摘要
针对燃煤电厂SCR入口氮氧化物(NOx)烟气连续监测系统存在较大延时性和维护过程无法连续测量的问题,提出一种考虑偏差补偿基于改进型PSO优化的BP神经网络软测量模型.首先,利用改进的PSO算法对网络参数进行优化,解决各神经元阀值与权值初始选择随机性强和训练过程易陷入局部极值的问题,从而提升网络的训练速度和精度;其次,构建模型误差补偿网络修正预测结果以提高网络的预测精度与泛化能力;然后将改进模型应用于SCR入口NOx软测量,基于某电厂DCS运行数据仿真计算,并与BP模型、标准PSO优化的BP模型和改进型PSO优化的BP模型进行了对比.仿真实验表明提出的软测量模型具有更高的收敛速度、更强的泛化能力、更佳的测量精度.
Aiming at the problem of large time delay and discontinuity of measurement within in the process of maintenance in the SCR inlet nitrogen oxides(NOx)flue gas continuous monitoring system of coal-fired power plants,a soft-sensing model of BP neural network based on improved PSO optimization considering deviation compensation is proposed.First,network parameters is optimized by the improved PSO algorithm to solve the problem that the initial selection of the threshold and weight of each neuron is strong and the training process is easy to fall into local extremes,so that the training speed and accuracy of the network is improved.Secondly,a model error compensation network is constructed to correct the prediction results and improve the prediction accuracy and generalization ability of the network,the improved model is then applied to the SCR inlet NOx soft sensor.Simulation calculation is established based on DCS operating data of a power plant,then compared with BP model,standard PSO optimized BP model and improved PSO optimized BP model.The simulation experiments show that the proposed soft-sensor model has higher convergence speed,stronger generalization ability,and better measurement accuracy.
作者
李忠鹏
姜子运
LI Zhong-peng;JIANG Zi-yun(School of Automation and Electrical Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)
出处
《兰州交通大学学报》
CAS
2020年第2期58-63,共6页
Journal of Lanzhou Jiaotong University
基金
国家自然科学基金(51767013)
甘肃省教育厅基金(2017A-020)。
作者简介
第一作者:李忠鹏(1991-),男,甘肃白银人,硕士研究生,主要研究方向为检测技术与自动化装置.E-mail:lzprocfly@foxmail.com.通信作者:姜子运(1981-),男,安徽六安人,副教授,工学博士,主要研究方向为过程控制系统建模与设计.E-mail:jiangziyun@mail.lzjtu.cn.
