Hours ahead automed long short-term memory (LSTM) electricity load forecasting at substation level: Newcastle substation

Abstract

Nowadays, electrical energy is of vital importance in our lives, every country needs this resource to develop its economy, factories, businesses, and homes are the basis of the economic structure of a country. In the city of Newcastle as in other cities are in constant development growing day by day in terms of industries, homes and businesses, these elements are the ones that consume all the electricity produced in Newcastle. Although Australia has strategically located substations that serve the function of supplying all existing loads with quality power, from time to time the load will exceed the capacity of these substations and will not be able to supply the loads that will arise in the future as the city grows. To find a solution to this problem, we use a deep learning model to improve accuracy. In this paper, a Long Short-Term Memory recurrent neural network (LSTM) is tested on a publicly available 30-minute dataset containing measured real power data for individual zone substations in the Ausgrid supply area data. The performance of the model is comprehensively compared with 4 different configurations of the LSTM. The proposed LSTM approach with 2 hidden layers and 50 neurons outperforms the other configurations with a mean absolute error (MAE) of 0.0050 in the short-term load forecasting task for substations.