Non‐linear MIMO identification of a Phantom Omni using LS‐SVR with a hybrid model selection

Abstract

Here, a multiple‐input–multiple‐output (MIMO) Phantom Omni robot made by SensAble Technologies Inc. is identified by using a least‐square support vector regression (LS‐SVR). To this end, a two‐stage hybrid optimisation strategy combining coupled simulated annealing as a priori optimisation strategy and a derivative‐free Simplex method as a complementary stage is proposed to solve the LS‐SVR model selection problem. This extra step is a fine‐tuning procedure to enhance the optimal tuning parameters and hence lead LS‐SVR to a better performance. Generalised v ‐fold cross‐validation is considered as the criterion of LS‐SVR model selection problem. The Phantom robot model is implemented via OPAL‐RT to assess the performance of the proposed algorithm compared with firefly algorithm and adaptive particle swarm optimisation in solving LS‐SVR model selection in practical application of the Phantom robot modelling. Finally, the proposed approach is validated and implemented in the hardware‐in‐the‐loop based on OPAL‐RT to integrate the fidelity of physical simulation as well as the flexibility of numerical simulations.