Short-Term Load Demand Forecasting Based on Weather and Influencing Factors Using Deep Neural Network Experts for Sustainable Development Goal 7

Due to the increasing essentiality of electricity in this modern world, its sustainable management is crucial for the economic development of a nation, and is allied with the Sustainable Development Goal 7, which focuses on ensuring access to affordable and reliable energy for all. Accurate day-ahead instantaneous load forecasting, also known as Short-Term Load Forecasting, plays a pivotal role in achieving higher efficiency and sustainability in power system planning and operation. Due to the volatility and randomness of electricity demand as well as high dependence on exogenous parameters such as weather, effectively capturing the dynamic change of load curves is extremely challenging. Including knowledge of influencing parameters can significantly enhance the forecasting accuracy, reduce service interruptions, and ultimately lead to cost reduction and reliability. Based on a systematic literature review, it is characterized that an optimal forecasting system learns patterns from historic load consumption during similar days along with consideration of various influencing parameters. Hence, we propose a novel load forecasting system MoDeNNFE that employs a Mixture of Experts to produce load forecast output. To optimize the forecasting accuracy, the system extracts different input features from historical data based on types of day, and each expert member undergoes Deep Learning to make specialized predictions accordingly. We perform experimental research for Indian metropolitan cities using data of its prominent power utility Tata Power and India Meteorological Department. The results and comparisons indicate that MoDeNNFE achieves superior predictive performance over other standard approaches and demonstrates a promising forecasting efficacy.