Two-level interacting multiple model algorithm

Due to the limitation of Markov parameters, the estimation accuracy of a standard interacting multiple model (IMM) algorithm in noise identification will deteriorate when too many models are chosen. A two-level IMM is proposed. In the proposed model, the model set including many models is divided into several model subsets. It is assumed that the transition of one model subset to another belongs one Markov chain and the transition of models in another Markov chain on the condition that the transition of corresponding model subsets happens. The formulas of input interaction, filtering, probability and output interaction are given. Simulation results for identifying the process noise with two abruptly varied statistical parameters show that the two-level IMM is better than the standard IMM in accuracy.