Comparison of Nested Simulated Annealing and Reactive Tabu Search for Efficient Experimental Designs with Correlated Data

Abstract

More complex analysis of correlated data has created a need for algorithms to find efficient designs for known correlation structures. Reactive Tabu Search (RTS) is adapted to the design of experiments with correlated error. It uses a scavenging descent local search and repeat returns to designs in the search path are used to invoke diversification strategies. Nested Simulated Annealing (NSA) is used in block design and adapted here for correlated data. NSA uses supercooling and reannealing to speed and continue the search. NSA is shown to be a less efficient algorithm than RTS as measured by interchanges tested and computing time.