Uncertainty Estimation in Power Consumption of a Smart Home Using Bayesian LSTM Networks

Smart cities and smart houses have recently seen a rise in power and energy consumption. Household energy demand is likewise ameliorating with population growth and enhanced living standards. For this reason, it is vital to predict power consumption demand for a smart home. Predicting the power consumption is a multivariate time series issue with multiple factors impacting power usage. The variables fluctuate depending on how the user uses them, impacting the amount of utilized power consumption. Predicting power consumption and estimating the uncertainty of these values provide a good idea to the residents and planners to prepare for future capacity or load forecasting demand. Time series forecasting relies heavily on accurate uncertainty estimations. Classical time series models are often combined with probabilistic formulations to estimate uncertainty. Adding exogenous variables and fine-tuning such models are complex tasks. As a response to the current resurrection of Long Short Term Memory networks, we provide a Bayesian deep neural network model that enables time series prediction and uncertainty estimation concurrently. We consider a real-world dataset (individual household electric power consumption data set) to test the effectiveness of the proposed approach.