基于人工神经网络的胶州湾浮游植物生物量模拟

文章首先介绍了BP网络数据标准化、隐层神经元选择、网络训练和有效性检验方法。以2003年胶州湾环境监测资料为基础,建立了多输入单输出的3层BP人工神经网络模型,采用8个水环境因子预测浮游植物生物量(Chla浓度)。检测集样本网络预测值与观测值的相关系数为0.8943,平均绝对误差为11.33%。为避免个别网络输入初值对输出的干扰,采取全局灵敏度的方法,分析了各水环境因子变化对浮游植物生物量的相对影响。结果表明,浮游植物生物量对各水环境因子变化响应的敏感系数顺序为DOCODPO4-PSSTpHOilDINSiO3-Si。

Simulation of phytoplankton biomass in Jiaozhou Bay by means of BP network model

In this paper,data standardization of sample,choice of hidden layer neurons,network training and validation methods of BP type artificial neural network have been illustrated.The ocean environment monitoring data of Jiaozhou Bay in 2003 were taken as training and testing samples to build a multi-input single-output three-layer BP type artificial neural network and eight water quality parameters were adopted to forecast biomass （Chl a concentration）changes.The correlation coefficient between the predicted biomasses of phytoplankton by the model and the observed values was 0.894 3 and the average absolute error was 11.33%.In order to avoid the disturbance of different initial value of inputs effectively,global sensitivity analysis method was adopted for calculating the relative influence degree of various environment indicators on phytoplankton biomass.The sensitivity analysis indicates that the parameters order of importance to phytoplankton biomass is DO〉COD〉PO4-P〉SST〉pH〉Oil〉DIN〉SiO3-Si.