Title

Effect of forecast accuracy on day ahead prediction of coincident peak days

Document Type

Article

Publication Date

10-24-2019

Department

Department of Computer Science

Abstract

Many Independent System Operators (ISOs) have programs with surcharges on days with highest system-wide demand, a coincident peak. Advance notice of these coincident peak days (CPs) may allow consumers to lower their demand and/or add generation to avoid the surcharge. In this paper, we focus on the prediction of five CP days per fiscal year (5CP) in an ISO. An all convolutional neural network classification model is used to predict 24 hours ahead whether a day is a 5CP. We consider four cases: I - knowing tomorrow's power demand exactly (an oracle), II & III - knowing some information about tomorrow (an oracle + increasing noise), and IV - no knowledge of future. Over 21 years of data, the prediction on oracle data has a mean precision of 0.509 and mean recall of 0.886. As expected, increasing the amount of noise or having no future knowledge (case IV) reduces the performance of the models.

Publication Title

2019 IEEE Innovative Smart Grid Technologies - Asia (ISGT Asia)

Share

COinS