A Novel Ensemble Deep Learning Model for Building Energy Consumption Forecast

The issue of energy limitation has gained attention as a crisis faced by societies. Buildings play a major role, in energy consumption making it crucial to accurately predict their energy usage. This prediction problem has led researchers to explore machine learning techniques in the field of energy efficiency. In this study we investigated the performance of used machine learning methods like Random Forest (RF) Multi Layer Perceptron (MLP) Linear Regression (LR) and deep learning methods for predicting building energy consumption. The findings revealed that deep learning outperformed methods in solving this problem. To address this we proposed a voting based solution that combines three CNN models with structures and a Deep Neural Network (DNN) method. We applied our proposed method to the WiDS Datathon dataset and achieved promising results. Each of the deep learning methods used in the proposed method provide suitable results and finally, the voting them is done by the averaging. Due to the fact that the proposed method obtains the final result from voting regression models with high accuracy, it is considered a robust model that will be able to provide a suitable prediction against new data.

Keywords:

energy consumption , Stacking , Regression , Deep Learning , Artificial intelligence