引用本文:贺婷婷,陆军,丁进良,刘长鑫.原油总碳含量的粒子群优化集成神经网络预测模型[J].控制理论与应用,2019,36(2):192~198.[点击复制]
HE Ting-ting,LU Jun,DING Jinliang,LIU Chang-xin.Prediction model of total carbon content of crude oil using ensemble random weights neural network optimized by PSO[J].Control Theory and Technology,2019,36(2):192~198.[点击复制]
原油总碳含量的粒子群优化集成神经网络预测模型
Prediction model of total carbon content of crude oil using ensemble random weights neural network optimized by PSO
摘要点击 2319  全文点击 1070  投稿时间：2017-10-26  修订日期：2018-04-27
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2018.70779
  2019,36(2):192-198
中文关键词  原油总碳含量  预测模型  粒子群  集成学习  神经网络  核磁共振
英文关键词  total carbon of crude oil  prediction model  PSO  ensemble learning  neural network  NMR
基金项目  国家自然科学基金(61590922, 61525302, 61621004)资助
作者单位E-mail
贺婷婷 东北大学流程工业综合自动化国家重点实验室  
陆军 东北大学流程工业综合自动化国家重点实验室  
丁进良* 东北大学流程工业综合自动化国家重点实验室 jlding@mail.neu.edu.cn 
刘长鑫 东北大学流程工业综合自动化国家重点实验室  
中文摘要
      原油评价新技术的研究和应用成为目前世界石油炼制企业致力发展的方向, 也是今后发展的必然趋势. 本文采用核磁共振光谱技术和粒子群优化集成神经网络(PSO-ERNN)模型建立了一种快速评价原油总碳物性指标的新方法. 该模型以随机向量函数连接网络(RVFL)作为基本模型, 采用正则化负相关学习策略集成基本模型, 并采用粒子群优化算法优化各基本模型的最优隐含层节点数(L)以及集成规模的最佳集成个数(M), 最后利用在线学习方法对模型进行更新. 实例验证表明, 所提出的模型显著提高了预报模型的精度, 避免了随机选择L和M对模型精度的影响, 对提高原油评价精度和效率和及时满足加工炼制要求具有应用价值.
英文摘要
      The research and application of the new technology of crude oil evaluation has become the trend of the world petroleum refining enterprises, and it is also the inevitable trend of development in the future. In this paper, a new method to rapidly evaluate total carbon of crude oil was established by using nuclear magnetic resonance (NMR) spectroscopy and ensemble neural network with random weights (ERNN) model which is optimized by particle swarm. The model uses the random vector functional link network (RVFL) as the basic model. A regularized negative correlation learning strategy is used to integrate the basic model. The particle swarm optimization (PSO) algorithm is used to optimize the hidden layer number (L) of each basic model and the number of integrated scale (M). Finally, the online learning method is used to update the model. Experiment results show that the proposed model significantly improves the prediction accuracy and avoids the influence of random choice of L and M on model precision. The model also improves the accuracy and efficiency of crude oil evaluation and it has a practical value to meet the requirements of oil refining in time.