BP神经网络在太湖不同湖区叶绿素a浓度短期预测的应用初探

摘要：人工神经网络具有强大的非线性能力，能对复杂的水环境系统中非线性行为进行准确有效地预测。本文选择太湖的梅梁湾和湖心区两个典型湖区为研究对象，分别设置4个和2个采样点。通过对其2006-2008年三年的常规水质监测参数进行主成分分析，选择合适的输入因子及最优的网络参数，建立优化的BP网络模型，以实现叶绿素a浓度的月预测。结果表明，梅梁湾湖区和湖心区的预测值与实测值的平均相对误差分别为71%和39%，两者预测精度均较低，原因与太湖的水动力条件、水文气象及藻型生态系统等因素有关。

Applications of back propagation neural network for Short-term predicting the concentration of chlorophyll-a of different regions in Lake Taihu

Artificial Neural Network(ANN) has powerful nonlinear capacity,can exactly predicting the non-linear behavior in the water environmental system.This article selected Meiliang Bay and centre of the lake Taihu as study objects,respectively set four and two sampling spots.We selected appropriate input factors and the optimal network parameters through principal component analysis of water quality monitoring parameters in year 2006-2008,then established optimal BP network model to achieve the monthly predict chlorophyll-a concentration.The results showed that the Meiliang bay and centre of the lake Taihu respectively had average relative error of 71% and 39%,and the primary reasons of the poor predicing accuracy were hydrodynamic conditions,hydrometeorologicalin of Taihu Lake and factors of eco-system of algae.