Abstract: An online identification algorithm for multiple model based on minimum entropy clustering was investigated. The number of local models and corresponding weights was calculated by the entropy based fuzzy subtractive clustering approach, and the regularity degree of the local system was considered along with the clustering process. Parameters of local models could be estimated online by the weighted recursive least square method. The waste heat recovery Organic Rankine Cycles system was used to demonstrate the algorithm. The result shows the identified multi-model not only can reach an accuracy and reliability identification result, but also has a stronger self-adaptability for uncertain external disturbances.

Key words: multiple model, nonlinear, minimum entropy, online identification