Expert knowledge in long-term groundwater monitoring optimization process: The interactive genetic algorithm perspective

In most practical water resources optimization applications, a number of important subjective issues exist that cannot be represented in numerical optimization procedures. Considering these issues only in a post-optimization analysis of solutions by the expert (engineers, stakeholders, regulators, etc.) does not ensure that the final set of optimal designs address all qualitative issues important to the problem. The Interactive Genetic Algorithm (IGA) promises to overcome these hurdles by involving the expert directly in the online search process to steer the genetic algorithm to a solution or set of solutions that address both quantitative and qualitative criteria. This paper investigates the effect on the overall search process when a single user interacts with the IGA system. Some of the salient control parameters that affect performance of such a framework are algorithmic control parameters (i.e. the GA settings, visualization interfaces, etc.), human control parameters (i.e. the user's cognitive perception, user's degree of risk aversion, human fatigue, etc.), and external control parameters (i.e. environmental noise and uncertainty, etc.). This work begins a rigorous assessment of the effects of different control parameters on the IGA search process by simulating the human decision making process using fuzzy logic models of human preferences as 'pseudo humans'. Comparison of such a system with a conventional optimization framework (that lacks progressive user feedback) is made for a long-term groundwater monitoring optimization problem, and related ramifications are highlighted. Copyright ASCE 2005.