Enhancing Large Language Model Induced Task-Oriented Dialogue Systems Through Look-Forward Motivated Goals