Neural network based approaches to very short term load prediction

In the environment of ongoing deregulated power industry, traditional automatic generation control (AGC) has become a set of ancillary services traded in separate markets other than the energy market. The performance of AGC is mandated to meet the NERC control performance standards (CPS). The CPS criteria are intended for the over-compliant power utilities to loosen control of their generating units. On the other hand, excess regulating capabilities can be sold for better financial well being. Generation companies have to optimize the portfolio of their generating assets to achieve maximum profitability. The optimization process involves economic allocation of generation over a period of time, which requires the load profile to be predicted for the dispatch period of minute level. This paper addresses the importance of very short term load prediction (VSTLP) in this context, and proposes three new schemes to make load predictions. The VSTLP is developed and implemented as part of Siemenspsila CPS based automatic generation control (AGC) product. Procedures involved in the proposed approaches are presented. Comparison between different approaches is made. Experimental studies are presented to demonstrate the effectiveness of the proposed approaches.