Current students

Hybrid systems can describe dynamically rich phenomena that combine continuous and discrete dynamics. An often used representation employs different continuous local models (modes), and a switching mechanism between them. In switched systems, the latter is determined by an exogenous finite-valued switching signal which identifies at all times which mode is active. Piecewise affine models instead switch according to a polyhedral partition of the state-input domain. Accordingly, the identification of hybrid systems involves both the estimation of the local dynamics and the switching mechanism, configuring a complex optimization problem. Currently, many identification frameworks have been developed based on classification approach, which ignores the influence from temporal order. The examples of independent data set range from speech recognition to finance models, video segmentation, etc. This inner complexity and nonlinearity bring a lot of challenges. Machine learning approaches could be a solution. For example, the probabilistic graphical model with time series can describe a set of dynamic models including the discrete stats, hidden Markov models, and the continuous ones, Kalman filter & particle filter. These frameworks provide us with sufficient theoretical support to solve the identification problems of hybrid systems.