Deep Learning-Based Load Forecasting in Smart Grid

The major concern for energy management in smart grids is the uncertainty in the power supply and demand on the grid. To address the challenge of the unpredictability of demand and supply, there is a need to accurately predict power supply and demand, which helps in the quality of power, saving energy, integrating renewable source, and reducing operational costs. This paper is mainly focused on Short-Term Load Forecasting (STLF), which is the prediction of electricity load over a relatively short period. Two deep neural network architectures are proposed, one that combines the features extracted from the 1D-CNN model and Bi-LSTM models and the other that combines the features extracted from the 1DCNN and GRU models. This ensemble learning approach helps increase forecasting accuracy. On comparing the results of their performance, the combination of CNN and GRU performs better in forecasting the load with a lower Mean Absolute Error.