CN-122026395-A - New energy power grid self-adaptive frequency modulation optimization method

CN122026395ACN 122026395 ACN122026395 ACN 122026395ACN-122026395-A

Abstract

The invention relates to the field of power grid frequency modulation and discloses a new energy power grid self-adaptive frequency modulation optimization method which comprises the steps of firstly obtaining multi-source operation time sequence data of a new energy power grid, extracting characteristics to generate a power grid operation characteristic vector, analyzing dynamic statistical deviation of the power grid operation characteristic vector in a preset time window to obtain a comprehensive frequency disturbance index, identifying a dynamic disturbance event and early warning, generating a candidate frequency modulation parameter set according to a warning result, associating and matching a disturbance-frequency modulation response mode library to search similar case optimization parameters in a historical event database to generate an optimal frequency modulation parameter sequence, combining power grid real-time state and external environment prediction data to generate a frequency modulation resource optimization configuration scheme set, inputting a power grid evolution prediction model, outputting each scheme prediction result and evaluation indexes, selecting an optimal scheme to execute according to the evaluation indexes, generating a warning signal and triggering a new round of optimization configuration to accurately identify disturbance and optimize frequency modulation resource configuration, and guaranteeing power grid stability and safety.

Inventors

LIU XINGJUN
FANG JUNWU
ZHAO LICHUN
YANG CHUNLI
ZHANG WEI
FU ZHENG
ZHANG QINGGUO
GE YU
ZHANG JIAN

Assignees

国家电投集团东北电力开发有限公司

Dates

Publication Date: 20260512
Application Date: 20251222

Claims (8)

