Analysis of the Consumption and Sensors Features Contribution to the Consumption Forecast Using Explainable AI in Buildings

The energy sector nowadays has a high impact in the distribution of energy around the world. Moreover, the uncertainty of the energy and user behavior are a threat that causes the difficulty on the planning of an optimization plan intended to improve the energy efficiency as high as possible. Data scientists and machine learning professionals recommend the use of deep learning and machine learning algorithms to train an historic with energy consumption from past time and additional variables (sensors of a building for example) to forecast the energy consumption for the intended target period. Several forecasting methods may be used such as Artificial Neural Networks, K-Nearest Neighbors, and XGBoost. Experts in the explainable artificial intelligence area recommend the use of methods such as Shapley Addictive Explanations (SHAP) in order to explain the input features of the historic dataset with a stronger influence on the output power consumption.