A deep learning method for household characteristic classification from smart meter data

Inhalt:
Redicting household socio-economic characteristics such as household income and cooking style from daily electricity consumption can be very useful for energy providers to offer personalized services and shift to more efficient energy management by rolling out effective energy-saving programs. In this paper, we proposed a deep learning method (CNN-LSTM) to automatically predict household characteristics from their energy consumption. The proposed CNN-LSTM model utilizes the strength of the CNN and LSTM models to learn temporal and spatial features from the time series of electricity consumption data. Our method is evaluated using the Irish Commission for Energy Regulation (CER) electricity customer behaviour trial data. The experiment demonstrates the effectiveness of deep learning-based methods, which show higher accuracy and F1-score over traditional methods such as manual feature engineering techniques.