Regression-Based Wintertime Energy Consumption Prediction for Cold Load Pick-Up Management

Abstract

Thermostatically Controlled Loads (TCL) have a significant impact on Cold Load Pick-up (CLPU) during distribution system service restoration. Widespread deployment of smart meter devices opens up new opportunities for data-driven load modelling. In this paper, we compare several linear regression approaches to robust short term prediction of hourly energy consumption as a function of the outdoor temperature during the low-temperature season. The goal is to estimate the energy that will not be delivered during an outage for the purpose of a further estimation of the CLPU peak and duration for consumers with TCLs. The prediction is based on smart meter load data and outdoor temperature data. The performance of the proposed regression approaches is analyzed for 25 residential homes from real measured data. Prediction is performed on an hourly basis. The quality of the regression results is compared with the Naive forecast benchmark method. The results show that autoregression approach outperforms the other methods, however, since this approach is highly depended on the existence of the sequence of the previous load measurements, as an alternative approach, ENS prediction is successfully performed using only temperature data.