不同环境条件下通风率标准棒修正算法的研究与应用  

Research and application of correction algorithm for ventilation rate standard rods under different environmental conditions

作　　者：史占东 杨荣超[1] 曾波[1] 王松 沈俊祎 张鹏飞[1] 于千源 范黎[1] 张凯[2] 李栋[1] 张勍[1] 苗芊[1] SHI Zhandong;YANG Rongchao;ZENG Bo;WANG Song;SHEN Junyi;ZHANG Pengfei;YU Qianyuan;FAN Li;ZHANG Kai;LI Dong;ZHANG Qing;MIAO Qian(Zhengzhou Tobacco Research Institute of CNTC,Zhengzhou 450001,China;China Jiliang University,Hangzhou 310018,China)

机构地区：[1]中国烟草总公司郑州烟草研究院,郑州高新技术产业开发区450001 [2]中国计量大学,杭州钱塘区310018

出　　处：《中国烟草学报》2025年第1期40-47,共8页Acta Tabacaria Sinica

基　　金：中国烟草总公司郑州烟草研究院青年人才托举工程计划项目“通风率标准棒测量结果准确性及调控技术的研究”(郑烟院科[2020]11号);国家烟草专卖局标准项目“低阻值吸阻标准棒和细支通风率标准棒的研制与验证”(国烟科[2021]55号)。

摘　　要：为了解决通风率标准棒在不同环境条件下使用过程中通风率发生偏离的问题,提出了一种基于BP神经网络的不同环境条件下通风率标准棒的修正方法。通过品牌1、品牌2各3个不同规格的通风率标准棒在不同环境条件下的测试数据,分别构建了BP神经网络模型、对模型在不同品牌间的应用效果进行了评估。结果表明:①基于品牌1通风率标准棒的测试数据建立的BP神经网络模型(BP)、粒子群优化的BP神经网络模型(PSOBP)、遗传算法优化的BP神经网络模型(GAOBP)均能逐步收敛,相应的验证样本均方误差分别为2.5E-06、3.2E-06和3.5 E-06,测试样本相对误差(Relative Error,RE)平均值分别为0.09%、0.02%、0.03%,3种神经网络模型均能较好的实现通风率标准棒在不同环境条件下的修正。②基于品牌2通风率标准棒的测试数据作为测试样本,BP、PSOBP和GAOBP 3种模型RE平均值分别为-1.72%、-1.45%、-0.67%,四分位值间距(IQR)分别为7.88%、19.14%、4.81%,说明神经网络模型在不同品牌通风率标准棒的应用过程中,GAOBP模型修正结果优于PSOBP和BP模型。该研究可为提高实际环境条件下通风率测量结果的准确性提供支持。To address the deviation of ventilation standards under different environmental conditions,this paper proposes a correction method based on a BP neural network for ventilation rate standard rods under varying environmental conditions.Using test data from three different specifications of ventilation rate standard rods from Brand 1 and Brand 2 under different environmental conditions,BP neural network models were constructed and the application effectiveness of these models across different brands was evaluated.The results show that:①The BP neural network model(BP),the particle swarm optimization BP neural network model(PSOBP),and the genetic algorithm optimization BP neural network model(GAOBP),all established using the test data from Brand 1's ventilation rate standard rods,were able to gradually converge,with mean squared errors(MSE)for the corresponding validation samples being 2.5E-06,3.2E-06,and 3.5E-06 respectively.The average relative errors(RE)for test samples are 0.09%,0.02%,and 0.03%,respectively,indicating that all three neural network models can effectively correct the ventilation standards under different environmental conditions.②Using the test data of ventilation standards from Brand 2 as test samples,the average RE values for BP,PSOBP,and GAOBP models are-1.72%,-1.45%,and-0.67%,respectively,with the interquartile range(IQR)values of 7.88%,19.14%,and 4.81%,respectively.This suggests that the GAOBP model provided better correction results than the PSOBP and BP models when applied to ventilation rate standard rods from different brands.This research provides support for improving the accuracy of ventilation rate measurements under actual environmental conditions.

关 键 词：通风率 通风率标准棒 BP神经网络 校准值 修正 

分 类 号：TP1[自动化与计算机技术—控制理论与控制工程]