Application of a genetic algorithm in the development and optimisation of a non-linear dynamic runoff model

Daily rainfall and runoff data for the Kashinagar watershed of the Vamsadhara river basin, Orissa, India collected for the years 1984-1995 have been used for the development and optimisation of a runoff prediction model. The prediction model has been developed by considering the process as non-linear and dynamic in nature. Model orders are determined by multiple correlation analysis and then based on 't-test' values. All positive values of the t-test, obtained through multiple correlation, are considered for model formulation. Genetic algorithm has been employed for the estimation of model parameters and function optimisation. The predicted values by the non-linear rainfall-runoff prediction model were 1-2 days ahead of the measured values. This time problem was corrected by applying backward shift operator technique which improved the correlation coefficient by about 15%. (C) 2003 Published by Elsevier Ltd on behalf of Silsoe Research Institute.