Deep learning-based SCUC decision-making: An intelligent data-driven approach with self-learning capabilities

Abstract

This paper proposes an intelligent Deep Learning (DL) based approach for Data-Driven Security-Constrained Unit Commitment (DD-SCUC) decision-making. The proposed approach includes data pre-processing and a two-stage decision-making process. Firstly, historical data is accumulated and pre-processed. Then, the DD-SCUC model is created based on the Gated Recurrent Unit-Neural Network (GRU-NN). The mapping model between system daily load and decision results is created by training the DL model with historical data and then is utilized to make SCUC decisions. The two-stage decision-making process outputs the decision results based on various applications and scenarios. This approach has self-learning capabilities because the accumulation of historical data sets can revise the mapping model and therefore improve its accuracy. Simulation results from the IEEE 118-bus test system and a real power system from China showed that compared with deterministic Physical-Model-Driven (PMD)-SCUC methods, the approach has higher accuracy, better efficiency in the practical use case, and better adaptability to different types of SCUC problems.