CN-122026358-A - Broadband active damper parameter self-tuning method based on resonance sensing

CN122026358ACN 122026358 ACN122026358 ACN 122026358ACN-122026358-A

Abstract

The invention discloses a broadband active damper parameter self-tuning method based on resonance perception, which comprises the steps of constructing an input eigenvector fused with multiple dimensions, outputting self-tuning parameters through a resonance perception-impedance remodelling type parameter self-tuning neural network, calculating to obtain PI controller parameters and second-order filter parameters, constraining and correcting the PI controller parameters according to a PLL closed-loop damping ratio, ensuring the dynamic stability of a PLL self-control loop, inputting the PI controller parameters and the second-order filter parameters in a stable state into an active damper, enabling the phase presented by the active damper to accurately offset the impedance phase offset problem of the output impedance of the original active damper caused by the fact that the original active damper tracks the power grid voltage phase through a phase-locked loop, and realizing ideal pure resistance characteristics.

Inventors

ZHAO JING
WANG JIWEI
WANG LINA
LUO XIAOMING
LIU LUQING
CHEN WEIZHOU
LI JINGYUE
PEI LIYING
Yu Mengtong

Assignees

江苏科能电力工程咨询有限公司

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (10)

1. A broadband active damper parameter self-tuning method based on resonance perception is characterized by comprising the following steps: S1, acquiring a time-frequency impedance spectrum of a full frequency band of 1-3000Hz, converter hardware parameters and a 50Hz power frequency signal at the current moment in real time, acquiring a plurality of resonance points according to the time-frequency impedance spectrum of the full frequency band, dividing the full frequency band into a target frequency band, a transition frequency band and a non-target frequency band, sequencing the resonance points in each frequency band from large to small according to resonance peaks, and sequentially selecting a preset number of resonance points from the target frequency band and the transition frequency band to serve as key resonance points; s2, preprocessing the hardware parameters of the converter, the 50Hz power frequency signals and the resonance characteristics of key resonance points to construct real-time characteristic vectors; s3, constructing a resonance sensing-impedance remodelling type parameter self-tuning neural network, inputting a feature vector, outputting corresponding self-tuning parameters reflecting the feature vector state, wherein the self-tuning parameters are used for calculating PI controller parameters and second-order filter parameters in the corresponding state, inputting the acquired real-time feature vector into the pre-trained resonance sensing-impedance remodelling type parameter self-tuning neural network, outputting self-tuning parameters reflecting the current real-time feature vector state for remodelling the impedance of the active damper, and calculating PI controller parameters and second-order filter parameters in the real-time state according to the self-tuning parameters, wherein the PI controller parameters comprise PI controller proportional gain of a PLL and PI controller integral gain of the PLL; S4, calculating a PLL closed-loop damping ratio according to the PI controller parameters obtained in the step S3, judging whether the active damper is stable or not according to the PLL closed-loop damping ratio, correcting the PI controller parameters to obtain PI controller parameters in a stable state if the active damper is unstable, and inputting the PI controller parameters in the stable state and the second-order filter parameters in the step S3 into the active damper to compensate the impedance phase of the active damper.
2. The method for self-tuning wideband active damper parameters based on resonance sensing as claimed in claim 1, wherein the step S1 is specifically as follows: For PCC voltage And active damper current Obtaining, namely obtaining a time-frequency impedance spectrum of a full frequency band at the current time through short-time Fourier transform STFT: Wherein the frequency ; Obtaining impedance amplitude |Z (f, t) |, impedance phase |Z (f, t) and resonance peak half-power bandwidth B (f) of each frequency point according to the full-band time-frequency impedance spectrum, obtaining a plurality of resonance points according to the three-dimensional judgment criterion of amplitude-phase-bandwidth, The amplitude criterion is that Z (f, t) is not less than 2Z_rated, Z_rated is rated impedance, The phase criterion is that (d < Z (f, t))/df < 3 >/Hz, | (d < Z (f, t))/df < is the rate of change of impedance phase with frequency, Bandwidth criterion, namely half-power bandwidth B (f) of resonance peak is less than or equal to 50Hz; And a plurality of frequency points meeting all conditions of the three-dimensional judgment criterion of amplitude-phase-bandwidth are marked as resonance points.
3. The method for self-tuning parameters of a wideband active damper based on resonance sensing as claimed in claim 1, wherein the number of key resonance points is 2.
4. The method for self-tuning wideband active damper parameters based on resonance sensing as recited in claim 3, wherein the step S2 is specifically: the converter hardware parameters comprise an active damper side inductance Net side inductor Filter capacitor DC side voltage Inductance Equivalent series resistance of (2) Inductance Equivalent series resistance of (2) Equivalent parallel resistor with capacitor ; The resonance characteristics comprise the frequency, impedance amplitude, impedance phase, phase slope and amplitude gradient of 2 key resonance points; The 50Hz power frequency signal comprises a 50Hz power network voltage power frequency signal and a 50Hz output current power frequency signal, and the 50Hz power network voltage power frequency signal and the 50Hz output current power frequency signal are combined and normalized to obtain one-dimensional power frequency characteristics; and carrying out normalization processing on the hardware parameters, the resonance characteristics and the one-dimensional power frequency characteristics of the converter to generate an 18-dimensional input characteristic vector.
