臺灣博碩士論文加值系統

天氣預測是利用大量的天氣數據如氣壓、氣溫、濕度及風速等參數來進行預測未來天氣的狀態判斷。而天氣預測可以區分成短期、中期以及長期等三種預測方式，其準確度會隨著預測的時間越長而越低，所以目前的天氣預測普遍為短中期預測為主。目前大部分的預測準確性平均可以達到80\% 左右。但是就算目前能達到如此高的預測準確度，但多時候在一些地區會隨著地理環境條件的不同，而影響預測的準確度，其預測的結果與實際的天氣狀況不同的情形也是屢見不鮮。隨著現今機器學習技術的高速發展，如果將這技術應用在預測天氣上，可以提高預測的精準度，因此本研究的目標為利用雲層的圖像以及一些輔助資訊去預測接下來的 30 分鐘內降雨的機率。

一般的天氣預測的方式與天氣一些參數有關，例如大氣氣壓，溫濕度、雲層等，其中『雲層』的判別是一項重要的因素。一般天氣預測會使用到衛星照相的方式得知雲層分布狀況及移動方式來預測該地區的降雨機率或者天氣情況，在古時候更有人以觀察雲層的方式預測會不會降雨，為了達到低成本來利用雲層的觀察，所以本研究會利用機器的『眼睛』——相機去觀察雲層的資訊再配合一般感測器就能得到的氣象參數輔助去預測降雨的機率。蒐集到的雲層資料以及氣象資料會做一些處理，雲層資料會將夜晚的資料過濾以及進行雲層特徵值的提取，而氣象資料會過濾有問題的資料以及將標籤(Labeled)替換成 30 分鐘後的降雨情況，以便作為接下來模型訓練的訓練標籤，最後處理好的資料就會作為接下來模型訓練的輸入資料。

本研究所提出的模型訓練的預測模型架構，將採用卷積神經網路(Convolutional Neural Network, CNN)與遞歸神經網路(Recurrent Neural Network, RNN)組合進行降雨預測，CNN會使用ResNet-152預訓練模型以及RNN内層使用長短期記憶(Long Short Term Memory Network, LSTM)的架構來進行模型訓練。訓練的流程會將雲層圖片透過 CNN 編碼出來出圖片特徵向量，其向量會與氣象資料當成 RNN 輸入進行訓練，最後其訓練之後的模型預測結果準確度為82\%。所以本研究提供了另一個預測降雨的方法，其成本上以及預測的時間上都比一般的預測方式來得有利，並且本研究也透過各種模型驗證的結果，證明了其方法的正確性以及可行性。

Weather forecasting is to predict the future state of weather based on huge amount of weather data, such as atmospheric pressure, temperature, humidity, and wind speed, etc. According to the length of time covered, weather forecasting can be categorized into three types: short-term, medium-term, and long-term. Normally, the prediction accuracy will decrease with the increasing of time. Therefore, current weather forecasts are mainly short-term and medium-term predictions. The average accuracy of most predictions can reach up to 80\%. Nevertheless, such high prediction accuracy still can be affected due to the conditions of different geographical areas. It is common that the predicted result and the actual weather are different. With the rapid development of machine learning technologies nowadays, the prediction accuracy is bound to be improved if we combine machine learning with weather forecast. Thus, the goal of this study is to predict the probability of precipitation within 30 minute based on the images of clouds and some auxiliary information.

Normally, the weather forecast is related to some parameters, such as atmospheric pressure, temperature, humidity, cloud coverage etc., in which cloud coverage is the most important. Ancient people observed the cloud to predict whether or not it would rain. Now, satellite imagery is used to display the movement or distribution of the clouds so as to predict the probability of precipitation or weather conditions in that area.

To observe the cloud coverage at a low cost, this paper uses the camera for cloud cover observation. Together with the meteorological parameters obtained by sensors, we therefore can predict the probability of precipitation. The collected cloud and meteorological data will be processed and the features of the cloud will be extracted but the nighttime data will be filtered out. Problematic information will be removed from the meteorological data and the label will be replaced by the precipitation 30 minutes later as the training level. Finally, all processed data will be used as the input of the next model training.

Our proposed model architecture is a combination of Convolutional Neural Network(CNN) and Recurrent Neural Network(RNN). CNN uses ResNet-152 pre-training model and RNN uses Long Short Term Memory Network(LSTM) layer for model training. The training process will extract the image feature vector by encoding the cloud image through CNN and the vector and meteorological data will be trained as the input of RNN. Finally, the accuracy of the model prediction result after training is 82\%. Therefore, this study provides another method for rainfall prediction, which is more beneficial in term of cost and prediction time compared to general prediction methods. The results have proven the correctness and feasibility of our proposed method.

