AN ARTIFICIAL NEURAL NETWORK METHOD TO FORECAST ELECTRIC LOAD

Autor/autori: Acad. Paul Dan CRISTEA, PhD, Prof. Rodica TUDUCE, PhD, Assoc. Prof. Dumitru Iulian NASTAC, PhD

Rezumat: Lucrarea prezinta un sistem care utilizeaza o retea neurala (RN) pentru invatarea eficienta a unei secvente nestationare. Sistemul are potentialul de a invata, clasifica si prezice orice semnale stationare sau nestationare, in particular curbele de variatie in timp a sarcinii din sistemele electro-energetice, la diferite scale de timp. O mare varietate de aplicatii de clasificare si predictie a secventelor nestationare pot fi rezolvate intr-un mod eficient utilizand aceasta tehnica de reantrenare adaptiva combinata cu arhitectura in doua trepte a sistemului.

Cuvinte cheie: predictia curbelor de sarcina, retele neurale artificiale, analiza componentelor principale, reantrenare

Abstract: The paper presents a system using an artificial neural network (ANN) to efficiently learn the shape of a non stationary sequence. The system has the potential to learn, classify and predict any stationary or non stationary signals, in particular the chronological load curves of power systems, at various time scales. A large variety of classification and forecasting applications to non stationary sequences can be efficiently solved by using this adaptive retraining technique combined with the two step system architecture.

Keywords: load forecasting, Artificial Neural Networks, Principal Component Analysis, retraining technique

DOWNLOAD PDF