LM-BP算法在径流预测中的应用

Application of LM-BP Algorithm on Runoff Prediction

为改善常规BP神经网络的性能,根据Nguyen-Widrow初始化规则对网络层的权值和阈值进行了初始化,利用黄金分割法对隐层节点数所在区间进行了寻优,并采用Levenberg-Marquardt优化算法改进了BP神经网络模型,然后利用经隐层单元优化的LM-BP网络模型对某流域的年径流量进行了预测检验。结果表明:经隐层单元优化的LM-BP网络收敛速度快;2001—2010年年径流量预测结果的相对误差均小于20%,合格率为100%。

In order to improve the peribrmance of basic BE neural network , using Nguyen - Widrow initialization rule initialized weights and thresholds of the network layer, also golden section algorithm was employed to expand the interval of nodes where the hidden layer optimiza- tion approach, Levenberg- Marquardt optimization algorithm was used to improve BP neural network model. Then it made a prediction on annual runoff of a certain river with the LM - BP network model, which had an optimized hidden layer. The results show that LM - BP algo- rithm with optimized hidden layer not only speeds up the network convergence, but also makes high prediction precision; the relative error of annual runoff prediction from 2001 to 2010 are all less than 20% and the pass rate is 100%.