A Hybrid Deep Learning Model for Day Ahead Short-Term Load Forecasting

Electricity load forecasting is crucial for the efficient and economical operation of power systems. It involves predicting the future demand for electricity to optimize grid operations, manage energy resources effectively and ensure stability across the network. Accurate forecasting helps in mitigating risks associated with energy supply and demand imbalances and facilitates the transition towards smarter, more sustainable energy grids. Addressing the challenges of accurate forecasting amidst non-linear, non-stationary time-series data associated with electricity usage, this paper employed a hybrid deep learning model combining convolutional neural networks (CNN) and bidirectional long short-term memory (Bi-LSTM) networks. The objective of this study was to train the model to predict the aggregated active power demand on an hourly basis over a 24 hour period, using data from the preceding 24 hours. The model's performance was validated using historical power load data collected from the Power Grid Company of Bangladesh Limited. We compared four deep learning models to assess their forecasting efficiency. Our analysis revealed that the CNN Bi-LSTM algorithm was the most effective, achieving the lowest RMSE of 0.71493 and thus the least prediction error among the tested methods.