基于PSO-BP神经网络的加氢脱硫柴油硫含量的预测研究

发布时间：2018-03-05 12:55

本文选题：人工神经网络　切入点：硫含量　出处：《石油化工》2017年01期 　论文类型：期刊论文

【摘要】：针对柴油加氢脱硫生产过程中出现的工艺参数和产品质量难以精准控制的问题,提出粒子群优化(POS-BP)神经网络。基于中国石油大庆石化公司1 300 kt/a柴油加氢脱硫装置生产工艺操作台账数据,选取生产过程中的易波动工艺参数构建训练样本集合和测试样本集合,采用PSO-BP神经网络预测生产操作参数变化时精制柴油产品中硫含量的变化,将POS-BP神经网络与神经网络(BP)和遗传算法优化(GA-BP)神经网络进行横向预测效果比较。实验结果表明,BP神经网络预测的均方误差为2.66×10~(-3),GA-BP神经网络预测的均方误差为2.94×10~(-5),PSO-BP神经网络预测的均方误差为2.41×10~(-5);PSO-BP神经网络预测值与实际值最为接近,且预测结果较佳,具有较好的稳定性和泛化能力,能够精确预测生产操作参数变化时精制柴油产品中硫含量的变化。
[Abstract]:In view of the problem that the process parameters and product quality are difficult to control accurately in the process of hydrodesulfurization of diesel oil, A particle swarm optimization (PSO) neural network for diesel oil hydrodesulfurization (HDS) was proposed. Based on the operating data of 1 300 kt/a diesel hydrodesulfurization unit of PetroChina Daqing Petrochemical Company, The training sample set and the test sample set are constructed by selecting the easily fluctuating process parameters in the production process, and the variation of sulfur content in refined diesel oil products is predicted by PSO-BP neural network. The POS-BP neural network is compared with the neural network (BP) and the genetic algorithm optimized neural network. The experimental results show that the mean square error of the BP neural network prediction is 2.66 脳 10 ~ (10) -3GA-BP neural network prediction mean square error is 2.94 脳 10 ~ (10) -5PSO-BP neural network. The mean square error of the network prediction is 2.41 脳 10 ~ (-5) ~ (-5). The predicted value of PSO-BP neural network is the closest to the actual value. The result of prediction is good, and it has good stability and generalization ability. It can accurately predict the change of sulfur content in refined diesel oil products when the operating parameters change.
【作者单位】：齐齐哈尔大学化学与化学工程学院;齐齐哈尔大学计算机与控制工程学院;
【基金】：黑龙江省自然基金项目(B201422)
【分类号】：TE624

【参考文献】