A new intelligent approach is developed for short-term loadforecasting (STLF). The technique consists of three basic modules. Thefirst module employs the clustering of daily load curves using amodified Kohonen algorithm (MKA). The second module determines the mostappropriate supervised neural network topology and associated initialweight values for each cluster extracted from a historical database, byusing a genetic algorithm (GA). In the third module, a geneticallyoptimized three-layered backpropagation (BP) network is trained and runto perform hourly load forecasting. The effects of each module on theforecasting accuracy are considered separately. The proposed system istested extensively with the load curves of the Turkish electrical powersystem in 1993 using different day types from different times of theyear, and promising results are obtained with approximately 1% meanerror for days distributed throughout the year
展开▼
