Self-organizing map neural network as a multiple model identifier for time-varying plants

The identification method of multiple modeling by the irregular self-organizing map (MMISOM) neural network is presented, which improves the authors’ previous method of MMSOM that uses the rectangular SOM. Inputs to the neural networks are parameters of the instantaneous model computed adaptively in each instant of time. The reference vectors of its output nodes are the parameters estimation of the multiple models. At each time, the model with nearest output to the plant output is chosen as the model of the plant. The irregular SOM used in MMISOM is a graph of all the nodes and some of the links that make a minimum spanning tree (MST) graph. It is possible to add new models if the number of models is initially less than the suitable one.