Predictive analysis of Low power DC loads in Residential Buildings

Abstract

Abstract There has been a constant rise in use of loads powered from direct current (DC) among the residential loads. In the traditional Alternating Current (AC) grid voltage undergoes conversion multiple times before it reaches the loads requiring DC, which leads to substantial power loss in the converters. Implementation of a DC grid in the residential buildings can reduce the losses caused due to the conversions. The paper presents a prediction model for domestic low power DC loads which can be powered from a battery. The data of usage of the load as well as the data of the parameters on which the load usage is dependent on was collected over a period of a year. This data was fed to a Random forest-based machine learning algorithm to predict the usage of the loads in future. The state of charge (SOC) of the battery used to power the loads was compared with the load prediction to decide whether the load can be powered from the available power in the battery or not. This allowed the system to decide in real time on the source to be used to power the DC loads and providing the users a choice to change the load usage time in order to prevent power from being drawn from the AC grid.