Authors
Kiguchi Ya , Heo Yb, Weeks Mc, Choudhary Rad
- a Department of Engineering, University of Cambridge, UK
- b School of Civil, Environmental, and Architectural Engineering, Korea University, South Korea
- c Faculty of Economics and Clare College, University of Cambridge, UK
- d Data-centric Engineering, Alan Turing Institute, UK
Publication Details
Energy, Volume 173, 15 April 2019, Pages 959-970
Received 17 February 2018, Revised 9 December 2018, Accepted 9 January 2019, Available online 18 January 2019.
Highlights
- • A model to forecast user adaptation under different Time-of-use tariffs.
- • Lifestyle constraints are considered as key inputs in the form of statistical moments.
- • The model requires only a half-hourly sampled historical smart meter data.
- • Random Forest outperforms Neural Network and Liner Regression models.
- • MAPE of the best model reports 2.05% for the weekday.
Abstract
The installation of smart meters enabling electricity load to be measured with half-hourly granularity provides an innovative demand-side management opportunity that is likely to be advantageous for both utility companies and customers. Time-of-use tariffs are widely considered to be the most promising solution for optimising energy consumption in the residential sector. Although there exists a large body of research on demand response in electricity pricing, a practical framework to forecast user adaptation under different Time-of-use tariffs has not been fully developed. The novelty of this work is to provide the first top-down statistical modelling of residential customer demand response following the adoption of a Time-of-use tariff and report the model’s accuracy and the feature importance. The importance of statistical moments to capture various lifestyle constraints based on smart meter data, which enables this model to be agnostic about household characteristics, is discussed. 646 households in Ireland during pre/post-intervention of Time-of-use tariff is used for validation. The value of Mean Absolute Percentage Error in forecasting average load for a group of households with the Random Forest method investigated is 2.05% for the weekday and 1.48% for the weekday peak time.