Comparing Search Algorithms for the Temperature Inversion Problem

Several inverse problems exist in the atmospheric sciences that are computationally costly when using traditional gradient based methods. Unfortunately, many standard evolutionary algorithms do not perform well on these problems. This paper investigates why the temperature inversion problem is so difficult for heuristic search. We show that algorithms imposing smoothness constraints find more competitive solutions. Additionally, a new algorithm is presented that rapidly finds approximate solutions.

• Type: Conference
• Year: 2004
• Where: GECCO
• Authors: Monte Lunacek, L. Darrell Whitley, Philip Gabriel, Graeme Stephens
• Keywords: Algorithms Imposing Smoothness, GECCO 2004, Several Inverse Problems, Standard Evolutionary Algorithms