Addressing missing smart meter data in electricity consumption using machine learning

This study investigates the potential of Long Short Term Memory neural networks to estimate smart meter electricity daily consumption data for a given household. Long Short Term Memory, LSTM for short, is a machine learning solution particularly well suited for building predictive models in a time-series context, due to its architecture and long-term memory capability. In the first step, the smart-meter electricity data undergoes a comprehensive data analysis, aimed at effectively preparing the data for training. In the second step, a feature engineering approach is applied to add relevant information such as the weekday, temperature or the occupancy of the household to the dataset. A univariate LSTM model, which utilizes past energy consumption data, alongside a multivariate LSTM model, that incorporates the engineered features is designed and trained. Finally, the results of the univariate LSTM model, the multivariate LSTM model, and a baseline method are evaluated and compared.