Operational Planning for Home Energy Management System using Machine Learning

Abstract

Demand Response (DR) that controls energy consumption based on a margin of power supply capability has been drawing increasing attention. In the residential sector, however, aggregators have to aggregate saving power because the capacity of saving power per customer is less than the industrial sector. Because of this, improving the calculation time that is spent on deriving an operation plan is necessary. Therefore, we derive operation plans using Convolutional Neural Network (CNN). And then, we analyze the classifier to clear up the efficient operational logic of residential energy equipment. We target a household that an air conditioner and polymer electrolyte fuel cell co-generation system (PEFC-CGS) are installed. To create CNN that reduces operational cost, we added a term for the operational cost to the loss function which is used during learning. After that, we analyze constructed CNN by LIME (Local Interpretable Model-agnostic Explanations). As a result, an average relative error rate from the optimal result is decreased to 5.67 %. We found that the hot water demand, the amount of remaining hot water and operation state have a big impact on the operation plan of PEFC while the electricity demand has no impact on it.