Short-Term Load Forecasting Using Long Short-Term Memory Based on EVN NLDC Data

Abstract

Load forecasting has always been a crucial part of an efficient power system planning an operation. Since the global electricity market has
developed rapidly in recent years, a load sequence has gradually become non-stationary and hence makes accurate forecasting difficult.
Nowadays, many techniques can be used for load forecasting such as fuzzy logic, similarday approach, expert systems, etc.. However, these
methods normally obtain undesired results due to considerable variation of electricity load. In this paper, a short-term load forecasting model is
proposed based on Long Short-Term Memory (LSTM) network combining linear regression algorithm and EVN-NLDC (National Load
Dispatch Centre) data. Short-term load forecasting has an enormous impact on unit commitment, strategic power reserve for national power
system and enhances the reliability of the power supply. National load data could be considered as a time series sequence consisting of two
components which are trend and residual. Linear regression (LR) is adopted for trend forecasting and LSTM is used to residual forecasting. In
addition, to evaluate the performance of the proposed model, artificial neural network (ANN) is utilized as a benchmark model. The results
show that the proposed model achieves the Mean Absolute Percentage Error (MAPE) ranged from 1% to around 4% for one-day ahead load
forecasting, whereas ANN model obtains over 4%.