Power Load Daily-Similarity and Time-Series Prediction Using the Selective Differential Polynomial Network

Abstract

Cooperation on the electricity grid requires from all providers to foresee the demands within a sufficient accuracy. Short-term electric energy estimations of a future demand are needful for the planning of generating electricity in regional grids and operating power systems. An over-estimating of a future load results in an unused spinning reserve. Under-estimating a future load is equally detrimental because buying at the last minute from other suppliers is obviously too expensive. Differential polynomial neural network is a new neural network type, which forms and solves a selective general partial differential equation of an approximation of a searched function, described by observations, with combination sum series of convergent relative polynomial derivative terms. An ordinary differential equation, which can model 1-variable function real time-series, is analogously substituted with partial derivatives of selected time-point variables. A new method of the short-term power demand forecasting, based on similarity relations of subsequent day progress cycles at the same time-series points is presented and tested on 2 datasets. Comparisons were done with the artificial neural network using the same prediction method.