臺灣博碩士論文加值系統

The propagation of air contaminant PM2.5 that threatens public health is hard to predict because it is affected by long short-term measurements involving many atmospheric variables. Air quality prediction systems provide initial information to increase public awareness and are expected to reduce the long-term health impact on public health. However, these heterogeneous sensory systems are not feasible. They are essentially incompatible and computationally expensive due to their massive deployments of sensory nodes. In this study, a collaborative prediction model is proposed to extract spatiotemporal features from real-world scientific datasets which are collected from government monitoring sites and from community-driven microsites in Taiwan. This study inherits the basic idea of horizontal aggregated learning to generate a more robust prediction model by enhancing features of the dataset. In this study, a prediction model i.e., called sparse-fault-tolerant deep learning (SFT-DL) model is designed using combinations of convolutional neural network (CNN) layers and long-short-term memory (LSTM) layers to forecast the PM2.5 propagations. In a nutshell, the proposed model achieves accurate predictive results than the baseline CNN and LSTM by considering the relationship among long short-time measurements. In addition, the collaborative learning framework boosts the robustness of the prediction model, which is assessed using point-based evaluation.

The propagation of air contaminant PM2.5 that threatens public health is hard to predict because it is affected by long short-term measurements involving many atmospheric variables. Air quality prediction systems provide initial information to increase public awareness and are expected to reduce the long-term health impact on public health. However, these heterogeneous sensory systems are not feasible. They are essentially incompatible and computationally expensive due to their massive deployments of sensory nodes. In this study, a collaborative prediction model is proposed to extract spatiotemporal features from real-world scientific datasets which are collected from government monitoring sites and from community-driven microsites in Taiwan. This study inherits the basic idea of horizontal aggregated learning to generate a more robust prediction model by enhancing features of the dataset. In this study, a prediction model i.e., called sparse-fault-tolerant deep learning (SFT-DL) model is designed using combinations of convolutional neural network (CNN) layers and long-short-term memory (LSTM) layers to forecast the PM2.5 propagations. In a nutshell, the proposed model achieves accurate predictive results than the baseline CNN and LSTM by considering the relationship among long short-time measurements. In addition, the collaborative learning framework boosts the robustness of the prediction model, which is assessed using point-based evaluation.

CONTENTS
Verification Letter from the Oral Examination Committee i
Abstract ii
Contents iii
List of Figures v
List of Tables vii
Chapter 1. Introduction 1
1.1. Research Background 1
1.2. Research Motivation 2
1.3. Research Purpose 3
1.4. Research Scope 4
1.5. Research Method 4
Chapter 2. Related Works 6
Chapter 3. Design of Regional Datasets Using Compressive Sampling 9
3.1. Dataset Collection 9
3.2. Design of Sparse Fault-tolerant Dataset 11
Chapter 4. Estimation of PM2.5 Propagations using Multivariate CNN-LSTM Network 17
4.1. Spatiotemporal Feature Extraction 17
4.2. Design of Prediction Model Using Convolutional LSTM Autoencoder 18
Chapter 5. Collaborative Learning Framework for Environmental Sensing 20
5.1. Weighing Coefficient in a Collaborative Learning Scheme 20
5.2. Spatiotemporal Feature Extraction 21
Chapter 6. Experiments 24
6.1. Evaluation 24
6.2. Tuning Energy Concentration Threshold 25
