Deep learning-based reactive voltage control method and system for power system

Abstract

The invention relates to a deep learning-based reactive voltage control method and system for a power system. The method comprises the steps of generating data samples corresponding to various operation scenes based on a reactive voltage control method; calling an ICNN deep learning method to carry out deep learning training based on the training sample, and after training is completed, performing deep learning model fitting effect testing based on the test sample; evaluating whether the currently generated deep learning model is suitable for actual control based on the sample average error index and the maximum single sample error index; and when the deep learning model passes index evaluation and then is applied to real-time online calculation, receiving various parameters of model input variables corresponding to a current operation scene in real time in the single calculation process, and completing deep learning single forward calculation to obtain control instructions of reactive equipment. According to the method, the timeliness of reactive voltage control can be greatly improved, and a quicker response can be made for local power grid reactive voltage fluctuation caused by large-scale distributed energy access.

Description

technical field [0001] The invention belongs to the field of power system control, in particular to a method and system for controlling reactive power and voltage of a power system based on deep learning. Background technique [0002] As the proportion of distributed energy such as wind power and photovoltaics in the power grid increases year by year, due to its output volatility and randomness, as well as the characteristics of local high-density grid connection, the reactive power and voltage of the local power grid often occurs in a short period of time. Substantial changes. Taking a wind farm in China as an example, in the operating data of the wind farm, the average fluctuation of the 220kV bus voltage exceeded 6kV within 10s, and the maximum fluctuation exceeded 5kV within 2s. These rapid fluctuations of reactive power and voltage brought about by the access of distributed energy resources have undoubtedly brought severe challenges to traditional reactive power and vo...

AI Technical Summary

Problems solved by technology

However, reinforcement learning is still an optimization method in essence. Some scholars have proved that even if the same hyperparameters and reinforcement learning algorithms are used, only because of the different random seed settings, reinforcement learning methods will appear when solving the same task at different times. In the case of different performance, there is a disadvantage of unstable optimization results

Images