基于组件技术和遗传小波神经网络的智能建模系统研究

Research on the Intelligent Modeling System Based on the Combination of the Genetic Wavelet Neural Networks and the Component Technology

本文对已有的人工神经网络、小波分析、遗传算法的建模方法进行组合利用和加以改进,建立了智能信息处理器。该系统将大量的观测数据进行小波去噪等预处理后,作为小波神经网络模型的输入训练样本数据,网络训练中利用遗传算法动态修改网络结构和参数,并避免神经网络训练速度慢、容易陷入局部极值的缺点,从而完成数据挖掘和复杂的非线性建模功能;同时以智能信息处理器为基础,基于GIS平台利用组件技术建立扩展性强的智能建模系统。最后以某灌区水资源管理过程中的径流预报为例进行仿真实验,验证了方案的可行性和有效性。

A combined intelligent information processor is developed based on recombining and improving artificial neural networks（ANN） ,wavelet transformation（WT）, and genetic algorithm（GA）. Firstly, mass historical data and field data gathered by multi-sensors on spot are preprocessed using wavelet analysis, which takes the preprocessed data as the input sample of the neural network model, and the synchronouslygenetic algorithm which has the ability of global optimization is adopted to dynamically modify the network structure and parameters and eliminate the rate tardiness of neural network training and relapse into local extremum. This processor can be used for accomplishing complex nonlinear modeling and data mining. Finally, an intelligent modeling system with good expansibility is established by integrating the combined intelligent information processor with GIS using the component technology, and the integrated scheme is described clearly in this paper. In order to verify the feasibility and validity of the modeling methods, a simulation example is given for the runoff forecasting of an irrigation catchment.