環保專案成果報告查詢系統

專案計畫編號
經費年度	112
計畫經費	4,190 千元
預算科目	數位建設-建構開放政府及智慧城鄉服務
專案開始日期	2023/01/03
專案結束日期	2023/12/31
專案主持人	蔡春進
主辦單位	環境部監測資訊司
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
專案分類	環境監測及檢測
中文關鍵字	感測器、低成本空氣感測器、監測、環境物聯網、多變數迴歸
英文關鍵字	PM2.5 sensor, low cost air sensor, monitoring, Environmental internet of things, Multiple variable regression
協同主持人	洪廣華
共同主持人
計畫聯絡人	劉邦珉
計畫聯絡電話	03-5712121#55549
計畫聯絡信箱	zero000000000003@gmail.com

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18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。本計畫統計108至111年間，共計7款PM2.5感測器通過型式驗證並設置在實場監測，以各廠牌為基準不分縣市在同一年度的平均數據品質滿意度，結果顯示Plantower的平均品質滿意度介於83.5~95.1%，數據品質滿意度呈現逐年遞減趨勢，使用3年後的品質滿意度已低於85%。Honeywell的平均品質滿意度介於87.0~92.2%，相較109-110年度的品質滿意度，在111年度略降5%左右。Sensirion的平均品質滿意度介於78.2~95.9%，該品牌在111年度品質滿意度明顯降低，主因為廠商以鄰近空氣品質測站的BAM-1020為參考標準作校正，導致現場PM2.5感測器測值明顯偏低，而數據品質變差。另4款PM2.5感測器包含Sharp、Amphenol、經昌1代及經昌2代，設置時間都少於2年，較難評估長期使用的品質滿意度，其中Sharp在108-109年度兩次的查核，平均品質滿意度僅74.9~76.0間，性能表現較差。本年度截至11月已完成949台PM2.5感測器的現地查核，除新竹縣數據品質滿意度85%略低外，其他縣市數據品質滿意度皆>90%，原因為新竹縣比對時PM2.5濃度較低，使得感測器測值誤差較大。本計畫以廣域科技SAQ-200及SAQ-210分別在屏東測站(BAM-1020)與揚塵測站(E-BAM)的PM10測值比對分析，發現將環境風速與微粒的粒徑分布作為校正參數，可以有效提升感測器測值的準確度。在屏東測站與揚塵測站的PM10感測器使用MLR-R校正後，RMSE由10.1~13.4μg/m3(原廠)降低為7.4~8.6μg/m3 (MLR-R)，數據品質明顯改善。Figaro TGS5141電化學式CO感測器經校正後有良好的數據品質，後續可應於移動源和工廠污染排放監測。而Figaro FECS44-100的電化學式NH3感測器在不同濃度區間，受環境溫濕度影響大導致測值明顯高估。利用NH3感測器與參考儀器半自動化氣體監測系統(PPWD-IC)在大氣環境中平行比對，發現R2僅為0.18，RMSE為127.44 ppb，相關性不佳且測值明顯高估。經本研究MLR(T＆RH)校正後，R2提升為0.65，RMSE降為2.08 ppb，有效提升數據品質。
中文關鍵字	感測器、低成本空氣感測器、監測、環境物聯網、多變數迴歸

