A data-driven framework for modelling community energy demand

To understand the dynamics of community-level energy demand, this paper aims to present a novel framework of data-driven approaches, developed using the Big Electricity Demand Dataset (BEDD). The BEDD is collected by the Community Energy Demand Reduction in India (CEDRI) project for the case study community “Auroville” located in India. The BEDD used in this study comprises of high-resolution electricity demand dataset, recorded at a temporal resolution of 1-minute over the period of November 2020 - September 2021 across 664 buildings.The modelling framework integrates a suite of advanced data analytics (ADA) approaches in an organised schematic that facilitates an opportunity for extracting underlying statistical dynamics and patterns of electricity profiles at individual building scales using an STL-based demand decomposition algorithm [1]. The key statistical features obtained are used to drive widely applied unsupervised machine learning techniques such as K-means and Hierarchical clustering for grouping electricity demand patterns in BEDD. A preliminary framework (developed by authors, referred to as STL_HMM_GP [2]) in association with a novel ‘climate module’ is applied to generate an ensemble of community-level energy demand profiles. The modeling framework is applied to access the impact of climate change on community-level energy demand.The key scientific innovation is the integration of advanced data-science/statistical approaches in a single framework that facilitate a systematic pre-processing, modelling, and post-processing of big energy demand dataset for the case-study community. These include 1) the application of STL-based demand decomposition approach for extracting underlying dynamics/patterns for individual buildings, ii) a k-mean based clustering approach for characterising individual electricity demand patterns and sampling of profiles; iii) stochastic modelling of individual demand profiles using a designed Hidden Markov Model (HMM) based algorithm; iv) application of extreme value distribution for an efficient simulation of peak demands and v) a percentile-based bias correction. This paper demonstrates the integration of STL_HMM_GP with a novel ‘climate module’ (underpinned by partial-least square regression) for associating climatic variables directly with demand profiles for a comprehensive climate impact assessment.The outcomes of the paper are intended to serve a range of applications, including, i) missing value imputation; ii) synthetic demand simulation; iii) long-term forecasting; iv) ensemble-based uncertainty analysis; v) climate change impact assessment of community energy demand. In this context, the proposed model should be relevant to wider building and energy-related topics, e.g. built-environment, community-level energy demand, big data, and advanced data-science approaches.The paper discusses potential applications of data-driven approaches for designing an efficient modelling framework that also includes a climate module for the simulation of climatic impact on community energy demand. The underpinning STL_HMM_GP modelling system has been previously shown to generate community-level energy demand profiles using a small sample of individual demand profiles. The present work presents a comprehensive application of the modelling framework which is also coupled with a climate module to simulate the dynamics of community-level energy demand patterns for future climate change. The outcomes of the STL_HMM_GP system, are highly dependent on the selection of individual demand profiles in the sample. The outcomes of this paper have been discussed to generate a range of potential scenarios for projecting future community-level energy demand, e.g. impact of selecting the low-medium-high volume of building with a varying ratio of buildings with an air-conditioning system in the sample to assess the penetration of AC in Indian communities. Further, the impact will be assessed for three different periods Winter, Spring, and Summer, and should be of high interest to future policymakers and planners.