6.3. Collaborative Learning Performance 27
Chapter 7. Discussions 31
Chapter 8. Conclusion 34
References 35

List of Figures
Figure 1. Schema of collaborative learning for environmental prediction sensing which involves datasets from government’s AQ monitoring sites and community-driven microsites 10
Figure 2. Compressive sampling technology is used to process incompatible datasets from different monitoring sites into a sparse fault-tolerant (SFT) dataset 12
Figure 3. SFT dataset consisting of 5 features: SO2 and NOX collected from WPA sites, wind speed collected from CWB; and PM2.5, humidity, and temperature collected from Airbox 15
Figure 4. (a) Univariate and (b) multivariate prediction model using LSTM network 19
Figure 5. The proposed multivariate spatiotemporal prediction model inherits the basic idea of autoencoder 19
Figure 6. The CL scheme maintains regional correlation by training the dataset for each region 22
Figure 7. Comparison between (a) error rate and (b) data saving ratio with various σ on the reconstructed data. Meanwhile (c) evaluates the optimal value of σ that achieves efficient data generation while maintaining data fidelity 27
Figure 8. The training performance of the proposed SFT-DL, LSTM and the CNN models 28
Figure 9. RMSE of the proposed models i.e. LSTM, LSTM, and SFT-DL with various of noise addition 28
Figure 10. The propagation map of PM2.5 generated by several DL models 29
Figure 11. Simulation of data generation on WSN that utilizes three different schemes with several data rates 32

List of Tables
Table 1. Statistical characteristics of the dataset 7
Table 2. CL scheme for PM2.5 spatiotemporal predictions using Cv-Avg algorithm 22

[1]C.-P. Chio et al., “Health impact assessment of PM2.5 from a planned coal-fired power plant in Taiwan,” Journal of the Formosan Medical Association, vol. 118, no. 11, pp. 1494–1503, Nov. 2019, doi: 10.1016/j.jfma.2019.08.016.
[2]Y. P. Wardoyo, U. P. Juswono and J. A. E. Noor, "Measurements of PM2.5 motor emission concentrations and the lung damages from the exposure mice," 2016 International Seminar on Sensors, Instrumentation, Measurement and Metrology (ISSIMM), Malang, 2016, pp. 99-103, doi: 10.1109/ISSIMM.2016.7803731.
[3]Md. F. Khan, Y. Shirasuna, K. Hirano, and S. Masunaga, “Characterization of PM2.5, PM2.5–10 and PM>10 in ambient air, Yokohama, Japan,” Atmospheric Research, vol. 96, no. 1, pp. 159–172, Apr. 2010, doi: 10.1016/j.atmosres.2009.12.009.
[4]Z. Wei, J. Peng, X. Ma, S. Qiu and S. Wang, "Toward Periodicity Correlation of Roadside PM$_{2.5}$ Concentration and Traffic Volume: A Wavelet Perspective," in IEEE Transactions on Vehicular Technology, vol. 68, no. 11, pp. 10439-10452, Nov. 2019, doi: 10.1109/TVT.2019.2944201.
[5]R. Borge, W. J. Requia, C. Yagüe, I. Jhun, and P. Koutrakis, “Impact of weather changes on air quality and related mortality in Spain over a 25 year period [1993–2017],” Environment International, vol. 133, p. 105272, Dec. 2019, doi: 10.1016/j.envint.2019.105272.
[6]C.-J. Huang, Y. Shen, P.-H. Kuo, and Y.-H. Chen, “Novel spatiotemporal feature extraction parallel deep neural network for forecasting confirmed cases of coronavirus disease 2019,” Socio-Economic Planning Sciences, p. 100976, Nov. 2020, doi: 10.1016/j.seps.2020.100976.
[7]H.-C. Chen, K. T. Putra, S.-S. Tseng, C.-L. Chen, and J. C.-W. Lin, “A spatiotemporal data compression approach with low transmission cost and high data fidelity for an air quality monitoring system,” Future Generation Computer Systems, vol. 108, pp. 488–500, Jul. 2020, doi: 10.1016/j.future.2020.02.032.
[8]X. Li and D. Long, “An improvement in accuracy and spatiotemporal continuity of the MODIS precipitable water vapor product based on a data fusion approach,” Remote Sensing of Environment, vol. 248, p. 111966, Oct. 2020, doi: 10.1016/j.rse.2020.111966.