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18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。本計畫統計108至111年間，共計7款PM2.5感測器通過型式驗證並設置在實場監測，以各廠牌為基準不分縣市在同一年度的平均數據品質滿意度，結果顯示Plantower的平均品質滿意度介於83.5~95.1%，數據品質滿意度呈現逐年遞減趨勢，使用3年後的品質滿意度已低於85%。Honeywell的平均品質滿意度介於87.0~92.2%，相較109-110年度的品質滿意度，在111年度略降5%左右。Sensirion的平均品質滿意度介於78.2~95.9%，該品牌在111年度品質滿意度明顯降低，主因為廠商以鄰近空氣品質測站的BAM-1020為參考標準作校正，導致現場PM2.5感測器測值明顯偏低，而數據品質變差。另4款PM2.5感測器包含Sharp、Amphenol、經昌1代及經昌2代，設置時間都少於2年，較難評估長期使用的品質滿意度，其中Sharp在108-109年度兩次的查核，平均品質滿意度僅74.9~76.0間，性能表現較差。本年度截至11月已完成949台PM2.5感測器的現地查核，除新竹縣數據品質滿意度85%略低外，其他縣市數據品質滿意度皆>90%，原因為新竹縣比對時PM2.5濃度較低，使得感測器測值誤差較大。本計畫以廣域科技SAQ-200及SAQ-210分別在屏東測站(BAM-1020)與揚塵測站(E-BAM)的PM10測值比對分析，發現將環境風速與微粒的粒徑分布作為校正參數，可以有效提升感測器測值的準確度。在屏東測站與揚塵測站的PM10感測器使用MLR-R校正後，RMSE由10.1~13.4μg/m3(原廠)降低為7.4~8.6μg/m3 (MLR-R)，數據品質明顯改善。Figaro TGS5141電化學式CO感測器經校正後有良好的數據品質，後續可應於移動源和工廠污染排放監測。而Figaro FECS44-100的電化學式NH3感測器在不同濃度區間，受環境溫濕度影響大導致測值明顯高估。利用NH3感測器與參考儀器半自動化氣體監測系統(PPWD-IC)在大氣環境中平行比對，發現R2僅為0.18，RMSE為127.44 ppb，相關性不佳且測值明顯高估。經本研究MLR(T＆RH)校正後，R2提升為0.65，RMSE降為2.08 ppb，有效提升數據品質。

中文關鍵字

感測器、低成本空氣感測器、監測、環境物聯網、多變數迴歸

類型	檔名	檔案大小	說明	最新版公開日期
期末報告	2023-EPA air sensor FINAL report 定稿V1.pdf	10MB		2024/1/10 上午 12:00:00

摘要	This project aims at utilizing BAM-1020 PM2.5 dates to compare and 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 eighteen 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. During year 108 to 111, a total of 7 types of PM2.5 sensors were tested to be qualified and were set up for field monitoring. For the average quality index each year of different types of sensors, the average quality index of Plantower ranged from 83.5 to 95.1%. The data quality deteriorated year by year, and the quality index is lower than 85% after three years of usage. The average quality index of Honeywell ranged from 87.0 to 92.2%. Compared to year 109~110, the quality index of the year 111 dcreased by 5%. The average quality index of Honeywell ranged from 78.2 to 95.9%. The quality index of this sensor decreased significantly in year 111 since the manufacturer calibrated the sensors’ output data by using the data of BAM-1020 deployed at the nearest air quality monitoring station, resulting the underestimation and the poor quality of the monitoring data of on-site PM2.5 sensors. The other 4 PM2.5 sensors include Sharp, Amphenol, the first and second generation of Vision. The deploy time of the sensors are less than two years, making it difficult to evaluate their quality of long-term usage. Among them, the averge quality of Sharp ranged from 74.9 to 76.0% during year 108 ~ 109, reporting poor performance. Until now, 949 PM2.5 sensors have been field-verified. Except for Hsinchu County with a data quality rate of 85%, the data quality rates in other counties and cities are above 90%. It indicates that the calibration coefficients did not cover higher environmental PM2.5 concentrations, leading to significant errors. This project conducts comparative analysis of PM10 measurements using the Ensens SAQ-200 and SAQ-210 sensors at the Pingtung monitoring station (BAM-1020) and the fugitive dust monitoring station (E-BAM). It was found that using environmental wind speed and particle size distribution as calibration parameters can effectively improve the accuracy of sensor readings. After MLR-R calibration to the PM10 sensors at the Pingtung monitoring station and the fugitive dust monitoring station, the RMSE decreased from 10.1~13.4μg/m3 (manufacture) to 7.4~8.6μg/m3 (MLR-R), resulting in a significant improvement in data quality. After calibration, the Figaro TGS5141 electrochemical CO sensor exhibits excellent data quality, making it suitable for subsequent applications in mobile sources and industrial pollution emission monitoring. Furthermore, the Figaro FECS44-100 electrochemical ammonia gas sensor is significantly prone to overestimating measurements in various concentration ranges due to the influence of environmental temperature and humidity. The ammonia gas sensor was compared with the reference instrument, a semi-automated gas monitoring system (PPWD-IC), in the atmospheric environment. It was found that the R2 value was only 0.18, with an RMSE of 127.44 ppb, indicating poor correlation and significant overestimation of measurements. After MLR (T＆RH) calibration was conducted in this study, the R2 value improved to 0.65, and the RMSE decreased to 2.08 ppb, resulting in an improvement in data quality.
英文關鍵字	PM2.5 sensor, low cost air sensor, monitoring, Environmental internet of things, Multiple variable regression

