LSTM Network for Predicting Medium to Long Term Electricity Usage in Residential Buildings (Rikkos Jos-City, Nigeria)

摘要

Recently, electricity consumption has significantly increased due to increasing urbanization. It is estimated that buildings consume about 40% of the electricity supply. To minimize the gap between supply and demand, there is need for efficient load forecasting thus, new paradigms have to be implored using automated methods that can dynamically forecast the buildings’ energy consumption. A number of technique and computational approaches have been used recently in order to improve the prediction accuracy. These techniques as presented in the literature would likely have decreased accuracy in application if the weather in the future was significantly different from the weather that was concurrent with the training data. Thus, this research implements an improved model for medium to long term forecasting of electricity usage in residential buildings by using NARX with LSTM while also including weather data as a factor due to its influence and uncertainties to the prediction. The dataset was collected using smart software tools on smart meters and aggregated over 34 buildings in Rikkos Jos- City, Nigeria. The NARX network was built on the top layer of the LSTM neural network to inputs electricity and weather data concurrently to the network. The network was trained using Bayesian regularization back propagation algorithm. The proposed model was evaluated against state-of-the-art prediction techniques use in energy forecasting using RMSE, MAPE and R on Matlab 2018a. The experimental result shows that the proposed model not only curtail future impact of weather uncertainty but also outperforms the existing models in terms of accuracy and model fitting achieving a lower value in terms of RMSE and MAPE.