[9]P. Siarry, A. K. Sangaiah, Y.-B. Lin, S. Mao, and M. R. Ogiela, “Cognitive Big Data Science over Intelligent IoT Networking Systems in Industrial Informatics,” IEEE Trans. Ind. Inf., pp. 1–1, 2020, doi: 10.1109/tii.2020.3024894.
[10]M. R. Ogiela, F. Palmieri, and M. Takizawa, “Transformative computing approaches for advanced management solutions and cognitive processing,” Information Processing & Management, vol. 57, no. 6, p. 102358, Nov. 2020, doi: 10.1016/j.ipm.2020.102358.
[11]S. Hur, “Short-term wind speed prediction using Extended Kalman filter and machine learning,” Energy Reports, Dec. 2020, doi: 10.1016/j.egyr.2020.12.020.
[12]B. Xu, W. Lin, and S. A. Taqi, “The impact of wind and non-wind factors on PM2.5 levels,” Technological Forecasting and Social Change, vol. 154, p. 119960, May 2020, doi: 10.1016/j.techfore.2020.119960.
[13]M.-Y. Lin et al., “An instantaneous spatiotemporal model for predicting traffic-related ultrafine particle concentration through mobile noise measurements,” Science of The Total Environment, vol. 636, pp. 1139–1148, Sep. 2018, doi: 10.1016/j.scitotenv.2018.04.248.
[14]Y. Zhang, Z. Huang, and S. Wen, “Spatiotemporal variations of in-cabin particle concentrations along public transit routes, a case study in Shenzhen, China,” Building and Environment, vol. 180, p. 107047, Aug. 2020, doi: 10.1016/j.buildenv.2020.107047.
[15]Y. Etzion and D. M. Broday, “Highly resolved spatiotemporal variability of fine particle number concentrations in an urban neighborhood,” Journal of Aerosol Science, vol. 117, pp. 118–126, Mar. 2018, doi: 10.1016/j.jaerosci.2018.01.004.
[16]P. S. S. de Souza et al., “Detecting abnormal sensors via machine learning: An IoT farming WSN-based architecture case study,” Measurement, vol. 164, p. 108042, Nov. 2020, doi: 10.1016/j.measurement.2020.108042.
[17]W. Wang and Y. Guo, "Air Pollution PM2.5 Data Analysis in Los Angeles Long Beach with Seasonal ARIMA Model," 2009 International Conference on Energy and Environment Technology, Guilin, Guangxi, 2009, pp. 7-10, doi: 10.1109/ICEET.2009.468.
[18]M. Oprea, M. Popescu and S. F. Mihalache, "A neural network based model for PM2.5 air pollutant forecasting," 2016 20th International Conference on System Theory, Control and Computing (ICSTCC), Sinaia, 2016, pp. 776-781, doi: 10.1109/ICSTCC.2016.7790762.
[19]B. B. Ustundag and A. Kulaglic, "High-Performance Time Series Prediction With Predictive Error Compensated Wavelet Neural Networks," in IEEE Access, vol. 8, pp. 210532-210541, 2020, doi: 10.1109/ACCESS.2020.3038724.
[20]Y. Tsai, Y. Zeng and Y. Chang, "Air Pollution Forecasting Using RNN with LSTM," 2018 IEEE 16th Intl Conf on Dependable, Autonomic and Secure Computing, 16th Intl Conf on Pervasive Intelligence and Computing, 4th Intl Conf on Big Data Intelligence and Computing and Cyber Science and Technology Congress(DASC/PiCom/DataCom/CyberSciTech), Athens, 2018, pp. 1074-1079, doi: 10.1109/DASC/PiCom/DataCom/CyberSciTec.2018.00178.
[21]D. Wang, Y. Yang and S. Ning, "DeepSTCL: A Deep Spatio-temporal ConvLSTM for Travel Demand Prediction," 2018 International Joint Conference on Neural Networks (IJCNN), Rio de Janeiro, 2018, pp. 1-8, doi: 10.1109/IJCNN.2018.8489530.
