COINCIDENT PEAK PREDICTION USING A FEED-FORWARD NEURAL NETWORK

Abstract

A significant portion of a business’ annual electrical payments can be made up of coincident peak charges: a transmission surcharge for power consumed when the entire system is at peak demand. This charge occurs only a few times annually, but with per-MW prices orders of magnitudes higher than non-peak times. A business is incentivized to reduce its power consumption, but accurately predicting the timing of peak demand charges is nontrivial. In this paper we present a decision framework based on predicting the day-ahead likelihood of peak demand charges. We train a feed-forward neural net-work to estimate the probability of system demand peaks and show it outperforms conventional forecasting methods using historical load. Using ERCOT demand and weather data from 2010-2017, we show the effectiveness of our framework.