Short-term system marginal price forecasting with hybrid module

In this paper, dynamic clustering and a BP neural network were combined to forecast short-term system marginal price (SMP). With the criterion of minimal distance between the sample data and the clustering center, sample data were divided into several classes with the dynamic clustering method. Then BP neural network modules with the same topology structure and different weights and thresholds were built for every class. Weights and thresholds of different layers' neurons in the BP neural network were revised in the backpropagation process. This set of forecasting modules were trained and tested on historical marginal price data from the American PJM power system. This kind of module can correctly predict short-term system marginal price.