电子测量与仪器学报

赵纯源.嵌入式人工智能甲醛传感器抗交叉干扰研究[J].电子测量与仪器学报,2025,39(12):239-247

嵌入式人工智能甲醛传感器抗交叉干扰研究

Research on cross-interference resistance for embeddedartificial intelligence formaldehyde sensors

DOI：

英文关键词:gas sensor embedded artificial intelligence cross-interference formaldehyde sensor

基金项目:福建省科技计划项目(2024T3085，2023H0048)资助

作者	单位
赵纯源	中国科学院城市环境研究所区域与城市生态安全全国重点实验室厦门361012

Author	Institution
Zhao Chunyuan	The Institute of Urban Environment Chinese Academy of Sciences, National Key Laboratory of Regional and Urban Ecological Security, Xiamen 361012, China

摘要点击次数: 650

全文下载次数: 461

中文摘要:

针对甲醛电化学传感器在室内环境监测过程中易受乙醇气体交叉干扰的问题，本研究以嵌入式人工智能芯片RV1126作为运算平台上，以电化学甲醛传感器和乙醇传感器作为感知器件获取环境要素输入研制甲醛检测仪器硬件系统，同时结合嵌入式人工智能技术，在设备端以算法补偿的方式来提高甲醛检测仪抗乙醇交叉干扰的能力。本研究以网格采样法获取数据集，并以此数据为基础，分别以线性方程回归法、常规机器学习回归法、神经网络回归法搭建回归模型。通过对各种模型预测的均方根误差进行对比实验，发现线性方程回归法搭建的模型误差最大约600 μg/m3，以常规机器学习回归法搭建的模型可以达到约100 μg/m3，而神经网络回归法搭建的模型预测精度最高，其预测误差均方差可以达到30 μg/m3。而根据世界卫生组织的《室内空气质量指南》室内空气中甲醛的长期平均浓度不应超过80 μg/m3，本研究研制的甲醛检测仪及以神经网络回归法搭建的模型能够在有一定浓度的乙醇气体交叉干扰的环境下实现对甲醛污染的检测要求。

英文摘要:

To address the issue of cross interference from ethanol gas in electrochemical formaldehyde sensors during indoor environmental monitoring, this study developed a formaldehyde detection system using the embedded AI chip RV1126 as the computing platform, with electrochemical formaldehyde and ethanol sensors as sensing components for environmental data acquisition. At the same time, combined with embedded artificial intelligence technology, algorithm compensation is used at the device end to improve the formaldehyde detection instrument’s ability to resist ethanol cross-interference. This study obtained a dataset using grid sampling method, and based on this data, regression models were constructed using linear equation regression, conventional machine learning regression methods, and neural network regression methods. Comparative experiments on root-mean-square error (RMSE) revealed that the linear regression model exhibited the highest prediction error (≈600 μg/m3), conventional machine learning models achieved ≈100 μg/m3 error, while the neural network regression model demonstrated superior accuracy with an MSE of 30 μg/m3. According to the World Health Organization’s Indoor Air Quality Guidelines, which recommend a long-term average formaldehyde concentration threshold of 80 μg/m3, the proposed formaldehyde detector, combined with the neural network model, effectively fulfills detection requirements in environments with ethanol cross-interference, ensuring reliable formaldehyde pollution monitoring.

查看全文查看/发表评论下载PDF阅读器