BP神经网络在河流叶绿素a浓度预测中的应用

Application of Back-propagation Neural Network for Predicting Chlorophyll-a Concentration in Rivers

采用BP神经网络改进算法,通过对津河、卫津河叶绿素a和其他10种水质因子的分析,建立了叶绿素a预测模型,并找出了对叶绿素a浓度变化影响较大的因子。结果表明,BP神经网络在河流系统叶绿素a含量的预测中,具有很好的泛化、推广能力,预测值与实测值的相关系数达到了0.837 6;预测过程中,增加训练样本量或增加输入变量都可以提高预测效果;pH、溶解氧、叶绿素a本底浓度和温度是叶绿素a预测模型的主要参数,四者的相对重要性指数之和达45.7%,对网络输出的准确度有较大影响。

By analyzing the chlorophyll-a concentration and its affecting factors in Jinhe River and Weijinhe River, a model to predict chlorophyll-a concentration was developed using an improved backpropagation （BP） neural network algorithm. The BP neural network is adaptable for predicting chlorophyll-a concentration in river system. The correlation coefficient between predicted values and observed values reaches 0.837 6. In the prediction process, the increasing of the training samples or input variable can improve the prediction effect. The pH value, dissolved oxygen （DO） , chlorophyll-a value and temperature are determining factors of chlorophyll-a. The relative importance index sum of the four numbers is 45.7%, and these factors affect the accuracy of the BP neural network.