Implementation of Improved Hybrid Genetic Algorithm in Parameter Optimization of Highly Non-linear Water Quality Model
Zhou Bin,Qian Xin,Wang Qin-Geng,Zhang Yu-Chao,Yin Fu-Cai
2007-01-01
Abstract:With the progress of numerical simulation, water quality model is becoming more and more complicated; its parameters have increased from several to even dozens now. These parameters are so important that just a small change of them will result in great difference in simulation. So a more efficient parameter calibration method is essential now. Genetic algorithm (GA) is a global searching evolutionary algorithm, which can directly deal with the optimized function, so the function continuousness and derivation are not needed. It is a robust global optimization method. So it has been used in water quality model for many years. But these models are mostly some simple models such as one dimensional river models. If it is used in complicated situations, its two shortcomings will be obvious: premature convergence to local area and low convergence speed. To find a good way in complicated water quality model’s parameter calibration, an improved GA method is needed. The improvements of GA are mainly in these four aspects:(1) Set the maximal modulus of objective function to reduce the raggedness response surface of model parameters, and adjust the average objective function value to a low difference range besides the peak valley in optimal solution space.(2) Use heuristic crossover and Gauss mutation operator to improve the selection tactics. (3) In searching in later stage, use simulated algorithm to enlarge the fitness difference, so that the advantage searching pop can be selected more easily.(4) Hybrid with simplex algorithm to improve the local accuracy when global optimal solution’s area has been ascertains, which is called Improved Real-encoded Hybrid Genetic Algorithm,(IRHGA).The effectiveness of IRHGA in optimization is firstly proved by application in Rastrigin test function. Rastrigin test function has a certain optimal solution that is (0,0) and the objective value is 0. This test shows that IRHGA is useful in non-linear optimization problem. At last, IRHGA is applied in optimization of a complicated water quality model, which is composed of more then ten equations and 37 parameters. In these 37 parameters, some parameters are very sensitive, By using IRHGA, a new series of parameters is found. When the model simulates with the new inputted parameters, it gets very close to the measured data result. It shows that IRHGA is useful and successful in optimizing high non-linear parameter optimization problems.