基于PCA和神经网络的农残含量预测模型研究

为提高快速检测农残含量的精度，针对建模数据特征发生明显变化的实际情况，提出了一种结合主成分分析（PCA）和神经网络的分段多模型方法。提取建模数据的前2个主成分作为模型的输入，分别使用主成分回归（PCR）和BP/RBF神经网络建立单一及分段多模型。通过计算模型验证集的输出总误差和误差百分比，对比模型检测精度。试验表明：与单一模型相比，利用神经网络建立的分段多模型可以显著降低农药含量的预测误差，使用BP和RBF网络建立的低浓度段模型的输出误差百分比分别为0.8%和0.4%,RBF网络效果更好。该方法可以在待测农药的较大浓度范围内实现定量检测，具有较强的实用性。

Study on Prediction Model of Pesticide Residue Content Based on PCA and Artificial Neural Network

In order to improve the accuracy of rapid detection pesticide residue content,a multi-section model is proposed based on principal components analysis (PCA) and neural network.The method can solve the problem that the modeling data characteristics changes obviously.The first two principal components extracted from modeling data are used as input of models.Single models and multi-section models are established by principal components regression (PCR) and BP/RBF neural network respectively.The accuracy of the models is compared by the total output error and output error percentage of the verification set.Experimental results show that the multi-section models built by using BP/RBF network can significantly reduce the prediction error compared with the single models.The output error percentage of the multi-section models established by BP and RBF network is reduced to 0.8% and 0.4% respectively,so the RBF model effect is better.This method can be used for quantitative detection in the larger concentration range of tested pesticides and has strong practicability.