An improved genetic algorithm for rainfall-runoff model calibration and function optimization

The standard binary-coded genetic algorithm (GA) has been improved using th e three strategies of automatic search space shifting to achieve hill-climb ing, automatic search space reduction to effect time-tuning, and the use of independent subpopulation searches coupled with shuffling to deal with the occurrence of multiple regions of attraction. The degrees of search space shifting and reduction are determined by the distribution of the best param eter values in the previous generations and are implemented after every spe cified number of generations. If the best parameter value in successive gen erations is clustering in a small part of the search range, a higher level of range reduction is used. The search shift is based on the deviation from the middle of the current search range of the best parameter values of a s pecified number of previous generations. With each independent subpopulatio n, a search is performed until an optimum is reached. Shuffling is then per formed and new subpopulation search spaces are obtained from the shuffled s ubpopulations. The improved GA performs remarkably better than the standard GA with three global optimum location problems. The standard GA achieves 1 1% success with the Hartman function and fails totally with the SIXPAR rain fall-runoff model calibration and the Griewank function while the improved GA effectively locates the global optima. Taking the number of function eva luations used to locate the global optimum as a measure of efficiency, the improved GA is about two times less efficient, three times more efficient, and 34 times less efficient than the shuffled complex evolution (SCE-UA) me thod for the SIXPAR rainfall-runoff model calibration, the Hartman function , and the Griewank function, respectively. The modified GA can therefore be considered effective but not always efficient. (C) 2001 Elsevier Science L td. All rights reserved.