Combined Use Of Genetic Programming And Decomposition
Techniques For The Induction Of Generalized Approximate
Throughput Formulas In Short Exponential Production
Lines With Buffers
Proceedings of the 30th International Conference on
Computers & Industrial Engineering, Tinos Island, Greece, (28 June-2 July 2002)Abstract
An attempt is made to combine standard decomposition
techniques and genetic programming approaches, for the
induction of generalized approximate throughput
formulas in short exponential serial production lines
with finite intermediate buffers. The domain of serial
production lines lacks the existence of general
formulas for acquiring useful measurements and line
characteristics, such as throughput. Throughput
approximation in literature takes place usually with
the aid of algorithmic computer-based decomposition
techniques. In this paper, decomposition-based data for
every different number of stations are used as training
cases into a genetic programming scheme, which tries to
generalize the calculation of throughput within a
single mathematical formula. The proposed formula,
obtains accuracy higher than 99% for the training
(i.e,. known) data, whereas, it deviates, on average,
5-15% from the accurate decomposed value, for testing
(i.e. unknown) production line characteristics.