基于差分进化算法的BP神经网络脱硫效率预测控制  

Desulfurization Rate Predictive Control of BP Neural Network Based on Differential Evolution Algorithm

作　　者：于沙家 程换新[1] 马彩青[1] 宋文瑞 郭锐捷 Yu Shajia;Cheng Huanxin;Ma Caiqing;Song Wenrui;Guo Ruijie(School of Automation and Electrical Engineering,Qingdao University of Science&Technology,Qingdao,266061,China;Qingdao Haiweisikang Electronic Technology Co.Ltd.,Qingdao,266000,China;Shanxi Bigdata Industry Development Co.Ltd.,Taiyuan,046000,China)

机构地区：[1]青岛科技大学自动化与电子工程学院,山东青岛266061 [2]青岛海威斯康电子技术有限公司,山东青岛266000 [3]山西大数据产业发展有限公司,山西太原046000

出　　处：《石油化工自动化》2023年第3期17-20,共4页Automation in Petro-chemical Industry

摘　　要：针对当前BP神经网络技术应用在大型燃煤热电厂以及燃煤脱硫机组效率数值预测分析计算模型中易陷入局部最优,预测效果可靠性差,收敛算法响应速度过慢,给出了一套结合自适应差分进化算法模型和BP神经网络理论的新一代高效率数值预测优化算法.该算法能自动优化BP神经网络的初始权值和阈值,将优化后的结果直接应用于BP神经网络模型训练中,进而获得脱硫效率的最佳预测网络模型.实验分析结果表明:该模型算法与目前传统的BP网络方法相比,算法收敛速度快,对网络参数的预测及精度估计更高,从而为脱硫系统的改进提出了必要的理论基础.Focusing on the disadvantages of easily falling into local optimal solution,poor reliability of prediction effect and too slow response speed for the BP neural network technology that is applied in the predictive analysis calculation model in the large-scale coal burning thermal power plant and desulfurization unit,a high efficiency numerical prediction algorithm merging flexible differential evolution algorithm&BP neural network is proposed.The algorithm can automatically optimize the initial weight and threshold of the BP neural network.The optimized results are directly applied in the training of BP neural network model,and the best predictive network model of desulfurization efficiency is obtained.The experiments reveal that,comparing to the currently used conventional BP network algorithm,the model algorithm achieves fast algorithm convergence speed,and the prediction and accuracy estimation for network parameter is better,which provides a necessary theoretical basis for the improvement of the desulfurization system.

关 键 词：DE-BP算法 脱硫系统 预测控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]