2020/11/10 by Dilusha Weeraddana, Weeraddana, Dilusha, Nguyen Lu Dang Khoa +7
Computer Science · Energy · Environmental Science · #Air Quality Monitoring and Forecasting #Artificial Intelligence (cs.AI) #Energy, Environment, and Transportation Policies #FOS: Computer and information sciences #Gaussian Processes and Bayesian Inference #Machine Learning (cs.LG)
paper · pdf · doi:10.48550/arxiv.2011.05519
openalex publication_date 2020/11/10 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
In this paper, the process of forecasting household energy consumption is\nstudied within the framework of the nonparametric Gaussian Process (GP), using\nmultiple short time series data. As we begin to use smart meter data to paint a\nclearer picture of residential electricity use, it becomes increasingly\napparent that we must also construct a detailed picture and understanding of\nconsumer's complex relationship with gas consumption. Both electricity and gas\nconsumption patterns are highly dependent on various factors, and the intricate\ninterplay of these factors is sophisticated. Moreover, since typical gas\nconsumption data is low granularity with very few time points, naive\napplication of conventional time-series forecasting techniques can lead to\nsevere over-fitting. Given these considerations, we construct a stacked GP\nmethod where the predictive posteriors of each GP applied to each task are used\nin the prior and likelihood of the next level GP. We apply our model to a\nreal-world dataset to forecast energy consumption in Australian households\nacross several states. We compare intuitively appealing results against other\ncommonly used machine learning techniques. Overall, the results indicate that\nthe proposed stacked GP model outperforms other forecasting techniques that we\ntested, especially when we have a multiple short time-series instances.\n