A Meta-Modeling Power Consumption Forecasting Approach Combining Client Similarity and Causality

Power forecasting models offer valuable insights on the electricity consumption patterns of clients, enabling development advanced strategies and applications aimed at energy saving, increased efficiency, smart pricing. The data collection process for client is not always ideal resulting datasets often lead to compromises in implementation models, as well suboptimal performance, due several challenges. Therefore, combinations elements that highlight relationships between clients need be investigated order achieve more accurate predictions. In this study, we exploited combined effects similarity causality, developed a power model utilizes ensembles long short-term memory (LSTM) networks. Our novel approach enables derivation different representations predicted based feature sets influenced by causality metrics. were used train meta-model, multi-layer perceptron (MLP), combine results LSTM optimally. This combinatorial achieved better overall performance yielded lower mean absolute percentage error when compared standalone do include causality. Additional experiments indicated combination resulted performant implementations utilizing only one element same structure.