Deep Neural Network Based Hierarchical Control of Residential Microgrid Using LSTM

Abstract

Microgrid is a hot topic for the present research as their role is significant in framing reliable and efficient power system. Major sources in a typical microgrid are renewable resources like solar, wind etc. Their intermittency and uncertainty in the load demands makes the control and smooth operation of microgrid challenging. This paper presents a Long Short Term Memory (LSTM) network for hierarchical control of a residential microgrid. This multi input multi output LSTM regression architecture is used to predict the optimal real time control action for the microgrid. The performance of the LSTM model is evaluated using root mean square error (RMSE), SMAPE, MRE, MAE, RMSE and loss function. Its performance is compared with other prominent techniques also.