用BP神经网络模型评价养殖区水域的富营养化——以湖州地区为例

Eutrophication Assessment for Open Pond Using Artificial Neural Networks : A Case Study in Huzhou City

根据养殖区水域富营养化程度主要影响因素和评价标准,用足够多的BP神经网络训练样本、检验样本和测试样本进行模拟学习,给出了区分养殖区富营养化程度的分界值,能够直观地进行不同等级富营养化程度的划分.所得到的神经网络模型具有较好的泛化能力和预测能力,减少了人为主观因素的影响,该模型有一定的客观性、通用性和实用性.实例分析表明,湖州地区养殖区外荡水域富营养化程度比较严重,处于富营养化和重富营养化状态.

The level of eutrophication is related to many between the eutrophication degree and some influencing factors. Due to the non-linear relationship noted factors, a BP artificial neural network （ANN） is designed for the assessment of the eutrophication for the open water-area. The paper gives detail description on technical issues such as set data generation, training, verification, testing and boundaries. The trained ANN-based model demonstrates the good potential of generalization and is initial-value-invariant of the connection weights. The model is characterized in robustness, practicality, and superior fault-tolerance compared with other alternative methods such as gray-clustering analysis. Using the proposed model, eutrophication level is found to above average for the aquatic-breeding pond in Huzhou city.