Electricity load forecasting using long short-term memory: Case study from Central Java and DIY

Forecasting power system loads refers to the study or use of a mathematical method for methodically processing historical and future loads, while taking capacity expansion into consideration to meet accuracy criteria. Improving load forecasting technology enables planned power management, which enables rational grid operation and unit maintenance planning, as well as the formulation of suitable power supply construction plans and facilitation of power improvement. Now, there are numerous approaches for electricity load forecasting. We present an electrical load forecasting model in this article that uses long short-term memory (LSTM) for case studies of electrical data in Central Java and DIY. The proposed implementation of LSTM is extremely well matched to the time series dataset, which can improve the training process’s accuracy of convergence. We experiment with different time steps to accelerate the training process’s convergence. It is discovered that by forecasting the different timesteps, all scenarios attain significantly varying forecasting accuracy. We present several variations on epochs and the number of hidden layers to identify the best model for the basic RNN and LSTM models.