Non-Linear Auto-Regressive Modeling based Day-ahead BESS Dispatch Strategy for Distribution Transformer Overload Management

Abstract

Distribution Transformer (DT) overload management is a promising application for Battery Energy Storage Systems (BESSs) in urban areas with space constraints and growing load. However, the BESS operation must be optimized so as to ensure minimum impact on both the battery cycling and the DT life. Hence, a day-ahead dispatch strategy can allow the BESS to maintain its state of charge and identify the crucial load-peaks to cater to ensuring minimum stress on the DT. The present work proposes a non-linear autoregressive with exogenous input (NARX) framework for short-term load forecasting. The nonlinearity is approximated by an artificial neural network. The proposed method uses past electricity consumption data of a distribution utility in New Delhi, India and the corresponding weather data to predict the future load demand on a particular DT serving a locality. The results obtained from the proposed method are used for defining the charging/discharging level of the BESS on a day-ahead basis to minimize the transformer loss-of-life. The results obtained from the proposed NARX model are encouraging and the model successfully forecasts the load for three days with a mean absolute percentage error (MAPE) of 6.17%.