NUMERIC GENETIC ALGORITHM .1. THEORY, ALGORITHM AND SIMULATED EXPERIMENTS

A numeric genetic algorithm (NGA) that optimizes value parameters, is described. The most attractive feature of NGA is that it is suitable for the optimization of a wide variety of problems in chemometrics, such as calibration, parameter estimation, non-linear model building and multi-dimensional data resolution. The representation of individuals and genetic operators such as crossover and mutation is designed to deal with value parameters. A new genetic operator, memory, is also developed to speed up and improve the evolution process. Two architectures of NGA are constructed and discussed in detail. One realizes bit operation of floating point and the other is numerical operation. The genetic parameters in NGA are also discussed in detail. These new algorithms are applied to find the global optima of some simulated mathematical functions and to optimize the parameters in a non-linear parameter estimation problem. The results show that for those mathematical functions NGA succeeds in converging to the global optima very fast and efficiently even when the optimization interval is enlarged and can always converge to the global optimum for these parameter estimation problems.