We present a Linear Parameter Varying (LPV) subspace identification method that takes advantage of the recent developments in the Machine Learning community. More specifically, a Radial Basis Function kernel is used to model the predictor’s impulse response of an LPV model and the involved hyperparameters are estimated via a marginal likelihood maximization algorithm. This step is followed by the estimation of the predictor’s impulse response coefficients, evaluated at the training points. Finally, these values are used to estimate the related coefficients of the LPV model. From this point, the algorithm follows the same steps as in the LPV-PBSIDopt algorithm. Simulation results verify that this algorithm can improve the accuracy of the estimated model with respect to the state-of-the-art LPV subspace methods.
- system identification
- linear parameter varying systems
- machine learning
- gaussian processes