Model-based adaptation for sample efficient transfer in reinforcement learning control of parameter-varying systems