A general modeling approach to online optimization with lookahead

• We embed online optimization with lookahead into the theory of optimization.
• We develop a precise definition of lookahead in the realm of online optimization.
• We decompose the effect of lookahead into two components.
• We establish the link between discrete event systems and online optimization.
• We formulate a generic modeling framework and provide a classification scheme.
• We use the framework for algorithm analysis in standard online problems.
• We build an information pool for observable lookahead effects.
• We apply wholistic performance measurement using counting distributions.

A vast number of real world problems are coined by an information release over time and the related need for repetitive decision making over time. Optimization problems arising in this context are called online since decisions have to be made although not all data is known. Due to technological advances, algorithms may also resort to a limited preview (lookahead) on future events. We first embed the paradigm of online optimization with lookahead into the theory of optimization and develop a concise understanding of lookahead. We further find that the effect of lookahead can be decomposed into an informational and a processual component. Based on analogies to discrete event systems, we then formulate a generic modeling framework for online optimization with lookahead and derive a classification scheme which facilitates a thorough categorization of different lookahead concepts. After an assessment of performance measurement approaches with relevance to practical needs, we conduct a series of computational experiments which illustrate how the general concept of lookahead applies to specific instantiations and how a knowledge pool on lookahead effects in applications can be built up using the general classification scheme.