Robust Probability Density Forecasts of Yearly Peak Load using Non-Parametric Model

We propose an approach for robust probability density forecast of yearly peak load. The probability density forecast is robust against influential observations and error in econometric projections. By using a method akin to jackknifing, we obtain multiple instances of the yearly peak load per scenario of explanatory variables. The density forecast of the YPL is obtained using kernel density estimation. There can be many parametric models for forecasting trend. We propose the use of alternating condition expectation (ACE) to discover trend without making any assumption on its functional form. We compare the ACE model and parametric trend models e.g., linear and exponential with the explanatory variables factored in them. Proposed approach is illustrated with real life data of an electricity distribution company.