Time-Series based Household Electricity Consumption Forecasting

Data analytics using machine learning technologies when applied to the energy consumption data can provide valuable inputs for maintaining the perfect supply demand balance in a smart electrical grid system. In particular, the accurate predictions of energy consumption for future periods of time aids significantly in cost-cutting and energy saving for utility companies. Making use of the popular method of time-series forecasting and the Artificial Neural Networks (ANN) models, here in this paper, one of the variants of the Recurrent Neural Networks (RNN) model, the Long Short Term Memory (LSTM) model is applied for household electricity consumption forecasting. Real datasets from consumption building are used for experimenting the model and applied through Tensorflow platform with the keras functions in Python. The results obtained show significantly accurate values in predicting future consumption derived from models training with actual values of current consumption. Hence, this work provides yet another proof that the LSTM machine learning forecasting methods can be efficiently applied for household electricity forecasting.