環境資源報告成果查詢系統

110年度空污感測器巡檢及數據分析應用計畫

中文摘要 本計畫的工作內容主要為利用BAM-1020比對與校正PM2.5感測器,另外巡檢抽查已布建在17縣市的PM2.5感測器性能,並分析與應用監測的數據,希望提升我國目前已布建的PM2.5及VOC感測器數據品質,對於感測器物聯網的推動及使用有所助益。 本研究發現RH及PM2.5濃度對PM2.5感測器的測值準確性影響很大,感測器測值都是隨著RH變高而增加,原廠數據也都呈現高估的現象,因此使用時需加以校正。另外本計畫也發現目前市售的PM2.5感測器(包含經昌電子、Honeywell、Plantower、Sensirion)的外殼未能密閉造成漏氣現象,反應時間(Response time, RT)都超過2分半以上(介於150~180秒之間)。當PMS5003感測器流量固定(0.1 mL/min)時,RT < 1分鐘,而流量無固定時,RT將明顯增加,最長達20分鐘。結果顯示感測器的採樣流量會影響環境濃度的量測,固定流量下可立即反應當時環境濃度變化且無數據延遲現象,數據表現也較佳。 本計畫利用BAM-1020比對Plantower PMS5003 PM2.5感測器的原始數據(未經布建廠商校正之數據),發現線性迴歸(Linear Regression, LR)判定係數R2分別為0.78 (基隆)、0.77 (忠明)、0.79 (台南)及0.85 (屏東)。經過非線性迴歸公式(Non Linear Regression, NLR)校正後,R2可提昇至0.81 (基隆)、0.81 (忠明)、0.84 (台南)及0.89 (屏東),且平均正規化偏差(MNB)可大幅降低至±10% (原為+25.01~+43.90%),平均正規化誤差(MNE)可大幅降至29%內(原為45.91~61.03%)。Sensirion SPS30 PM2.5感測器的原始數據(未經布建廠商校正之數據),LR判定係數R2分別為0.83 (平鎮)及0.80 (士林)。經過NLR校正後,R2可提昇至0.89 (平鎮)及0.83 (士林),MNB可大幅降低至±5% (原為-33.02~-44.57%),MNE可大幅降至23%內(原為40.00~47.02%)。另Honeywell HPMA115S0 PM2.5感測器的原始數據(未經布建廠商校正之數據),發現LR判定係數R2分別為0.81 (板橋)及0.80 (前金)。經過NLR校正後,R2可提昇至0.88 (板橋)及0.84 (前金),MNB可大幅降低至±5% (原為+13.64~+20.18%),MNE可大幅降至22%內(原為31.31~36.46%)。 本研究利用空品測站的NMHC比對Sensirion SGP30 VOC感測器的原始數據(未經布建廠商校正之數據),發現LR判定係數R2為0.52 (忠明)、0.52 (新竹) 與0.37 (臺南),經過MLR校正後,R2可大幅提昇至0.78 (忠明) 、0.70 (新竹) 與0.67 (臺南)。忠明測站MNB由406.32±280.12% 降至10.70±44.94%,MNE由406.32±280.12% 降至10.70±44.94%,新竹測站MNB由245.44±155.19% 降至 -32.51±65.50%,MNE由247.28±152.24% 降至 52.68±50.71%,臺南測站MNB由185.04±169.59% 降至 -25.36±58.71%,MNE由187.37±167.01% 降至 42.14±48.10%。以忠明及台南等測站上長期佈掛的VOC感測器測值求得VOC感測器的LOD為138.19 ppb。MLR校正後在不同的濃度區間(>100ppbv)結果顯示,忠明測站MNB由261.33±150.55% 降至 -1.80±29.68%,MNE由261.33±150.55% 降至 22.92±18.95%、新竹測站MNB由150.39±97.68% 降至 -19.95±27.02%,MNE由151.03±96.68% 降至 27.72±18.98%、臺南測站MNB由127.34±119.03% 降至 -15.63±29.57%,MNE由128.80±117.45% 降至 27.69±18.77%,已達美國環保署規範建議的熱區追蹤應用等級(MNE<30%)。 本計畫另個重點為巡檢已布建在17縣市的PM2.5感測器,已完成巡檢抽查792台感測器完成率100%。發現各縣市的數據品質滿意度介於78.7~98.4%,大部分縣市都高於80%以上,新竹市在同址查核時僅74.6%,主要係因數據管理疏失所導致,台南市數據品質平均滿意度僅78.7,因較長時間未有維運作業所致。本研究發現感測器廠商多以空品測站的BAM-1020作參考標準,導致現場感測器測值偏低,建議後續可擴充特定區域或工業區的現地查核數量,以利研擬未來現地查核的標準及作業規範。
中文關鍵字 PM2.5感測器、低成本空氣感測器、空氣品質監測、環境物聯網、多變數迴歸

基本資訊

專案計畫編號 經費年度 110 計畫經費 3930 千元
專案開始日期 2021/01/01 專案結束日期 2021/12/31 專案主持人 蔡春進 講座教授
主辦單位 監資處 承辦人 鐘偉瑜 執行單位 台灣PM2.5監測與控制產業發展協會

成果下載

類型 檔名 檔案大小 說明
期末報告 20211227- final report-v2.pdf 18MB

110th year project on the comparison, calibration and field auditing of air pollution sensors

