BP神经网络在水华短期预测中的应用

为解决影响因素多、作用关系复杂的水华预测问题,将BP神经网与水体环境因子的高频实测数据相结合,构建了巢湖水华的短期动态预测模型,该模型准确地预测了每次水华发生的时间,预测值与实际观测值相关系数可达0.608 4;在分析BP神经网络自身局限性的基础上,研究了建模过程中输入输出数据预处理、网络结构设计、训练模式选择等问题,给出了水华预测中确定环境因子和建模方案的具体方法.该方法容易移植到其它湖库,提高了模型的实用性和通用性.

Application of BP Neural Network in Algal Blooms Short-Term Forecast

Forecast of algal blooms is a difficult problem because it is influenced by many factors that are in complex relation. This paper for the first time presents a short-term forecast model of the algal blooms in Chaohu Lake based on the combination of high-frequency measured values of environmental factors in water with BP neural network technique. The model could accurately forecast the time of each bloom, and the correlation coefficient between the forecasted and observed values is up to 0.608 4. Further, the issues in the modeling process were analyzed such as the limitation of BP neural network, the input and output data preprocessing, network structure design, training mode selection and other aspects. Finally, a specific method to select environmental factors and its modeling scheme were determined. The proposed model could be easily used on other lakes showing its strong practicability and universality.