A Novel Dynamic Model Selection Method for VSIMM Filter

This paper proposes a novel dynamic model selection method for variable structure interacting multiple model (VSIMM) filter based on Jensen-Shannon (JS) divergence and sojourn-time-dependent semi-Markov chain (STDM). Traditional interacting multiple model (IMM) algorithms struggle to effectively adapt to frequent changes due to their reliance on fixed-structure model sets and fixed mode transition probabilities, and its performance critically depends on model set selection. To address this, we introduce JS divergence as a novel symmetric and bounded measure for evaluating the discrepancy between candidate models and true states. The JS divergence-based mechanism dynamically prunes redundant models, significantly reducing computational redundancy and model mismatch errors. Furthermore, in the proposed semi-Markov chain-driven VSIMM algorithm, the sojourn time of motion modes follows a stochastic distribution with memory effects, where transition probabilities are dependent on sojourn time. This mechanism enables the model to more accurately reflect the actual states of targets, thereby enhancing model adaptability. Simulation results demonstrate that the proposed algorithm outperforms conventional methods.