1. The adaptive frequency modulation optimization method for the new energy power grid is characterized by comprising the following steps of: s1, acquiring multi-source operation time sequence data of a new energy power grid, and extracting characteristics of the acquired multi-source operation time sequence data to generate a power grid operation characteristic vector; S2, obtaining a comprehensive frequency disturbance index by analyzing dynamic statistical deviation of the power grid operation feature vector in a preset time window, thereby identifying dynamic disturbance events which threaten the power grid frequency and triggering disturbance event early warning; S3, according to a triggering result of disturbance event early warning, carrying out association matching on the power grid operation feature vector and a preset disturbance-frequency modulation response mode library to generate a group of candidate frequency modulation parameter sets; s4, according to the candidate frequency modulation parameter set, multi-mode similarity retrieval is carried out in a historical event database to obtain a group of historical treatment cases with similar frequency modulation parameters applied as historical treatment evidence, and the candidate frequency modulation parameters are optimized based on the historical treatment evidence to generate an optimal frequency modulation parameter sequence; S5, generating a corresponding frequency modulation resource optimal configuration scheme for each frequency modulation parameter in the optimal frequency modulation parameter sequence according to the optimal frequency modulation parameter sequence by combining the real-time state of each frequency modulation resource in the current power grid and external environment prediction data in a preset time window, so as to form a frequency modulation resource optimal configuration scheme set; S6, inputting each frequency modulation resource optimization configuration scheme in the frequency modulation resource optimization configuration scheme set and the current power grid operation feature vector into the constructed power grid evolution prediction model, directly outputting the recovery track of the power grid frequency and the diffusion risk prediction result of the disturbance event after the power grid frequency is executed, and generating the prediction effect evaluation index of each power grid; And S7, selecting a frequency modulation resource optimal allocation scheme with the best effect as a final scheme to be executed according to the prediction effect evaluation index, generating a dynamic early warning upgrading signal if the prediction effect of all schemes in the scheme set does not reach a preset threshold value, and returning to S4 to trigger a new round of frequency modulation optimal allocation.
2. The adaptive frequency modulation optimization method of a new energy power grid according to claim 1, wherein the step S1 is to synchronously collect output time sequence data, power grid frequency deviation and change rate time sequence data, regional load fluctuation time sequence data and key tie power and node voltage phase angle time sequence data of wind, light, water, fire and storage multi-type power sources in real time through a wide area measurement system and a multi-source data collection terminal which are deployed in the new energy power grid to form a multi-source heterogeneous time sequence data set; Performing self-adaptive decomposition on each data sequence in the multi-source heterogeneous time sequence data set by adopting variation modal decomposition to separate out eigen-modal functions of different time scales, calculating the corresponding sample entropy of each eigen-modal function to quantify the complexity of the sequence, and simultaneously extracting time domain statistical characteristics, frequency domain spectral characteristics and nonlinear dynamics characteristics of each original data sequence; and finally, fusing the sample entropy of each eigenmode function, the time domain statistical characteristic, the frequency domain spectral characteristic and the nonlinear dynamic characteristic to generate a power grid operation characteristic vector.
3. The adaptive frequency modulation optimization method of a new energy power grid according to claim 2, wherein S2 is characterized in that based on the generated power grid operation feature vector, in a preset sliding time window, the mean value and covariance matrix of each dimension feature in the feature vector are calculated to construct a reference feature vector representing the current steady state operation level of the power grid, then the mahalanobis distance between the power grid operation feature vector at the current moment and the reference feature vector is calculated in real time, wherein the mahalanobis distance is used for comprehensively measuring the overall deviation degree of the feature vector in a multidimensional space, the mahalanobis distance is input into an S-shaped function to be normalized, and a value range is generated And when the comprehensive frequency disturbance index exceeds a preset early warning threshold value, the preset early warning threshold value is determined based on historical disturbance data statistics, then a dynamic disturbance event which threatens the stability of the power grid frequency is identified, and disturbance event early warning is immediately triggered, otherwise, the power grid frequency is continuously monitored.
4. The self-adaptive frequency modulation optimization method of the new energy power grid according to claim 3, wherein after the disturbance event early warning is triggered, S3 inputs a power grid operation characteristic vector generated in real time into a disturbance-frequency modulation response mode library which is constructed in advance for matching, the disturbance-frequency modulation response mode library divides the power grid operation characteristic vector in the historical disturbance data into a plurality of typical disturbance scenes through a clustering algorithm, and an optimal frequency modulation parameter combination which is obtained based on historical frequency modulation effect evaluation index optimization is stored for each typical scene; the matching process specifically comprises the steps of calculating the similarity between a current power grid operation feature vector and clustering center vectors of typical scenes in a mode library by adopting a weighted cosine similarity algorithm, wherein the weight is preset according to the influence degree of each dimension in the feature vector on the frequency stability, and finally, selecting optimal frequency modulation parameter combinations corresponding to the first N typical scenes with the highest similarity to form a group of candidate frequency modulation parameter sets.
5. The adaptive frequency modulation optimization method of a new energy grid according to claim 4, wherein S4 uses each group of parameters in the candidate frequency modulation parameter set as a query index, and performs multi-mode similarity search in a historical event database, wherein the search is implemented by calculating a comprehensive similarity index, and the comprehensive similarity index is a weighted sum of euclidean distance of frequency modulation parameter vectors, cosine similarity of initial running state vectors of the grid and matching degree of disturbance type labels, so as to obtain a set of historical treatment cases with highest comprehensive similarity as historical treatment evidence; The method comprises the steps of constructing a multi-objective optimization function which aims at minimizing frequency recovery time and frequency overshoot and considering frequency modulation cost based on historical treatment evidence, solving the multi-objective optimization function through an iterative optimization algorithm, wherein the iterative optimization algorithm is configured to be capable of processing a plurality of mutually conflicting targets, reserving optimized non-inferior solutions in an iterative process, finally outputting a candidate solution set consisting of a plurality of non-inferior solutions, and finally selecting one solution from the candidate solution set as a final optimal frequency modulation parameter sequence according to preset dynamic performance and economic weight coefficients.
6. The adaptive frequency modulation optimization method of a new energy power grid according to claim 5, wherein S5 takes the optimal frequency modulation parameter sequence as a total frequency modulation instruction, and constructs a multi-objective distribution model with the objective of minimizing the overall frequency modulation cost and frequency modulation resource adjustment deviation and with the constraint of the adjustment rate, response time and real-time available capacity of each frequency modulation resource; acquiring real-time state data of each frequency modulation resource in a current power grid through a real-time data acquisition system, and acquiring external environment prediction data in a preset time window from a prediction data interface to be input into the multi-target distribution model, wherein the external environment prediction data are used for predicting availability change of future frequency modulation resources and comprise at least one of weather prediction data, sunlight intensity prediction data, wind speed prediction data, temperature prediction data or holiday load characteristic data; And solving the multi-target distribution model by adopting a quadratic programming algorithm, and dynamically calculating the optimal output distribution proportion of different frequency modulation resources for each parameter in the optimal frequency modulation parameter sequence, thereby forming a frequency modulation resource optimal configuration scheme set.
7. The adaptive frequency modulation optimization method of a new energy power grid according to claim 6, wherein each scheme in the frequency modulation resource optimization configuration scheme set is input together with a current power grid operation feature vector into a pre-trained power grid evolution prediction model to predict power grid dynamic response after each scheme is executed, wherein the power grid evolution prediction model is a sequence-to-sequence prediction model constructed based on a long-short-term memory network or a gating circulation unit, and the model is trained by using a corresponding relation between frequency modulation actions recorded in historical operation data and power grid frequency response through a supervised learning algorithm; And finally, generating a prediction effect evaluation index of each scheme based on the frequency recovery track and the diffusion risk quantization index.
8. The adaptive frequency modulation optimization method of a new energy power grid according to claim 9, wherein the step S7 is characterized in that all schemes in the frequency modulation resource optimization configuration scheme set are ordered according to the prediction effect evaluation indexes thereof, and the scheme with the optimal prediction effect evaluation indexes is selected as a final scheme; Comparing the predicted effect evaluation index of the optimal scheme with a preset safety risk threshold, if the predicted effect evaluation index is larger than or equal to the preset safety risk threshold, judging that the current scheme set cannot meet the safety and stability operation requirement of the power grid, generating a dynamic early warning upgrading signal at the moment, and triggering to return to S4 so as to start a new frequency modulation optimization configuration.

