Data-driven short term load forecasting with deep neural networks: Unlocking insights for sustainable energy management

In today’s smart grid and building infrastructure, it is strongly suggested to implement short-term demand forecasting for future power generation. There is a growing demand for improved accuracy in forecasting, from the level of individual users to that of the power system in the context of the emerging energy market and the creation of a smart grid infrastructure. There is also a need for better rules to control the supply and demand equilibrium. This study demonstrates and incorporates a deep learning neural network model with time series analysis and feature selection in order to forecast the complex and variable hourly load demand and involved a comprehensive comparison between DNN approach and other alternatives for short-term load forecasting. The deep feed-forward and wide and deep neural network models incorporate with stacking hidden layers of CNN such as LSTM/GRU units for handling uncertainty. While, random forest regression with GridSearchCV has been used for the identification of optimal set of feature selection. This proposed model has increased temporal dependencies, a large influence, and a strong capacity for learning in order to accurately predict loads based on the hourly driven power consumption of IEEE dataset. Examining the resulting limitations and influencing factors allows for the validation of the experimental findings.