Emerging and Evaluating Long Short Term Memory (LSTM) Network for Load Forecast in Java Bali System

The aftermath of the COVID-19 pandemic in 2019 resulted in a decrease in Java Bali's total load instead of the usual increasing trend. The loads also exhibit different characteristics in their daily, weekly, and monthly load profiles in each region. The basic statistical coefficient method used to forecast the load introduces a higher possibility of error and inaccuracies in operational planning. A different approach is necessary to achieve higher accuracy in load forecasting. One method to predict reliable trends is deep learning, a subfield of machine learning, which can synthesize the learning curve based on available data. A method called Long-Short Term Memory (LSTM), included in Deep Learning and popularized by researchers since 2000, has shown better accuracy in forecasting. This paper focuses on reviewing the LSTM method for short-term load forecasting in the Java Bali power system using several additional inputs. The method demonstrates an accurate learning curve after the addition of several input parameters.