基于神经网络的降雨预报系统及其改进

降雨预报采用标准BP网络的连续值,用第1、2天数据预报第3天雨量.包括气象因子选取与数据预处理及人工神经网络模拟预报.系统的改进采用增加动量项训练网络,以解决局部极小,提高网络效率.实验表明,该算法使BP网络的设计及训练得到较好解决,特别在网络结构适合情况下能避免BP算法陷入局部极小问题.

Rain Forecast System and Its Improvement Based on Neural Network

The rain forecast adopted the serial value of standard BP neutral network; it used the data of the first and second day to forecast the precipitation rain fall in the third day. The rain forecast includes weather factor choosing, data pretreatment and manual network simulation forecast, The system improvement used the increased dynamic items to train network for solving the extreme small part and improving the network efficiency, The test showed that this method cold improved the design and training of BP network; and it could avoid the extreme small part problem of BP algorithm in suited network structure.