基于样本选择与PSO-ANN的葡萄酒酒精浓度预测

为了提高拉曼光谱定量分析模型的准确性以及稳健性，提出了一种新的样本选择算法——KM法.实验中以40组葡萄酒光谱为分析对象，将KM法与传统的RS，KS，SPXY样本选择算法相比较.实验结果表明: KM法获得的|RMSEP-RMSEC|要优于其他三种方法，剩余预测偏差(RPD)存在显著性差异，说明KM法具有很好的预测准确度.同时，针对BP神经网络易陷入局部极值的问题，将粒子群优化算法用于优化人工神经网络的参数(PSO-ANN)，通过与遗传算法、人工鱼群算法及混合蛙跳算法比较，发现PSO-ANN较之于其他三种方法，能够提高BP神经网络泛化性能，具有收敛速度快、稳健性强及预测精度高等优势.

Prediction of Wine Alcohol Concentration Based on Sample Selection and PSO-ANN

In order to improve the accuracy and robustness of the quantitative analysis model， a new sample selection algorithm named KM was proposed. In the experiment， 40 samples of wine were used as the analysis objects， and the KM algorithms was compared with traditional sample selection algorithms， such as RS， KS and SPXY. The experimental results show that |RMSEP-RMSEC| obtained by KM algorithm is superior to the other three algorithms， and there are significant differences in RPD， which indicates that KM method has good prediction accuracy. In order to overcome the neural network training algorithms drawbacks that BP neural networks converge slowly and is easy to fall into local optimum， the particle swarm optimization algorithm was used to optimize the parameters of artificial neural network (PSO-ANN). The results show that PSO-ANN algorithm can improve the convergence velocity of training， robustness and the accuracy of classification than genetic algorithm， artificial fish swarm algorithm， and shuffled frog-leaping algorithm.