Active Learning in Gaussian Process State-Space Model