%0 Journal Article
%T 基于改进极限学习机的光谱定量建模方法<br>Quantitative spectral modeling method based on improved extreme learning machine
%A 周美灵
%A 郑德智
%A 娄格
%A 刘峥
%J 北京航空航天大学学报
%D 2017
%R 10.13700/j.bh.1001-5965.2016.0459
%X 摘要 依据近红外光谱（NIR）产生原理，提出了粒子群优化极限学习机（PSO-ELM）算法，运用于小样本氨水浓度定量分析。通过优化极限学习机（ELM）隐藏节点参数，解决了极限学习机由于输入权值和隐含层偏差随机产生的建模结果具有随机性的问题，提高了预测模型的稳定性、精确度和泛化性能。经实验验证，优化后的PSO-ELM相比ELM，模型预测集均方根误差由0.011 66减小至0.003 22，预测集相关系数由0.995 1提高至0.997 9。将优化后的模型预测结果与支持向量机（SVM）、BP神经网络算法等传统方法的建模结果进行对比，优化后的PSO-ELM算法具有较高的精确度和良好的泛化性能，模型预测效果优于传统的定量回归分析算法。<br>Abstract：According to the principle of near infrared spectrum (NIR), an optimized extreme learning machine algorithm with particle swarm optimization (PSO-ELM) was proposed and used in ammonia concentration quantitative analysis of small sample. By optimizing the hidden node parameters of extreme learning machine (ELM) algorithm, the problem of randomly generated input weight and hidden layer of ELM leading to random modeling results has been solved. At the same time, the model stability, accuracy and generalization performance were improved. Through the experimental verification, by the optimized PSO-ELM, compared to ELM, the root mean square error of prediction set reduces to 0.003 22 from 0.011 66 and the correlation coefficient of prediction increases from 0.995 1 to 0.997 9. After comparing the optimized model prediction results with the modeling results of traditional support vector machine (SVM) regression and BP neural network algorithm, optimized PSO-ELM offers high accuracy and excellent generalization performance. Model prediction effect is superior to the traditional quantitative regression analysis algorithm.
%K 近红外光谱(NIR)
%K 定量分析
%K 小样本
%K 极限学习机(ELM)
%K 粒子群优化(PSO)算法&lt
%K br&gt
%K near infrared spectrum (NIR)
%K quantitative analysis
%K small sample
%K extreme learning machine (ELM)
%K particle swarm optimization (PSO) algorithm
%U http://bhxb.buaa.edu.cn/CN/abstract/abstract14087.shtml