Hybrid ARIMA-LSTM Forecasting of Residential and Regional Energy Consumption

The effective energy management in both residential and regional contexts has become increasingly important for reducing the costs of energy generation, transmission, and consumption to sustainable levels. Accurate energy consumption prediction is therefore key to achieving optimal energy and capital management. The current literature on energy consumption patterns indicates that no single model is able to adequately encode the intricate time series complexities inherent in energy consumption. Here, we demonstrate that a model combining ARIMA and Long Short-Term Memory (LSTM) enables the highly-accurate forecasting of energy consumption in scenarios consisting of residential and regional settings. Our computational experiments, which involve training these hybrid models on shortterm time series data influenced by weather and family routines in residential environments, as well as on long-term energy consumption data from large regional areas, show that the ARIMA-LSTM hybrid model achieves improved predictive performance. Our findings have the potential to advance the development of hybrid forecasting models for noisy energy consumption time series, contributing to optimal data-driven energy management, enhanced grid efficiency, and environmentally conscious energy prediction strategies.