Energy Consumption Forecasting using a Deep Learning Energy-Level Based Prediction

Smart grids benefit the consumers via enabling informed participation. As a result, customers have more choices, which drives a range of buying trends and actions. One such important application is accurate forecasts of energy demands (loads) at individual sites, but load forecasting is still a challenge. Most load forecasting research has used traditional numerical value estimation, which is a regression style prediction; however, energy consumers will not know if the forecasted load is low or high. Level prediction can be performed by performing regression first and then classifying the values into energy levels, or by using a classification model to predict the levels directly. The Random Forest (RF) classification algorithm is shown to have outperformed other machine learning algorithms for level predictions. This paper provides a deep learning-based classification approach. The results of this deep learning-based classification were compared to those of the RF classification algorithm, it was observed that the deep learning method outperformed the RF method, particularly for high numbers of energy levels.