Application Of CUDA In The Very Short Term Load Forecasting Of Multi-node

Load forecasting is to estimate and predict on the power load, which is the foundation of accomplishment to power system security and economic scheduling. As one kind of the forecasting, the very short term load forecasting is based on power load variation in certain seconds or minutes in one hour, taking an more and more important role in power system operation and control, such as dynamic economic coordination, online security monitor, real-time power price achievement and automatic generating control. Power load forecasting on several important stations is conducive to understanding the power load distribution and trend of critical stations in power system, and maintaining the stable operation of power system to have high-quality electricity with a view to health operation among power market.Recently, with the more and more mature technique on Graphics Processing Unit (GPU), its strong floating-point calculation and parallel operation leads to the potentials cultivation in this field and makes it popular among non-graphic field, for example, mode recognition, earthquake forecasting and synthesizing radar. The research on General Purpose CPU (GPGPU) based on graphics processing chip is going deeper and deeper. With the launch of GeForce 8800 by NVIDIA, its computer Unified Device Architecture (CUDA) spread widely and now there are considerable thesis and research reports publicized.Due to the real-time and complicacy of very short-term power load forecasting, in the very short term load forecasting of multi-node, the normal CPU can't afford thus GPGPU is necessary to bring in.The test procedure of this thesis is based on the descendible calculation of neural network to the very short term load forecasting of multi-node. Moreover, in order to keep the real-time of multi-node load forecasting, the CUDA will be implemented in GTX460 by NVIDIA to forecast the very short term power load based on neural network calculation among 33 towns in Dongguan power system. The test proves that, by implementing CUDA, the result of very short term load forecasting is good with real-time feature.