Short Term Power Load Forecasting Based on Deep Neural Networks

Abstract

The purpose of this study is to find the most appropriate forecasting method by applying the machine learning and deep learning techniques that have recently been representing an outstanding performance in various fields to power load forecasting and evaluating their performance. Forecasting model has been realized by using logistic regression, decision tree, support vector machine (SVM) algorithm as the machine learning technique, and deep neural network (DNN) algorithm as deep learning technique and compared with each other. In order to find the most appropriate method for power load forecasting, the performance of machine learning and deep learning model was compared and evaluated. Performance was evaluated by realizing total 7 forecasting models including 3 machine learning-based single forecasting models, 1 deep learning-based single forecasting model, and 3 complex forecasting models. As for complex forecasting model, forecasting rate turned out to be 96.91% for logistic regression-based complex forecasting model, 97.08% for decision tree-based complex forecasting model, and 96.43% for support vector machine-based forecasting model that the complex forecasting model combined with decision tree and deep neural network represented the most outstanding performance. With this study, it is anticipated to precisely forecast power load saving the electronic energy while preparing for a plan to efficiently distribute and utilize energy in connection with smart grid technology such as Energy Storage System (ESS) or Energy Management System (EMS).