cooperative control method according to claim 4, characterized by, The resonance frequency, fractional order parameter and real-time frequency shift rate are input, and the damping ratio and gain after optimization are combined to obtain the dynamic phase compensation signal and alignment signal through fractional order phase compensation, adaptive step adjustment and high-frequency noise suppression processing, including: According to the target resonant frequency and the fractional order parameter, a phase offset synthesizer is constructed, and a phase offset synthesizer parameter and a transfer function are obtained , the transfer function is as follows: ， wherein, represents a proportionality coefficient, represents a Laplacian operator, represents a cut-off frequency, represents an adjustable differential order; Adjustable differential order by real-time frequency shift rate adjustment adaptive update of phase offset synthesizer parameters with adjustment step size The phase compensation signal obtained by the phase offset synthesizer is subjected to high-frequency noise suppression through an IIR low-pass filter to output the dynamic phase compensation signal and the alignment signal.

8. The dual high-frequency resonance suppression cooperative control method according to claim 1, characterized by, The suppression bandwidth, damping ratio and gain, real-time observation data and prior history are input, parameter decoupling recursive identification is performed, and three-layer optimization adaptive optimization is performed to output the globally optimal high-confidence primary and secondary modal parameter set, including: The collected observation data and initial parameters are input, and the high-confidence and smooth fusion primary and secondary modal parameter set is output through parameter decoupling, primary and secondary modal recursive estimation, robustness discrimination and fusion processing. The primary and secondary modal parameters are used as initial values, the system observation data, optimization target and historical cases are input, and adaptive efficient optimization of the parameters is realized through a three-layer hierarchical optimization architecture to output the globally optimal high-confidence primary and secondary modal parameter set meeting the multi-objective constraints.

9. The dual high-frequency resonance suppression cooperative control method according to claim 8, characterized by, The collected observation data and initial parameters are input, and the high-confidence and smooth fusion primary and secondary modal parameter set is output through parameter decoupling, primary and secondary modal recursive estimation, robustness discrimination and fusion processing. Construct the parameter sensitivity matrix Φ, compute its Gram matrix , solve for eigenvalues using Jacobi iteration , obtain all eigenvalues and eigenvectors; The parameter corresponding to the eigenvector with an eigenvalue greater than the eigenvalue threshold is mapped to an independent subspace to obtain the decoupled primary and secondary modal parameter vector. A short-time window strong tracking recursive algorithm is used to recursively update the primary modal parameter vector, and the update rate is dynamically adjusted through the Kalman gain to obtain the estimation result of the current primary modal 4-dimensional parameter, and the state update equation is as follows: ， wherein, denotes the parameter vector estimate at the current time instant k, denotes the parameter estimate at the previous time instant, denotes the gain matrix, denotes the actual observed output at time instant k, denotes the model output mapping; Robust estimation of the sub-modal parameter vector is performed using a long time window, and a Huber loss function is used The Huber loss function is used to suppress the interference of abnormal values and obtain an estimated result of the current 8-dimensional sub-modal parameter The expression is as follows: , wherein, represents the difference between observation and model prediction, represents a switching threshold; Computing the manhattan distance of the primary and secondary modal parameter vector estimates When >2.5, freeze the low-confidence parameter updates, resulting in an effectively updated set of trusted parameters; The updated confident parameter set is weighted and smoothed to obtain the smooth fusion high-confidence primary and secondary modal parameter set.

10. The dual high-frequency resonance suppression cooperative control method according to claim 8, characterized by, The system observation data, optimization target and historical cases are input, and adaptive efficient optimization of the parameters is realized through a three-layer hierarchical optimization architecture to output the globally optimal high-confidence primary and secondary modal parameter set meeting the multi-objective constraints. With the primary and secondary modal parameter set as the initial value, in the 4-dimensional gain parameter space, the improved particle swarm optimization algorithm with quantum tunneling probability is adopted to calculate the difference between the current fitness of the particle and the global optimal solution , the quantum tunneling probability is calculated according to the difference , and the global optimal parameter set is obtained by optimization according to the quantum tunneling probability, and the calculation formula of the quantum tunneling probability is as follows: ， wherein represents the Boltzmann constant, represents a temperature parameter, represents an exponential function; The global optimal parameter set is taken as input, a fuzzy logic rule is applied to combine weights in a 6-dimensional control target space, multi-targets are coordinated and scheduled, and a membership function is constructed The weights are adaptively adjusted according to operating conditions, and a multi-target constraint parameter set is output, and a calculation formula of the membership function is as follows: ， wherein denotes the expected value, denotes the tolerated bandwidth, denotes the steepness of the adjustment function, denotes the actual input variable to be attributed; With multi-objective constraint parameter set as input, the decision case library of historical working conditions and optimal parameters is constructed, and the cosine similarity is used to measure the similarity between current working condition and historical case. If the similarity is greater than the similarity threshold, the historical optimal parameter is directly selected as the initial solution of optimization. If the similarity is lower than the similarity threshold, the global randomization initialization is automatically switched.

Images