From: Paul
Date: Sun, 20 May 2001 00:19:46 +0100
Subject: [TSSP] Genetic optimisation
One of the difficulties ahead is that by adding in the primary resonator we now have quite a few extra dimensions to the resonator configuration space. The virtual secondary database is spanned by seven dimensions: h, b, h/d, awg, sr, td, and tb. We must add another five dimensions to accomodate flat spiral and conical primaries, and goodness knows how many are required to describe the gap and cap. As a means of zeroing in on an 'optimum system' I was thinking of trying the following method: Preparation: Start by generating say 1000, more or less randomly chosen designs. Selection: Compute the performance of all the coils and select the best say 100. Discard the remaining 900. Propagation: Take the design parameters of the successfull 100 and intermix them to produce another 900 offspring. Mutation: Randomly tweak by a small amount the parameters of some of the systems. Loop: Back to the selection step. Begins with a fairly high level of mutation, gradually reducing over say 100 generations. Would need about 100k coils to be computed, that's around 10 times the vsd effort, but surely considerably less than visiting at least 2 or 3 points on perhaps 12 or 15 axes. Not that any of the results would be particularly realistic, as I don't suppose we can take into account the breakout characteristics of the toploads involved. Is there any hope of being able to calculate a reasonable estimate of the voltage at which a particular topload will discharge? If not then any attempt at optimisation is just a piece of recreational mathematics. Food for thought. Comments welcome. Cheers, -- Paul Nicholson, Manchester, UK. --
Maintainer Paul Nicholson, paul@abelian.demon.co.uk.