Evolving Smart Meter Data Driven Model for Short-Term Forecasting of Electric loads

Abstract

Short-term forecasting of electric loads is an essential function required by Smart Grids. Today increasing amount of smart metering data is available enabling the development of more accurate and adaptive data-driven models for short-term load forecasting. Until now, a plethora of models have been developed ranging from simple statistical regression models to more advanced models such as artificial neural networks (ANNs) and support vector machines (SVMs). Despite the relatively high accuracy obtained, data-driven models are still perceived to be highly complex and nontransparent, thus not allowing engineers and system operators to interpret and understand properly their behavior. Therefore it is important to develop optimization schemes, which can be used to facilitate the selection of appropriate data-driven model structure, and thus improve the acceptance of data-driven models in the domain. This study presents an optimization scheme based on multi-objective genetic algorithm (GA) for designing simple but accurate data-driven models for short-term forecasting of electric loads using smart metering data. The optimization scheme is demonstrated for an ANN model, and the performance of the resulting ANN model is assessed in terms of several performance indices.