Identification of Hybrid Systems with Particle Filtering and Expectation Maximization
This talk addresses the problem of parameter identification for a class of hybrid systems with continuous states and discrete time-varying parameters that can take different values from a finite set at each time instance. The identification of such systems typically results in non-convex formulation. Although these problems can be solved as a mixed integer program, the resulting complexity may be intractable. Another approach involves heuristics in order to deliver approximate solutions. An offline (batch) algorithm is introduced, that combines Particle Filter and Expectation Maximization for the identification of such systems. The performance of the method is demonstrated on simulated systems, and on experimental diauxic bacterial growth data.