System of statistical approaches for community energy demand modelling

Modelling transient community-level peak energy demand event is often challenging, as it requires the acquisition and systematic analysis/modelling of electricity demand data across a large number of buildings. Electricity demand data with diverse demand characteristic can be analysed/modelled/aggregated (in time) to understand the impact of various micro-level activities (specifically, peak demand household-level activities occurring simultaneously across multiple dwelling at a specific time) on the community-level demand curve. However, in real-life applications, the availability of good-quality electricity demand data across a large number of multiple dwellings within a community is often challenging. This paper is aimed to investigate the potentials of k-means clustering approach for developing a systematic sampling, weighting and demand aggregation strategy for projecting community-level demands with high precision, just by using a small sample of buildings and easily accessible contextual information (e.g. average monthly demand or various activity periods during a day). These selected samples of dwellings are processed with a novel system of demand synthesis model developed by authors, referred to as HMM_GP. Five different variants of k-means clustering are developed using statistical mean, median and proportion of demand during four different periods of days. Corresponding to each variant five aggregation schemes are constructed. The HMM_GP model is underpinned by a hidden Markov model (HMM) for simulating synthetic demand and a Generalised Pareto (GP) distribution to effectively model dynamics of peak demand events. Aggregation schematics are demonstrated for 30-minutely demand dataset collected over four weeks in July 2017 for 74 dwellings for a case-study community of Fintry (Scotland).