Search

CN-122020287-A - Hydroelectric generating set low-frequency oscillation multidimensional identification and intelligent early warning method based on PMU data, electronic equipment and readable storage medium

CN122020287ACN 122020287 ACN122020287 ACN 122020287ACN-122020287-A

Abstract

The invention discloses a low-frequency oscillation multidimensional identification and intelligent early warning method of a hydroelectric generating set based on PMU data, and belongs to the technical field of source network coordination. The method comprises the steps of obtaining a stable signal through CEEMDAN denoising and related coefficient screening through a four-stage process of preprocessing, characteristic enhancement, intelligent identification and hierarchical early warning, optimizing VMD parameters through CWOA, combining the random resonance enhancement weak oscillation characteristics of a Duffing oscillator, extracting 12-dimensional characteristics of a time domain, a frequency domain and an energy domain, inputting an Attention-LSTM model to realize 3-class oscillation mode classification and parameter calculation, constructing a four-stage early warning system based on a damping ratio and an oscillation amplitude, and predicting trend by combining an exponential smoothing method. The invention solves the difficult problems of non-stationary signal processing and weak feature extraction, the mode identification accuracy rate reaches 98.3%, the single-group data processing time is less than or equal to 300ms, the weak signal extraction success rate is more than or equal to 95.2% under 10dB signal-to-noise ratio, millisecond response and accurate early warning can be realized, and the invention is suitable for high-proportion new energy access scenes.

Inventors

  • LIU HAIBO
  • WANG BENHONG
  • GUO JIANG
  • ZHANG FANGQING

Assignees

  • 中国长江电力股份有限公司

Dates

Publication Date
20260512
Application Date
20260114

Claims (10)

