Heuristic optimization of experimental designs

We propose to integrate different algorithms for constructing D-optimum designs for linear models. Our emphasis is on efficiency gain and on applicability to larger models than those currently considered in the literature. We implement a one-exchange algorithm and use a generalized simulated annealing. This method does not require to construct or to enumerate each point of the candidate set, whose size grows exponentially with the number of variables. In order to handle more complex problems, we develop a procedure generating guided starting designs. A comparison of our results with those found in the literature shows that the simultaneous integration of these algorithms turns out to be very effective. As compared to results from the literature, our algorithmic process allows an increase in efficiency while, for larger models (up to 20 parameters), we attain a 90% D-efficiency level. (C) 2002 Elsevier Science B.V. All rights reserved.