洪泽湖水位预测模型的研究

A Water Level Prediction Model of Hongze Lake

水位预测是进行洪水监测的关键,而洪泽湖水面面积又随着水位的上升呈不规则扩大趋势,因此,其水位预测满足不规则非线性函数关系,不易使用某个函数进行逼近。采用了BP神经网络对历年的水文信息进行学习、建模,实现了对这种不规则函数的拟合,并支持在线学习及适时调整。另外,使用改进的粒子群优化算法(PSO)对常规的BP网络进行训练。实验结果表明使用由改进的粒子群优化算法进行训练的BP神经网络进行的水位预测的精度有显著提高,并且在训练过程中尽可能地避免收敛于局部最优值。

The prediction of water level is the key to flood monitoring, and the surface area of Hongze Lake expands irregularly with the rising of water level. Therefore, the prediction of water level is corresponding to irregular nonlinear function, and is not easy to use a function approximation. In this article, BP neural network learning and modeling was used to achieve irregular function fitting. It supports the timely adjustment and online learning. In ad- dition, the improved particle swarm optimization （IPSO） was used to train the BP network. The experimental results show that the accuracy of water level prediction has been improved remarkably by using IPSO for training the neural network, and in the training process, it avoids the convergence to the local optimum value as much as possible.