Identification and Adaptive Control of Change-Point ARX Models Via Rao-Blackwellized Particle Filters

By proper choice of proposal distributions for importance sampling and of resampling schemes for sequentially updating the importance weights, we address the problem of on-line identification and adaptive control of autoregressive models with exogenous inputs (ARX models) with Markov parameter jumps. Particle filters that can be implemented online via parallel recursions are developed by making use of explicit formulas of the posterior means of the time-varying parameters. Theoretical analysis and simulation studies show improvements of this approach over conventional methods