气体预警穿戴系统的传感器校正及浓度预测

发布时间：2018-06-27 14:32

本文选题：气体预警穿戴系统 + 电化学气体传感器　；参考：《东华大学》2017年硕士论文

【摘要】：随着社会经济的不断发展,国内各种行业建设规模的不断提高,各种类型的作业现场,如工业生产,市政维护,矿业开采等,随之而来也伴随着各种有害气体对作业人员的生命与健康的危害。其中CO气体由于其产生范围较广,且无色无味无刺激并带有剧毒性,不易被人体察觉,对现场作业人员的安全与健康造成严重的危害。得益于传感器,机器学习,电子及计算机行业的迅猛发展,针对特殊作业场合的毒害气体检测系统也有了较大的发展。其中穿戴式作业现场毒害气体预警系统将CO气体传感器嵌入到现场作业防护装备中,无需人工干预而独立工作,提高作业人员安全与健康的时间分辨率和空间分辨率。本论文针对穿戴式作业现场毒害气体预警系统对其使用的CO传感器进行灵敏度校正和浓度预测研究与实现。本论文研究使用的开放计算平台是树莓派Raspberry PI Zero,CO传感器采用德国Solidsens公司的CO1000 Micro3,由于Raspberry PI Zero运行Linux系统,支持Python语言,且由于Python简洁性、易读性以及可扩展性,拥有强大且丰富的库,如Numpy,Matplotlib,Pandas和Skelarn,还集成了GUI等工具,相比于Matlab更适用于实际工程中。由于电化学传感器的灵敏度会随着环境中温度,湿度及气压的影响,会对测量结果造成一定的影响,本文通过改进的最小二乘法岭回归方法对电化学气体传感器的灵敏度进行预测。由传感器采集到的CO气体浓度是系列随着时间变化的序列。对时间序列的预测有多种,比如移动平均法,指数自回归模型等经典方法,近年来随着机器学习,人工神经网络等先进算法的发展,机器学习也逐渐运用到时间序列的预测中去,本文将使用决策树回归,支持向量回归方法,移动平均模型,指数将加权移动平均模型对CO气体浓度进行预测,并比较四种种方法的效果,在综合考虑处理器的运行速度和算法拟合精度的条件下,支持向量回归的表现更加优越。
[Abstract]:With the continuous development of social economy, the construction scale of various industries in China is increasing, and various types of operation sites, such as industrial production, municipal maintenance, mining and mining, etc. Followed by a variety of harmful gases to the lives and health of workers. Due to its wide range of production, colorless, odorless and irritant, and highly toxic, CO gas is not easily detected by human body, which causes serious harm to the safety and health of field workers. Thanks to the rapid development of sensors, machine learning, electronics and computer industry, the toxic gas detection system for special work situations has also made great progress. The wearable gas warning system embeds the CO gas sensor into the field protection equipment and works independently without manual intervention to improve the temporal and spatial resolution of the safety and health of the workers. In this paper, the sensitivity correction and concentration prediction of CO sensors used in wearable field poison gas warning system are studied and realized. The open computing platform used in this thesis is the raspberry Pi Zeroy CO sensor using CO1000 Micro3 of Solidsens Company of Germany. Because Raspberry Pi Zero runs Linux system, supports Python language, and because of Python simplicity, readability and extensibility. It has powerful and rich libraries, such as Numpy MatplotlibPandas and Skelarn, and integrates tools such as GUI, which is more suitable for practical engineering than Matlab. Because the sensitivity of the electrochemical sensor will be affected by the temperature, humidity and pressure in the environment, it will have a certain impact on the measurement results. In this paper, the sensitivity of electrochemical gas sensor is predicted by the improved least square regression method. The concentration of CO gas collected by the sensor is a series of changes over time. There are many methods to predict time series, such as moving average method, exponential autoregressive model and so on. In recent years, with the development of advanced algorithms such as machine learning, artificial neural network, etc. Machine learning is also gradually applied to the prediction of time series. In this paper, decision tree regression, support vector regression, moving average model, exponential weighted moving average model are used to predict CO concentration. Compared with the results of the four methods, the performance of support vector regression is superior under the condition of considering the processor speed and the algorithm fitting accuracy.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212

【参考文献】