Short Term Load Forecasting Using LSTM Neural Networks

Abstract

Electricity load forecast plays an important role in smart grid applications to promote the decision-making process for energy generation and consumption. Long-term forecasting is not feasible as there may be uncertainty in the forecast owing to an increase in the inclusion of renewable sources into current grids. Since the behavior of the load is highly non-linear and seasonal, Neural Networks is the best model for studying non-linear behavior within data and for forecasting purposes. Hence this paper presents an enhanced Long Short-Term Memory (LSTM) neural network model, which is used to forecastthe closing electricity load for the future interval. This paperinvestigates the performanceanalysis for optimal selectionof layers withhidden units’ combination andalso discussesstatistical analysis for selection of the optimal LSTM architecture. Finally, deploy the best-adapted configuration with the lowest absolute percentage error and optimized network architecture.