The WINDY domain - A challenging real-world application of integrated planning and scheduling

Abstract

Many renewable sources of energy can harness greater up-time and power output when located in remote and potentially hostile locations. One example of this is wind power, wherein turbines positioned at offshore locations can experience higher and more sustained windspeeds than their on-shore counterparts. However, these traits also lead to increased load and degradation upon components, which in turn means that regular maintenance is required. While on-shore maintenance costs are relatively trivial, the costs associated with offshore maintenance can be several orders-of-magnitude greater. Traditionally, the scheduling of these repairs is performed by hand using a set of pre-determined plans for specific fault-categories (e.g. trivial/minor/major component replacement). This paper formulates this problem as a PDDL domain which encapsulates all of the individual pre-defined plans in a single representation, such that multiple levels of response can be integrated in a single plan. The domain presented is complex in that it contains not only numeric and temporal planning aspects, but that a subset of the domain is heavily geared towards pure scheduling. We include performance results on how a state-of-the-art planner performs on various example scenarios. Copyright © 2013, Association for the Advancement of Artificial Intelligence. All rights reserved.