In the field of over-subscribed planning (OSP), goal utility dependencies are more useful than a single goal utility used to improve the plan quality, if goals are not independent. However, existing description models do not follow the grammatical specification of standard planning domain description language (PDDL), so they cannot be used in other OSP planning systems yet. To solve this, this paper presents a new way of describing goal utility dependencies with derived predicate rules and goal preferences, both of which are essential elements of PDDL. The goal of the process is to transform GAI (general additive independence) models into these two elements, where a derived predicate rule is used to describe the explicitly triggering conditions of a goal sub-set. A preference is used to depict explicitly its utility or value and both are indispensable. This compilation mechanism can not only maintain the characteristic of ease-of-use and straightness of GAI models in describing utility dependencies, but can also expend the ability of handling utility dependencies for general OSP planning systems. Also, this paper proves the semantic conservation in the compilation process. Experimental results in some OSP benchmark domains show that the algorithm is feasible and useful for improving the plan quality. It is the first time to describe goal utility dependency with PDDL elements in order to overcome the limitations of existing models.