Energy Consumption Forecasting Using a Stacked Nonparametric Bayesian Approach

In this paper, the process of forecasting household energy consumption is studied within framework nonparametric Gaussian Process (GP), using multiple short time series data. As we begin to use smart meter data paint a clearer picture residential electricity use, it becomes increasingly apparent that must also construct detailed and understanding consumer’s complex relationship with gas consumption. Both patterns are highly dependent on various factors, intricate interplay these factors sophisticated. Moreover, since typical low granularity very few points, naive application conventional time-series techniques can lead severe over-fitting. Given considerations, stacked GP method where predictive posteriors each applied task used in prior likelihood next level GP. We apply our model real-world dataset forecast Australian households across several states. compare intuitively appealing results against other commonly machine learning techniques. Overall, indicate proposed outperforms tested, especially when have instances.