Neural Network Based Very Short Term Load Prediction

摘要

This paper presents novel neural network based very short term load prediction (VSTLP) schemes. The VSTLP has developed and implemented as part of Siemens' CPS based Automatic Generation Control (AGC) Scheme. The load prediction is formulated mathematically to form a basis for the neural network based VSTLP. The neural network based VSTLP is different from conventional neural network based Short Term Load Forecast (STLF) in that: (1) VSTLP provides predictions of minutely load for the very near future while STLF forecasts load with a much longer lead time of one hour up to seven days; and (2) The minutely forecasted load values by VSTLP are intended for use in dispatching generation in a predictive manner in real time. The neural network based VSTLP takes into consideration the load dynamics in the immediate past, the variations in load dynamics during the course of a day, and the weather factors as well. Mathematical formulation of the problem and the architecture of the neural network based load prediction schemes are studied. Experimental experiences in this study are also discussed.