Non-parametric probability density forecast of an hourly peak load during a month

Abstract

The Load Serving Entity (LSE) requires, for its power procurement portfolio management, accurate peak load forecast in medium term (upto six months ahead). A complete description of the random variable, i.e., load, is provided by probability density function. Hence, we consider the problem of forecasting probability density function of hourly peak load during a month. First, we propose a non-parametric model based on the Alternating Conditional Expectation (ACE) to obtain point forecast. Then, by considering multiple scenarios of the weather variables i.e., temperature-humidity tuples, we obtain probability density forecast of the peak load. Out-of-sample testing is used to demonstrate efficacy of the proposed approach.