Predictive Modelling for Multi-Location Deep Learning based Load Forecasting: An Integrated Approach

Accurate prediction of short-term future energy demand is crucial to modern energy grids for uninterrupted supply of electricity to the consumers. The advancement in accuracy and robustness of state-of-the-art Machine Learning (ML) and Deep Learning (DL) algorithms, as well as computational power of computers have created the perfect platform for ML to perform predictions to a great extent of accuracy. Bangladesh, being a lower-middle income and still aspiring to be a developing nation, has to use its resources to the utmost effectiveness, of which, power load prediction is an integral part. Using DL algorithms like- DNN, LSTM and GRU, prediction models were trained and tested using historical daily load data of all the 9 major metropolitan areas of Bangladesh. Load data from January 2014 to April 2023 was split into train and test sets after cleaning and normalization, which was then validated for each metropolitan area. The load prediction model using DNN performed better compared to the other tested models, as well as the other competing models, with the highest performing metropolitan area having $\mathrm{R}^{2}$ of 0.991 in Sylhet and MAPE of 0.011 in Mymensingh and Rajshahi. While the LSTM and GRU models’ R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> values, at 0.9827 and 0.9753, respectively, for Sylhet, are excellent by themselves, the exceptionally accurate performance of the DNN model, puts behind the state-of-theart techniques.