摘要

This project aims at utilizing BAM-1020 PM2.5 dates to compare and 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 eighteen 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. During year 108 to 111, a total of 7 types of PM2.5 sensors were tested to be qualified and were set up for field monitoring. For the average quality index each year of different types of sensors, the average quality index of Plantower ranged from 83.5 to 95.1%. The data quality deteriorated year by year, and the quality index is lower than 85% after three years of usage. The average quality index of Honeywell ranged from 87.0 to 92.2%. Compared to year 109~110, the quality index of the year 111 dcreased by 5%. The average quality index of Honeywell ranged from 78.2 to 95.9%. The quality index of this sensor decreased significantly in year 111 since the manufacturer calibrated the sensors’ output data by using the data of BAM-1020 deployed at the nearest air quality monitoring station, resulting the underestimation and the poor quality of the monitoring data of on-site PM2.5 sensors. The other 4 PM2.5 sensors include Sharp, Amphenol, the first and second generation of Vision. The deploy time of the sensors are less than two years, making it difficult to evaluate their quality of long-term usage. Among them, the averge quality of Sharp ranged from 74.9 to 76.0% during year 108 ~ 109, reporting poor performance. Until now, 949 PM2.5 sensors have been field-verified. Except for Hsinchu County with a data quality rate of 85%, the data quality rates in other counties and cities are above 90%. It indicates that the calibration coefficients did not cover higher environmental PM2.5 concentrations, leading to significant errors. This project conducts comparative analysis of PM10 measurements using the Ensens SAQ-200 and SAQ-210 sensors at the Pingtung monitoring station (BAM-1020) and the fugitive dust monitoring station (E-BAM). It was found that using environmental wind speed and particle size distribution as calibration parameters can effectively improve the accuracy of sensor readings. After MLR-R calibration to the PM10 sensors at the Pingtung monitoring station and the fugitive dust monitoring station, the RMSE decreased from 10.1~13.4μg/m3 (manufacture) to 7.4~8.6μg/m3 (MLR-R), resulting in a significant improvement in data quality. After calibration, the Figaro TGS5141 electrochemical CO sensor exhibits excellent data quality, making it suitable for subsequent applications in mobile sources and industrial pollution emission monitoring. Furthermore, the Figaro FECS44-100 electrochemical ammonia gas sensor is significantly prone to overestimating measurements in various concentration ranges due to the influence of environmental temperature and humidity. The ammonia gas sensor was compared with the reference instrument, a semi-automated gas monitoring system (PPWD-IC), in the atmospheric environment. It was found that the R2 value was only 0.18, with an RMSE of 127.44 ppb, indicating poor correlation and significant overestimation of measurements. After MLR (T＆RH) calibration was conducted in this study, the R2 value improved to 0.65, and the RMSE decreased to 2.08 ppb, resulting in an improvement in data quality.

英文關鍵字

PM2.5 sensor, low cost air sensor, monitoring, Environmental internet of things, Multiple variable regression