  1. 1. A hydroelectric generating set low-frequency oscillation multidimensional identification and intelligent early warning method based on PMU data is characterized by comprising the following steps: Step S1, data preprocessing and self-adaptive denoising, namely adding Gaussian white noise to PMU original signals to construct a noise auxiliary signal set, decomposing through CEEMDAN modes to obtain a plurality of IMF eigenmode functions, calculating correlation coefficients ρi of each IMF and the original signals, and reserving component reconstruction that ρi is more than or equal to 0.3 to obtain preprocessed signals ; CWOA-VMD parameter optimization and characteristic enhancement, namely optimizing the decomposition layer number K and penalty factor alpha of Variation Modal Decomposition (VMD) through a chaos improvement whale optimization algorithm (CWOA), performing secondary decomposition on x' (t), inputting a modal component containing dominant oscillation characteristics into a Duffing vibrator system, adjusting driving force frequency to realize stochastic resonance, and extracting instantaneous amplitude A (t) and instantaneous frequency f (t) through Hilbert transformation; step S3, extracting multi-dimensional characteristics, namely extracting time domain 6-dimensional characteristics, frequency domain 4-dimensional characteristics and energy domain 2-dimensional characteristics, wherein the total number of the 12-dimensional characteristic vectors is 12; Step S4, intelligent recognition, namely inputting a characteristic vector into an Attention-LSTM model to realize 3 types of mode classification of hydraulic oscillation, electromechanical oscillation and power grid disturbance oscillation and frequency and damping ratio sigma parameter calculation; And S5, carrying out hierarchical early warning and trend prediction, namely constructing a four-level early warning system based on sigma and A, predicting the variation trend of the oscillation amplitude by adopting an exponential smoothing method, and outputting early warning information comprising an oscillation mode, early warning grades and scheduling suggestions.
  2. 2. The method for identifying and intelligently pre-warning the low-frequency oscillation multidimensional of the hydroelectric generating set based on PMU data according to claim 1, wherein the chaos improved whale optimizing algorithm in the step S2 generates an initial population through Logistic chaos mapping, and the mapping formula is as follows: , ; Wherein, the The value of the chaotic variable generated for the (i+1) th iteration is that mu is a chaotic control parameter and takes the value of 4, For the value of the chaotic variable for the ith iteration, The self-adaptive inertial weight w is introduced as a chaos initial value, and the calculation formula is as follows: ; Wherein the method comprises the steps of =0.9, =0.4, T is the current iteration number, =100。
  3. 3. The method for identifying and intelligently pre-warning the low-frequency oscillation multidimensional of the hydroelectric generating set based on PMU data according to claim 1, wherein in the step (2), the parameter search range of the VMD is K [ epsilon ] [2,10], alpha [ epsilon ] [100,2000], and the fitness function of CWOA is as follows: ; wherein λ=0.6, snr is the modal component signal-to-noise ratio, B k is the bandwidth of the kth modal component, max # ) Is the maximum bandwidth value among all modal components.
  4. 4. The method for identifying and intelligently pre-warning the low-frequency oscillation multidimensional of the hydroelectric generating set based on PMU data according to claim 1, wherein in the step S2, the equation of the Duffing vibrator is as follows: ; Wherein the damping coefficient c=0.1, the driving force amplitude f=0.5, the fundamental frequency ω 0 =1 hz, s (t) is a weak oscillation signal after VMD decomposition, and n (t) is residual noise.
  5. 5. The method for identifying and intelligently pre-warning the low-frequency oscillation multidimensional of the hydroelectric generating set based on PMU data according to claim 1 is characterized in that in the step S3, the time domain features comprise peaks, kurtosis, peak factors, waveform factors, mean values and variances, the frequency domain features comprise fundamental frequency f 0 , harmonic amplitude ratios, frequency spectrum barycenters and frequency band energy ratios, the energy domain features comprise modal energy duty ratios and sample entropy, wherein the embedding dimension=2 of the sample entropy and the similarity tolerance=0.2 times of standard deviation.
  6. 6. The method for identifying and intelligently pre-warning the low-frequency oscillation multidimensional of the hydroelectric generating set based on PMU data according to claim 1, wherein in the step S4, the input sequence length=30 of the Attention-LSTM model comprises 2 LSTM hidden layers, 64 neurons in each layer are activated as tanh, an Attention mechanism adopts additive Attention, and Attention weight is calculated: output context feature ; Wherein, the Is an attention score; 、 、 a trainable weight matrix for the model; is a bias term; a hidden layer state for LSTM; Is an input feature vector; is the normalized attention weight; Is a normalized exponential function; A weighted context feature vector; Is an input feature.
  7. 7. The method for identifying and intelligently warning the low-frequency oscillation multidimensional degree of the hydroelectric generating set based on PMU data according to claim 1 is characterized in that in the step S5, the threshold standard of the four-stage warning system is that a normal state sigma is more than or equal to 0.05 and A is more than or equal to 0.05p.u., blue warning is more than or equal to 0.03 and less than or equal to sigma is more than or equal to 0.05 and A is more than or equal to 0.1p.u, yellow warning is more than or equal to 0.01 and less than or equal to sigma is more than or equal to 0.03 and A is more than or equal to 0.2p.u, and red warning is more than or equal to sigma is more than or equal to 0.01 and A is more than or equal to 0.2p.u.
  8. 8. The method for identifying and intelligently pre-warning the low-frequency oscillation multidimensional of the hydroelectric generating set based on PMU data according to claim 1, wherein in the step S5, a prediction formula of the exponential smoothing method is as follows: Â(t+T)=αA(t)+(1-α)Â(t); Wherein  (t+T) is the predicted value of oscillation amplitude at the moment T in the future, alpha is a smoothing coefficient, A (T) is the actual observed amplitude at the moment,  (T) is the smoothed predicted value at the moment; is the prediction step size.
  9. 9. An electronic device, the electronic device comprising: A processor (22) and a memory (21), wherein the memory (21) is in communication connection with the processor (22); The memory (21) is configured to store executable instructions executed by at least one processor (22), and the processor (22) is configured to execute the executable instructions to implement a method for identifying and intelligently pre-warning low-frequency oscillation multidimensional of a hydroelectric generating set based on PMU data according to any one of claims 1 to 8.
  10. 10. A computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and when the computer program is executed by a processor, the method for identifying and intelligently pre-warning the low-frequency oscillation multidimensional of the hydroelectric generating set based on PMU data is realized according to any one of claims 1 to 8.

