A Comparison of applying Multiple Methodologies for short-term Load Forecasting

EasyChair Preprint 135

Abstract

We present five methodologies for probabilistic load forecasting which are a method based on Bayesian estimation, a rank-reduction operation based on principle component analysis, least absolute shrinkage and selection operator (Lasso) estimation, ridge regression, and a supervised learning algorithm called scaled conjugate gradient (SCG) neural network. These five models considered can be regarded as a variety of competitive approaches for analyzing hourly electric load and temperature. The modeling approaches incorporates the load and temperature effects directly, and reflect hourly patterns of the load. We provide empirical studies based on the Global Energy Forecasting Competition 2014 (GEFCom 2014). In this research, we use historical load data only to forecast the future load. The study performs the estimation comparison of the five methodologies, showing that ridge regression has a marginal advantage over the others.

Keyphrases: Mean Squared Error, Methodologies, load forecasting, relative error percentage

@booklet{EasyChair:135,
  author    = {Mingsui Sun and Mahsa Ghorbani and Edwin Chong and Siddharth Suryanarayanan},
  title     = {A Comparison of applying Multiple Methodologies for short-term Load Forecasting},
  doi       = {10.29007/7ksb},
  howpublished = {EasyChair Preprint 135},
  year      = {EasyChair, 2018}}