Very short-term solar forecasting using multi-agent system based on Extreme Learning Machines and data clustering

This paper proposes a new multi-agent system to solve very short-term solar forecasting problems. The system organizes the training data into clusters using Part and Select Algorithm. These clusters are used to generate different forecasting models, where each one is performed by a different agent. Finally, another agent is responsible for deciding which model will be applied at each forecasting situation. Results present improvements in forecasting accuracy and training performance if compared to other forecasting methods. A discussion of how to use this architecture for the implementation of a more comprehensive model is also addressed.