Extracting controllable heating loads from aggregated smart meter data using clustering and predictive modelling

Modelling of controllable loads is a necessary function required by demand side management, and specifically load control of smart grids. A large amount of smart metering data and other supporting data are available, enabling the development of new, intelligent data-driven fashions for recognising and modelling loads. However, it is a challenge to extract useful information from this massive, often aggregated data in a reliable and understandable fashion. In this paper we present a data-driven approach, for recognising and modelling of controllable heating loads of small customers. Main computational methods used include self-organizing map (SOM), k-means algorithm and support vector regression (SVR). The approach consists of two major stages, namely (i) the recognition of customers that have electrical heating using clustering based on extracted behavioural features and (ii) the predictive regression modelling of controllable heating loads in recognised customer segment. One year of hourly metered electricity consumption data from 525 customers having heterogeneous heating systems, combined with available hourly measured outdoor temperatures and site-specific building information, were used as the base data in the model development and validation.