【摘要】 为了确定合理数据处理策略,提高基于神经网络的叶绿素a含量预测精度,采用7种数据处理方案和5种神经网络输入参数组合,研究了不同数据处理策略对叶绿素a含量预测精度的影响。结果表明:数据处理可有效提高基于神经网络的水体中叶绿素a含量预测精度,不同数据处理策略得到的主成分不同,对预测精度的提高程度不同;在输入参数数量相同情况下,以格拉布斯准则处理异值点,再采用局部多项式回归进行数据平滑所得的神经网络预测精度最高;4个输入参数情况下,预测精度在进行数据处理后最高可达到0.986,比采用原始数据提高23.25%。更多还原

【Abstract】 To determine a reasonable data processing strategy for improving the prediction accuracy of chlorophyll-a content based on neural network, the effects of different data processing strategies were studied by using seven data processing schemes and five neural network input parameters. The results show that data processing can effectively improve the prediction accuracy of chlorophyll-a content in water. Different data processing strategies have different main components, thus contributing different improving effects to the prediction. In the condition of inputting the same parameters, the neural network prediction accuracy is maximized by treating the outliers with the Grubbs criterion and using "Savitzky-Golay" to smooth the data. In the case of inputting 4 parameters, the prediction accuracy can reach 0.986 after data processing, which is 23.25% higher than directly using the original data.更多还原