基于BP神经网络的洞庭湖水位变化的预测

Prediction of the Water Level Fluctuation of Dongting Lake Based on BP Neural Network

摘要: 研究三峡水库运行后洞庭湖水位的变化情况，对洞庭湖湿地修复具有重要意义。本文利用三峡出库流量和对应时间段的城陵矶水位数据作为训练样本，基于Levenberg-Marquardt优化算法建立一个模拟精度较高的四层BP神经网络。并运用该网络对2010年10月份城陵矶水位进行了预测，结果表明：实测值的变化趋势与预测值的变化趋势基本一致，最大误差为3.89%，平均误差为0.91%，所建立的四层BP神经网络的能有效地应用于洞庭湖水位的预测及变化趋势的预报系统中。

Abstract: It is of vital significance to study the water level fluctuation of DongtingLakeespecially after the operation of the Three Gorges Reservoir (TGR), which may provide useful data and necessary information for the repairing works of the wetland. In this study, the historical time series of outflow discharge of the TGR and water level of Chenglingji were taken as training samples. Based on Levenberg-Marquardt (LM) algorithm, a BP neural network with four layers was established, which well-expressed the unknown but literally existed relationship between outflow discharge of the TGR and the water level of Chenglingji. Then it was applied to the water level prediction in October 2010 of Chenglingji. It is indicated that the trend of actual value and forecast value are in substantial agreement and the maximum and average errors are 3.89% and 0.91%, respectively. It is shown that BP neural network has fairly good simulation accuracy and can be satisfactorily utilized to predict the water level fluctuation of theDongtingLake.