COMPARISON OF LOCAL AND GLOBAL OPTIMIZATION TECHNIQUES FOR DIFFUSION BATTERY DATA-ANALYSIS

The recovery of a size distribution from a set of diffusion battery de position measurements is a typical example of a data inversion problem . A range of solution methods have been proposed and most rely on an i terative optimisation procedure. We used a non-linear regularisation a pproach and investigated the differences between local and global opti misation techniques. For local search methods the final solution depen ds on the starting point of the search and the probability of finding a good solution decreases as the search space becomes more complex. We present a global optimisation method based on simulated annealing. It is shown that good solutions can be consistently found using this met hod, but, that considerable computing time is required. The same solut ions were found far more rapidly by performing repeats of a local sear ch method from a range of start points. We conclude that the simulated annealing technique offers little advantage in the present applicatio n, although it has some attractive theoretical and practical propertie s and may prove to be useful in other aerosol data inversion problems. (C) 1997 Elsevier Science Ltd.