基于传感器阵列和LightGBM-SR模型的危化品泄露监测方法研究  

Dangerous Chemicals Monitoring Method Based on Sensor Array and LightGBM-SR Model

作　　者：王莉[1] 汤旭翔 周熙乾 WANG Li;TANG Xuxiang;ZHOU Xiqian(Office of Laboratory and Asset Management,Zhejiang Gongshang University,Hangzhou Zhejiang 310018,China;School of Business Administration,Zhejiang Gongshang University,Hangzhou Zhejiang 310018,China;Office of Teaching Adminstration,Zhejiang Gongshang University,Hangzhou Zhejiang 310018,China)

机构地区：[1]浙江工商大学实验室与资产管理处,浙江杭州310018 [2]浙江工商大学工商管理学院,浙江杭州310018 [3]浙江工商大学教务处,浙江杭州310018

出　　处：《传感技术学报》2024年第7期1290-1294,共5页Chinese Journal of Sensors and Actuators

基　　金：浙江省公益技术研究项目(LGF19G010004)。

摘　　要：探索了使用传感器阵列和LightGBM-SR模型的危化品泄露监测方法,采用多个传感器实时获取危化品监测数据,并且采用非线性随机共振(stochastic resonance,SR)模型对监测数据调整获取特征信息。选取ExtraTrees、XGBoost、KNN和LightGBM模型作为研究对照模型,分别使用传感器阵列原始数据和SR调理数据对四种模型进行自主学习拟合,然后对测试集数据进行回归预测。研究结果证明未经非线性模型调理的原始传感器阵列监测数据与四种模型的匹配度有所不足。数据经非线性SR算法处理后代入训练,LightGBM-SR模型准确率由LightGBM模型的78.75%提升至98.34%,ExtraTrees-SR稳定性最佳但实际依然存在用时较长,XGBoost-SR和KNN-SR泛化能力与稳定性良好,但是平均准确率不高。LightGBM-SR模型展现了较高的平均准确率,更适合危化品泄露监测的应用场景。The dangerous chemicals monitoring method using the sensor array and the LightGBM-SR model is studied.Multiple sensors are used to obtain the real-time laboratory safety monitoring data.Non-linear stochastic resonance(SR)model is used to adjust the raw monitoring data.ExtraTrees,XGBoost,KNN and LightGBM models are selected as pattern recognition models.The sensor array raw data and SR adjusted data are input into four pattern recognition models,respectively.The regression prediction is conducted based on the test set data.Results indicate that the raw sensor array monitoring data with the four models prestent low predicting accuracy.Non-linear SR adjusted data with the recoginition models have higher predicting accuracy.The accuracy of LightGBM-SR model is improved from 78.75% to 98.34%.ExtraTrees-SR has the best stability but still needs for longer time.The generalization ability and stability of XGBoost-SR and KNN-SR are accetable,but the average accuracy is relatively low.LightGBM-SR model presente higher average accuracy,which is more suitable for the application of dangerous chemicals monitoring.

关 键 词：危化品泄露监测 毒害气体泄露 非线性模型 LightGBM模型 

分 类 号：TP212.9[自动化与计算机技术—检测技术与自动化装置]