An Intelligent Hybrid Genetic Annealing Neural Network Algorithms for Runoff Forecasting

This study tackles the problem of modeling of the complex, non-linear, and dynamic runoff process. To overcome local optima and network architecture design problems of ANN to make runoff forecasting of catchment more accurate and fast, an hybrid intelligent genetic annealing neural network (IHGANN ) algorithms is established by recombining and improving artificial neural network(ANN) and genetic algorithm (GA). The typical approach can be regarded as a hybrid evolution and learning system which can combine the strength of back propagation (BP) in weight learning and GA's capability of global searching the architecture space. However, the standard genetic algorithm(SGA) adopts constant crossover probability as well as invariable mutation probability. It has such disadvantages as premature convergence, low convergence speed and low robustness. Common adaptation of parameters and operators for SGA is hard to obtain high-quality solution, though it promotes the convergence speed. To address this problem, the IHGANN algorithm applies the simulated annealing algorithm to increase the fitness properly, the self adaptation technology to adjust the value of crossover probability and mutation probability. Meanwhile, a fitness normalization formula is introduced and it always gets a positive value. The new formula can guide the population to a proper direction and increase the press for selection of individuals. The similarity is defined to increase the varieties of individuals without increasing the size of population, thus solving the problem of local optimized solution. Moreover, IHGANN's real encoding scheme allows for a flexible and less restricted formulation of the fitness function and makes fitness computation fast and efficient. This makes it feasible to use larger population sizes and allows IHGANN to have a relatively wide search coverage of the architecture space. In order to verify the feasibility and validity of the IHGANN, we give an example for some watershed located on the Jinsajiang river basin, Yunan province, southwest China and carry out serial simulation experiments by using BP, the IHGANN separately. The simulations showed that problems faced by both back propagation algorithm and standard genetic algorithm were overcame by IHGANN. Compared with BP, the IHGNN has faster convergence speed and higher robustness. Lastly, an dynamic intelligent interactive interface of the runoff forecasting system is developed by using the VC.net programming language.