A MARKET CLEARING PRICE PREDICTOR BASED ON MODULAR NETWORKS
Daozhuo Jiang · Proceedings of the CSEE · 2002
An innovative method to forecast MCP based on a modular network is presented. The architecture consists of one integrating unit and several local experts. The integrating unit based on FCM algorithm soft-partitions the input space into several regions while the local experts specialize to learn only the regions for which they are responsible. The output is given by a competition mixture mechanism. The predictor reduces the learning workload, and promotes performance. Finally, a simulation on real-world data acquired from California PX web site is given. This model gives out an effective bidding strategy, and may be a useful tool for monitoring market power.