electricityPricePredictionProject
Main menu:
An Ensemble Model for Day-ahead Electricity Demand Time Series Forecasting
Wen Shen, Vahan Babushkin, Zeyar Aung, Wei Lee Woon
In the electrical power industry an accurate prediction of electricity demand or price is essential both for producers and suppliers. We consider the problem of day-ahead electricity demand prediction via the prism of Pattern Sequence-based Forecasting (PSF) algorithm, which predicts the future evolution of a time series based on the similarity of patterns. The proposed Pattern Forecasting Ensemble Model (PFEM) incorporates an ensemble of five PSF-style forecasting algorithms, namely the K-means (PSF itself), Self Organizing Map, Hierarchical Clustering, K-medoids and Fuzzy C-means. Iterating on weighted predicted values of these models allows to obtain superior results compared with all the other five individual components in terms of both MER and MAE.
We evaluated the performance of PFEM on three publicly available electricity demand datasets from the New York Independent System Operator (NYISO), the Australia’s National Electricity Market (ANEM) and the Ontario’s Independent Electricity System Operator (IESO). It was observed that the PFEM was able to provide more accurate and reliable forecasts than the five individual models, including PSF. To read more, please follow the link.
Best prediction of PFEM for NYISO dataset (2009).
Worst prediction of PFEM for NYISO dataset (2009).
Best prediction of PFEM for ANEM dataset (2009).
Worst prediction of PFEM for ANEM dataset (2009).
Best prediction of PFEM for IESO dataset (2009).
Worst prediction of PFEM for IESO dataset (2009)