5. The method for self-tuning wideband active damper parameters based on resonance sensing as set forth in claim 1, wherein the PI controller parameters and the second order filter parameters are calculated in step S3, specifically as follows: The self-tuning parameters are: , Calculating proportional correction coefficient of PI controller according to self-tuning parameters Integral correction coefficient of PI controller Second order filter parameters, second order filter center frequency Damping ratio of second order filter And the quality factor Q of the second order filter, Wherein Is a Sigmoid function; Presetting an initial proportional gain of a PI controller of a PLL And an initial integral gain ; Calculating PI controller parameters, namely PI controller proportional gain of PLL PI controller integral gain for PLL 。
6. The method for self-tuning wideband active damper parameters based on resonance sensing as claimed in claim 1, wherein the process of processing the real-time input eigenvectors by the resonance sensing-impedance remodelling type parameter self-tuning neural network in step S3 is as follows: The neural network architecture comprises an input layer, an attention residual layer, a double convolution layer, a full connection layer and an output layer; The input layer receives real-time input feature vectors; The attention residual layer assigns different weights to key resonance points of different frequency bands, Adding residual connection, namely reserving original input characteristics through the residual connection; The double convolution layers are used for extracting the local coupling mode of the input feature, wherein the kernel size of the first convolution layer is 3, the output channel is 32, and the activation function is GELU, the kernel size of the second convolution layer is 3, the output channel is 16, and the activation function is GELU, and the double convolution layers are used for compressing the feature dimension; the full connection layer is used for fitting the nonlinear mapping of the characteristics to the parameters; The output layer outputs 5-dimensional self-tuning parameters [ o1, o2, o3, o4, o5].
7. The method for self-tuning wideband active damper parameters based on resonance sensing as recited in claim 1, wherein in step S4, the PLL closed loop damping ratio is: Wherein, the method comprises the steps of, Is the PI controller proportional gain of the PLL, Integrating the gain for the PI controller of the PLL; PLL closed loop damping ratio The active damper is judged to be stable, and if not, the active damper is judged to be unstable; Correcting parameters of the PI controller, specifically: 、 ; for the PI controller proportional gain of the corrected PLL, For the PI controller integral gain of the corrected PLL, Is the PLL closed loop damping ratio.
8. The method for self-tuning parameters of a wideband active damper based on resonance sensing as recited in claim 1, wherein the parameter tuning process of steps S1-S4 is performed cyclically with a fixed control period of 10ms.
9. The method for self-tuning wideband active damper parameters based on resonance perception according to claim 1, wherein the method further comprises calculating the phase of the active damper target frequency band according to PI controller parameters and second-order filter parameters in a stable state, judging whether the phase compensation requirement of the active damper target frequency band is met, if not, triggering a resonance perception-impedance remodelling type parameter self-tuning neural network online optimization mechanism executed asynchronously with steps S1-S4, if so, recording a current real-time feature vector and a self-tuning parameter in a corresponding stable state, taking the real-time feature vector as input, and taking the self-tuning parameter in the stable state of the active damper as expected output, and constructing a new training sample; the self-tuning parameters in the steady state are obtained according to the PI controller parameters and the second-order filter parameters in the steady state of the active damper; The online optimization mechanism specifically comprises the following steps: According to the data set, the resonance sensing-impedance remodelling type parameter self-setting neural network is updated on line with the aim of minimizing the preset multi-frequency band loss function value, and the updated resonance sensing-impedance remodelling type parameter self-setting neural network is applied to the step S3 of the next control period after the updating is completed; wherein, multiband constraint loss function specifically is: ; open loop impedance at angular frequency for active damper Phase at; For the target frequency band Is a set of angular frequencies of (a); a rated phase of open loop impedance of the active damper; Is of transition frequency band Is a set of angular frequencies of (a); for non-target frequency bands Is a set of angular frequencies of (a); tracking the phase for the target of the PLL; is the actual output phase of the PLL; Is the PLL closed loop damping ratio.
10. The method for self-tuning wideband active damper parameters based on resonance sensing as claimed in claim 9, wherein convergence conditions of the resonance sensing-impedance remodelling type parameter self-tuning neural network are as follows: 、 And ; If any one of the above three convergence conditions is not satisfied, it is determined that the convergence is not achieved.