Description

New energy power grid self-adaptive frequency modulation optimization method Technical Field The invention relates to the technical field of power grid frequency modulation, in particular to a new energy power grid self-adaptive frequency modulation optimization method. Background With the promotion of the 'double carbon' target, the permeability of new energy sources represented by wind power and photovoltaic in a power grid continuously rises, and inherent volatility and randomness of the new energy sources bring unprecedented challenges to the frequency stabilization of the power grid. The traditional frequency modulation technology mainly relies on a synchronous generator set, and the defects of slow response and single resource are difficult to cope with high-frequency and large-amplitude power disturbance. For this reason, the prior art has turned to diversified frequency modulation resources including energy storage, virtual power plants, etc., and introduced optimization algorithms based on Model Predictive Control (MPC) or Reinforcement Learning (RL). However, the above method still has the following drawbacks: Firstly, the traditional method relies on a fixed threshold value or an off-line model, so that dynamic disturbance such as new energy output fluctuation, load mutation and the like is difficult to capture in real time, and frequency modulation instruction generation delay and frequency recovery speed are slow; Secondly, the existing method generally adopts preset fixed frequency modulation parameters (such as dead zone and proportionality coefficient) and cannot be dynamically adjusted according to disturbance types and power grid states, so that frequency modulation resource waste or insufficient effect is caused; Thirdly, single resources such as energy storage, thermal power and the like are independently scheduled by the traditional method, and the resource capacity, response speed, cost and external environment (such as meteorological data) are not comprehensively planned, so that the configuration scheme is poor in economy or not feasible; Fourth, the existing method only focuses on the immediate correction of frequency deviation, and the influence of frequency modulation action on other nodes of the power grid is not evaluated, so that disturbance can be diffused to weak links, and cascading failure is caused. Disclosure of Invention In order to overcome the defects in the prior art, the invention provides a new energy power grid adaptive frequency modulation optimization method, which aims to solve the problems in the background art. The invention provides a new energy power grid self-adaptive frequency modulation optimization method, which comprises the following steps: s1, acquiring multi-source operation time sequence data of a new energy power grid, and extracting characteristics of the acquired multi-source operation time sequence data to generate a power grid operation characteristic vector; S2, obtaining a comprehensive frequency disturbance index by analyzing dynamic statistical deviation of the power grid operation feature vector in a preset time window, thereby identifying dynamic disturbance events which threaten the power grid frequency and triggering disturbance event early warning; S3, according to a triggering result of disturbance event early warning, carrying out association matching on the power grid operation feature vector and a preset disturbance-frequency modulation response mode library to generate a group of candidate frequency modulation parameter sets; s4, according to the candidate frequency modulation parameter set, multi-mode similarity retrieval is carried out in a historical event database to obtain a group of historical treatment cases with similar frequency modulation parameters applied as historical treatment evidence, and the candidate frequency modulation parameters are optimized based on the historical treatment evidence to generate an optimal frequency modulation parameter sequence; S5, generating a corresponding frequency modulation resource optimal configuration scheme for each frequency modulation parameter in the optimal frequency modulation parameter sequence according to the optimal frequency modulation parameter sequence by combining the real-time state of each frequency modulation resource in the current power grid and external environment prediction data in a preset time window, so as to form a frequency modulation resource optimal configuration scheme set; S6, inputting each frequency modulation resource optimization configuration scheme in the frequency modulation resource optimization configuration scheme set and the current power grid operation feature vector into the constructed power grid evolution prediction model, directly outputting the recovery track of the power grid frequency and the diffusion risk prediction result of the disturbance event after the power grid frequency is executed, and generating the prediction effect eval