Description

Hydroelectric generating set low-frequency oscillation multidimensional identification and intelligent early warning method based on PMU data, electronic equipment and readable storage medium Technical Field The invention relates to the technical field of source network coordination, in particular to a hydroelectric generating set low-frequency oscillation multidimensional identification and intelligent early warning method based on PMU data. Background At present, the related technology of low-frequency oscillation identification and early warning of a power system mainly focuses on three directions of signal processing, parameter optimization and mode identification, and related research has partial shortages. In the aspect of signal processing, the prior art mainly adopts methods such as Empirical Mode Decomposition (EMD), variational Mode Decomposition (VMD), S transformation and the like. For example, patent CN115270342a discloses a low-frequency oscillation mode identification method of an electric power system based on CEEMDAN algorithm, which decomposes signals through CEEMDAN and extracts oscillation parameters by combining hilbert yellow transformation, but the method does not solve the problem of parameter self-adaptive optimization of mode decomposition and has limited extraction capacity on weak oscillation signals, and literature 'low-frequency oscillation transient detection method based on optimization window function improvement S transformation' proposes improvement of S transformation through double-parameter optimization Gaussian window function, thereby improving frequency resolution, but still has the problem of insufficient time-frequency focusing in non-stationary signal processing. In terms of parameter optimization, traditional optimization algorithms such as Particle Swarm Optimization (PSO) and basic Whale Optimization Algorithm (WOA) are applied to VMD parameter selection, but these types of algorithms are prone to be in local optimization, so that modal decomposition accuracy is insufficient. For example, the decomposition layer number K and the penalty factor α of the VMD in the prior art are determined by empirical setting or a basic optimization algorithm, and are difficult to adapt to the dynamic change characteristics of the PMU data of the hydropower station. In terms of pattern recognition and early warning, the prior art mostly adopts a single feature or a traditional machine learning model. Patent CN119754986A proposes to eliminate forced oscillation by monitoring the pressure pulsation frequency of a draft tube and adjusting the opening degree of a guide vane, but the method focuses on the elimination of oscillation instead of early warning, and does not consider a multi-unit collaborative monitoring scene, patent CN118920341B adopts an LSTM neural network to perform modal identification, but does not introduce a characteristic attention mechanism, and has insufficient attention to key oscillation characteristics, so that the identification accuracy is low under complex working conditions. Disclosure of Invention The invention aims to provide a low-frequency oscillation multidimensional identification and intelligent early warning method AAA of a hydroelectric generating set based on PMU data, so as to solve the problems in the background technology. In order to achieve the purpose, the invention provides the following technical scheme that the low-frequency oscillation multidimensional identification and intelligent early warning method for the hydroelectric generating set based on PMU data comprises the following steps: Step S1, data preprocessing and self-adaptive denoising, namely adding Gaussian white noise to PMU original signals to construct a noise auxiliary signal set, decomposing through CEEMDAN modes to obtain a plurality of IMF eigenmode functions, calculating correlation coefficients ρi of each IMF and the original signals, and reserving component reconstruction that ρi is more than or equal to 0.3 to obtain preprocessed signals ; CWOA-VMD parameter optimization and characteristic enhancement, namely optimizing the decomposition layer number K and penalty factor alpha of Variation Modal Decomposition (VMD) through a chaos improvement whale optimization algorithm (CWOA), performing secondary decomposition on x' (t), inputting a modal component containing dominant oscillation characteristics into a Duffing vibrator system, adjusting driving force frequency to realize stochastic resonance, and extracting instantaneous amplitude A (t) and instantaneous frequency f (t) through Hilbert transformation; step S3, extracting multi-dimensional characteristics, namely extracting time domain 6-dimensional characteristics, frequency domain 4-dimensional characteristics and energy domain 2-dimensional characteristics, wherein the total number of the 12-dimensional characteristic vectors is 12; Step S4, intelligent recognition, namely inputting a characteristic vector in