Description

Broadband active damper parameter self-tuning method based on resonance sensing Technical Field The invention relates to the technical field of smart grids, in particular to a grid-connected system, and specifically relates to a broadband active damper parameter self-tuning method based on resonance sensing. Background With the large-scale access of new energy power generation such as wind power, photovoltaic and the like, the grid-connected inverter is used as a core interface for electric energy conversion, and the stability of the grid-connected inverter faces serious challenges. Under the condition of weak current network, interaction is easy to occur between the impedance of the power network and the output impedance of the grid-connected inverter, broadband resonance is caused, grid-connected current harmonic wave exceeds standard, bus voltage oscillation is caused, and safe and stable operation of power generation equipment and the power network is seriously threatened. Currently, the dominant resonance suppression methods include passive damping and active damping. The passive damping consumes resonance energy through passive elements such as series and parallel resistors in the filter circuit, but introduces extra loss, and suppresses frequency band fixation and poor adaptability. The traditional active damping method simulates virtual resistance in a control loop to restrain resonance by improving an inverter control algorithm, but the method generally needs to change the original control structure of the inverter, increases the complexity of the system and is difficult to coordinate when a plurality of machines are connected in parallel. To solve the above problem, the parallel connection of independent active dampers at the common connection point becomes an effective solution. The method can realize centralized suppression of resonance of a plurality of grid-connected inverters without changing the control structure of the original inverter. However, such active dampers typically require tracking of the grid voltage phase by means of a phase locked loop for synchronous control. The introduction of the phase-locked loop adds a parallel feedback path on the basis of the original control loop, so that the equivalent output impedance of the active damper generates obvious phase shift. In particular in weak networks where the network impedance exhibits a wide frequency variation, this phase shift can severely deteriorate the pure resistive properties that the active damper is expected to exhibit, thereby greatly impairing its ability to suppress mid-to-high frequency resonances and possibly even introducing new risks of instability. The prior art focuses on optimizing the phase-locked loop itself to improve the stability of the grid-connected inverter, but fails to effectively solve the specific problem of phase shift of the output impedance of the active damper caused by the phase-locked loop. Therefore, a method for sensing the resonance characteristics of the system on line and automatically adjusting the parameters of the phase-locked loop and the damper is needed to dynamically compensate the impedance phase, so that the active damper can always maintain excellent pure resistance characteristics and resonance suppression effects in full frequency band, especially in the complex working condition of weak current network. Disclosure of Invention The invention aims to provide a broadband active damper parameter self-tuning method based on resonance perception, which solves the problems that under the condition of weak power grid, the phase-locked loop PLL introduces an additional parallel feedback loop to cause the output impedance of the active damper to generate phase shift, so that the output impedance deviates from ideal pure resistance characteristic and the resonance inhibition capability is weakened. According to the invention, by constructing the resonance sensing-impedance remodelling type parameter self-tuning neural network, sensing the broadband resonance characteristic of the system on line, and dynamically tuning the phase-locked loop and the damper compensation parameter, the phase shift of the output impedance is accurately compensated, so that the active damper always maintains excellent pure resistance characteristic under the complex working condition, and finally stable and efficient broadband resonance suppression is realized. The technical scheme adopted by the invention is as follows: A broadband active damper parameter self-tuning method based on resonance sensing, the method comprising the steps of: S1, acquiring a time-frequency impedance spectrum of a full frequency band of 1-3000Hz, converter hardware parameters and a 50Hz power frequency signal at the current moment in real time, acquiring a plurality of resonance points according to the time-frequency impedance spectrum of the full frequency band, dividing the full frequency band into a target frequency band, a transi