[22]S. Song, J. C. K. Lam, Y. Han and V. O. K. Li, "ResNet-LSTM for Real-Time PM2.5 and PM₁₀ Estimation Using Sequential Smartphone Images," in IEEE Access, vol. 8, pp. 220069-220082, 2020, doi: 10.1109/ACCESS.2020.3042278.
[23]Civil IoT Taiwan, Available online: https://ci.taiwan.gov.tw/dsp/en/environmental_en.aspx. (accessed on 1 February 2021).
[24]T. Li, H. Shen, C. Zeng and Q. Yuan, "A Validation Approach Considering the Uneven Distribution of Ground Stations for Satellite-Based PM2.5 Estimation," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 13, pp. 1312-1321, 2020, doi: 10.1109/JSTARS.2020.2977668.
[25]Z. Tan, X. Li, M. Gao and L. Jiang, "The Environmental Story During the COVID-19 Lockdown: How Human Activities Affect PM2.5 Concentration in China?," in IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1-5, 2022, Art no. 1001005, doi: 10.1109/LGRS.2020.3040435.
[26]B Z. Zou, C. Cheng and S. Shen, "The Complex Nonlinear Coupling Causal Patterns Between PM2.5 and Meteorological Factors in Tibetan Plateau: A Case Study in Xining," in IEEE Access, vol. 9, pp. 150373-150382, 2021, doi: 10.1109/ACCESS.2021.3123455.
[27]J. He, G. Christakos and P. Jankowski, "Comparative Performance of the LUR, ANN, and BME Techniques in the Multiscale Spatiotemporal Mapping of PM2.5 Concentrations in North China," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 12, no. 6, pp. 1734-1747, June 2019, doi: 10.1109/JSTARS.2019.2913380.
[28]K. T. Putra et al., “Federated Compressed Learning Edge Computing Framework with Ensuring Data Privacy for PM2.5 Prediction in Smart City Sensing Applications,” Sensors, vol. 21, no. 13. MDPI AG, p. 4586, Jul. 04, 2021. doi: 10.3390/s21134586.
[29]Y. Yuan, W. Liu, T. Wang, Q. Deng, A. Liu and H. Song, "Compressive Sensing-Based Clustering Joint Annular Routing Data Gathering Scheme for Wireless Sensor Networks," in IEEE Access, vol. 7, pp. 114639-114658, 2019, doi: 10.1109/ACCESS.2019.2935462.
[30]J. Zhang, D. Zhao, C. Zhao, R. Xiong, S. Ma and W. Gao, "Image Compressive Sensing Recovery via Collaborative Sparsity," in IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol. 2, no. 3, pp. 380-391, Sept. 2012, doi: 10.1109/JETCAS.2012.2220391.
[31]J. Zammit and I. J. Wassell, "Adaptive Block Compressive Sensing: Toward a Real-Time and Low-Complexity Implementation," in IEEE Access, vol. 8, pp. 120999-121013, 2020, doi: 10.1109/ACCESS.2020.3006861.
[32]W. Fu, T. Lu and S. Li, "Context-Aware Compressed Sensing of Hyperspectral Image," in IEEE Transactions on Geoscience and Remote Sensing, vol. 58, no. 1, pp. 268-280, Jan. 2020, doi: 10.1109/TGRS.2019.2936229.
[33]J. Ma, Y. Ding, V. J. L. Gan, C. Lin and Z. Wan, "Spatiotemporal Prediction of PM2.5 Concentrations at Different Time Granularities Using IDW-BLSTM," in IEEE Access, vol. 7, pp. 107897-107907, 2019, doi: 10.1109/ACCESS.2019.2932445.
[34]M.-Y. Lin et al., “An instantaneous spatiotemporal model for predicting traffic-related ultrafine particle concentration through mobile noise measurements,” Science of The Total Environment, vol. 636, pp. 1139–1148, Sep. 2018, doi: 10.1016/j.scitotenv.2018.04.248.