Modeling-Learning-Based Actor-Critic Algorithm with Gaussian Process Approximator