混合遗传神经网络算法研究及其在径流预报中的应用

Daily Runoff Forecast System Based on Combined Genetic Algorithm and ANN

根据流域径流预报的特点,针对神经网络原有固定结构学习方法的缺陷,通过对人工神经网络、遗传算法进行组合利用和加以改进,建立了混合遗传神经网络模型,并对其神经网络的结构和权值阈值同时进行了优化。然后以某灌区径流预报为例,分别利用BP(back propagation)算法、本文方法进行仿真试验,验证方案的可行性和有效性,结果表明本文算法既克服了神经网络结构选取的盲目性,给出了优秀的初始权值,又克服了遗传算法耗时的缺点,最终达到了提高网络收敛性能和收敛速度的目的。

According to the characteristics of runoff forecasting in the catchement, an intelligently optimized algorithm based on recombining and improving artificial neural network（ANN）, genetic algorithm（GA） is presented in this paper. This combined algorithm can optimize the structure of neural network（NN）, as well as its weights and threshold values by using the genetic algorithm which has the ability of global optimization to dynamically modify the structure and parameters of ANN and to eliminate rate tardiness of neural network training and relapsing into local extremum. Then, in order to verify the feasibility and validity of the combined intelligent algorithm, authors take some irrigation catchment for example and carry out serial simulation experiments by using BP, the combined intelligent algorithm respectively. The analysis results show that the combined algorithm overcomes the defects of both the blindness of structure choice and the GA＇s time-consuming, and improves the network＇s performance and increases the speed of the network＇s convergence effectually. Lastly, an dynamically intelligent interactive interface of the runoff forecasting system is developed by using the VC. net programming language.