국민DR 참여활성화를 위한 머신러닝 기반의 전력수요 예측에 관한 연구

Abstract

Power demand forecasting requires a number of independent variables, and there are many prior technologies that predict through deep learning and machine learning in consideration of weather, temperature, humidity, and environmental effects. There are many artificial intelligence models for power demand prediction, but among them, errors occurred in inferring the dependent variable affecting the independent variable by constructing a regression analysis and a multilayer neural network, and in order to solve this problem, the experiment was conducted by adopting a method of reducing the error rate by repeatedly using machine learning. Machine learning was conducted based on one year of data from factories in industrial complexes on actual power use data. In this study, without adopting independent variables, a multilayer neural network was constructed through a training dataset with actual power use data to infer power predictions through machine learning, and daily, weekly, and monthly power usage was predicted. Through this, we present a direction for activating participation in the small commercial DR(Demand Response)