Hidden Markov Model-based Warm-start of Active Set Method in Model Predictive Control

We propose a new idea of statistical warm-start of the active set method (ASM) utilized for solving the problem of model predictive control (MPC). The idea is based on the Hidden Markov Model (HMM). In our HMM, the emissions are defined as the system states. The hidden states of HMM are represented by the active sets of the MPC optimization problem. Given the measurement or estimation of system states, using the HMM we can determine which active set is the most likely optimal. This active set is then used for the initialization of the ASM. We show that our approach can streamline the iterative process of ASM compared to the cold-start or conventional warm-start method using the result from the previous control step.