英文摘要 This project aims at utilizing BAM-1020 PM2.5 values to calibrate the PM2.5 sensor data. The calibration results were then applied to analyze the measured PM2.5 data of the randomly selected sensors deployed at seventeen counties or cities to evaluate their performance. It is expected to promote the smart IoT sensor network's application by enhancing the performance of the PM2.5 and VOC sensors. The results from the project indicated that the relative humidity (RH) and PM2.5 concentration had a significant effect on the PM2.5 sensor readings, which increased with increasing RH. Besides, the original manufacture’s data consistently overestimated PM2.5 concentrations. Therefore, calibration of the sensors data was required before being used. In addition, this project found that the outer casing of the commercial PM2.5 sensors, including Vision, Honeywell, Plantower, and Sensirion, were not sealed completely, causing the air leakage. As a result, the sensors' response time (RT) was longer than two and a half minutes (varying from 150 to 180 seconds). The RT of PMS5003 sensor is smaller than one minute when the flowrate of the sensor is fixed (0.1 mL/min). However, the RT will increase significantly to 20 minutes if the flowrate of the sensor is not fixed. This result indicates that the sampling flowrate of sensor will affect the measurement of PM concentration in the environment. The sensor can response to the change of the concentration in the environment immediately without delay. The performance of the data of the sensor also becomes better. This project also calibrated the Plantower PMS5003 PM2.5 sensor data (raw or manufacturer-calibrated data) using the BAM-1020 data. The results showed that the obtain R-squared (R^2) values of the linear regression (LR) were 0.78, 0.77, 0.79, and 0.85 for Keelung (KL), Zhongming (ZM), Tainan (TN), and Pingtung (PT), respectively. By using the non-linear regression (NLR) equation, the R2 values for KL, ZM, TN, and PT were increased to 0.81, 0.81, 0.84 and 0.89, respectively. Meanwhile, the mean normalized bias (MNB) values were decreased substantially to ±10% from +25.01~+43.90% and the mean normalized error (MNE) values were decreased to 29% from 45.91~61.03%. For the Sensirion SPS30 PM2.5 sensor data (raw or manufacturer-calibrated data), the R2 values of the LR were 0.83 and 0.80 for Penzhen (PZ) and Shilin (SL), respectively. In comparison, the R2 values of the NLR for PZ and SL were increased to 0.89 and 0.83, respectively, while the MNB and MNE values were decreased substantially to ±5% from -33.02~-44.57% and 23% from 40.00~47.02%, respectively. For the Honeywell HPMA115S0 PM2.5 sensor data (raw or manufacturer-calibrated data), the R^2 values of the LR were 0.81 and 0.80 for Banqiao (BQ) and Qianjin (QJ) stations, respectively. On the other hand, the R2 values of the NLR for BQ and QJ were increased to 0.88 and 0.84, respectively, while MNB and MNE values were decreased substantially to ±5% from+13.64~+20.18% and 22% from 31.31~36.46%, respectively. In this study, the Sensirion SGP30 VOC sensor data (manufacturer calibration) was also calibrated using the NMHC data from the air quality station of EPA. The R2 was 0.52 for ZM station, 0.52 for HC station and 0.37 for TN station when LR was used for calibration. When the MLR was used for calibration, the R2 of ZM, HC and TN station was increased to 0.78, 0.70 and 0.67 respectively. While MNB and MNE values of ZM station were decreased substantially to +10.70±44.94% from +406.32±280.12% and 32.70± 32.63% from 406.32±280.12%, respectively. Values of MNB and MNE of HC station were decreased substantially to -32.51±65.50% from 245.44±155.19% and 52.68±50.71% from 247.28±152.24%. Values of MNB and MNE of TN station were decreased substantially to -25.36±58.71% from 185.04±169.59% and 42.14±48.10% from 187.37±167.01%. The limit of detection (LOD) of the VOC sensors is calculated to be 138.19 ppbv based on the measured values of the long-term VOC sensors deployed on ZM and TN station. The results of MNB and MNE in different VOC concentration intervals shows that, the MNB and MNE values of ZM station were decreased to -1.80±29.68% from 261.33±150.55% and 22.92±18.95% from 261.33±150.55% respectively. in the interval which VOC concentration is larger than 100 ppbv. For HC station, the MNB and MNE values were decreased to -19.95±27.02% from 150.39±97.68% and 27.72±18.98% from 151.03±96.68% respectively. For TN station, the MNB and MNE values were decreased to -15.63±29.57% from 127.34±119.03% and 27.69±18.77% from 128.80±117.45% respectively. The result has met the standard of hot spot tracking application recommended by the US EPA (MNE<30%). Another focus of this project is to audit the performance of PM2.5 sensors deployed in 17 counties or cities. Until now, the auditing of 792 sensors has been finished accounting for 100% of 600 deployed sensors. The quality index varied from 78.7 to 98.4%, which was higher than 80% in most counties and cities except Hinschu city, which quality index was only 74.6% due to negligence in data management. It was found that the sensor manufacturers used the BAM-1020 data at the air quality monitoring stations as the reference values for the deployed sensors, which resulted in under-measurement when these sensors were deployed at the field. It is recommended that further auditing could be conducted at some specific places or industrial zones to establish the auditing criteria and examination regulations for field auditing in the future.
英文關鍵字 PM2.5 sensor, low cost air sensor, Air quality monitoring, Environmental internet of things, Multiple variable regression