Market clearing price forecasting based on dynamic fuzzy system

Abstract

The problem of forecasting power market clearing price is addressed based on dynamic fuzzy system (DFS). Dynamic fuzzy system is a modified model of fuzzy system. It separates the rule reasoning layer into two layers: antecedent layer and consequence layer. DFS can learn rules by itself from sample data rapidly. As all sample data were input into DFS completely, the rule set of DFS are established automatically. Therefore, DFS performs well in dynamic condition. In addition, hybrid-and-length-varied-encoding genetic algorithms (GAs) is used to train the DFS. New crossover operator and mutation operator for this scheme are presented. The whole model, not only the membership function parameters, but also the fuzzy quantities numbers of input and output variables are adjusted automatically based on sample data. The problem of forecasting market clearing price (MCP) in power market is very important and very complicated. In this paper, DFS is used to solve this problem. With the consideration of California Power Market MCP data, the test results shows that DFS performs very well in this varied condition.