摘要 - I
Abstract - II
誌謝 - III
目錄 - IV
圖目錄 - VI
表目錄 - VIII
1 緒論 - 1
1.1 前言 - 1
1.2 研究動機與目的 - 2
1.3 論文結構 - 2
2 相關文獻及背景介紹 - 3
2.1 天氣預測 - 3
2.1.1 傳統式天氣預測(Traditional Weather Forecasting) - 4
2.1.2 非傳統式天氣預測(Advanced weather forecasting) - 4
2.2 機器學習(Machine Learning) - 5
2.2.1 監督學習(Supervised Learning) - 5
2.2.2 無監督學習(Unsupervised Learning) - 6
2.2.3 強化學習(Reinforcement learning) - 7
2.3 遷移學習(Transfer learning) - 7
2.3.1 歸納遷移學習(Inductive Transfer Learning) - 8
2.3.2 轉換遷移學習(Transductive Transfer Learning) - 8
2.3.3 無監督遷移學習(Unsupervised Transfer Learning) - 8
2.4 類神經網路(Neural Network, NN) - 8
2.4.1 卷積神經網路(Convolutional Neural Network, CNN) - 10
2.4.2 遞歸神經網路(Recurrent Neural Network, RNN) - 12
2.5 文獻探討 - 15
2.5.1 分類模型驗證 - 15
2.5.2 雲層特徵提取處理 - 18
2.5.3 CNN-RNN 組合模型 - 18
2.5.4 深度殘差網絡(Deep Residual Network, ResNet) - 19
3 以雲層資訊利用機器學習進行降雨預測 - 20
3.1 雲層圖像與天氣資訊蒐集 - 21
3.2 雲層資訊處理 - 23
3.3 天氣資訊處理 - 24
3.4 模型編碼器與解碼器 - 25
4 實驗結果與分析 - 27
4.1 雲層圖片& 氣象資料 - 27
4.2 模型訓練 - 29
4.3 模型訓練結果 - 30
4.4 模型預測結果 - 30
5 結論與未來展望 - 35
參考文獻 - 36

[1] NOAA SciJinks, “How Reliable Are Weather Forecasts?.” Available: https://scijinks.
gov/forecast-reliability/ [Accessed: 2019-06-26].
[2] G. E. Hinton, “Deep Belief Networks,” Scholarpedia, vol. 4, no. 5, p. 5947, 2009.
[3] I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. C. Courville,
and Y. Bengio, “Generative Adversarial Nets,” 27th International Conference Advances in Neural
Information Processing Systems: Annual Conference on Neural Information Processing Systems,
pp. 2672–2680, 2014.
[4] C. J. C. H. Watkins and P. Dayan, “Technical Note Q-Learning,” Machine Learning, vol. 8,
pp. 279–292, 1992.
[5] W. K. Hastings, “Monte Carlo Sampling Methods Using Markov Chains and their Applications,”
Biometrika, vol. 57, pp. 97–109, 04 1970.
[6] S. J. Pan and Q. Yang, “A Survey on Transfer Learning,” IEEE Trans. Knowl. Data Eng., vol. 22,
no. 10, pp. 1345–1359, 2010.
[7] M. Wang and W. Deng, “Deep Face Recognition: A Survey,” Computing Research Repository,
vol. abs/1804.06655, 2018.
[8] C. Szegedy, S. Ioffe, V. Vanhoucke, and A. A. Alemi, “Inception-v4, Inception-ResNet and
the Impact of Residual Connections on Learning,” Thirty-First AAAI Conference on Artificial
Intelligence, pp. 4278–4284, 2017.
[9] A. Sehgal and N. Kehtarnavaz, “A Convolutional Neural Network Smartphone App for Real-
Time Voice Activity Detection,” IEEE Access, vol. 6, pp. 9017–9026, 2018.
[10] A. Nagrani, S. Albanie, and A. Zisserman, “Seeing Voices and Hearing Faces: Cross-Modal
Biometric Matching,” IEEE Conference on Computer Vision and Pattern Recognition, pp. 8427–
8436, 2018.
[11] H. H. Pajouh, A. Dehghantanha, R. Khayami, and K. R. Choo, “A Deep Recurrent Neural
Network Based Approach for Internet of Things Malware Threat Hunting,” Future Generation
Comp. Syst., vol. 85, pp. 88–96, 2018.
[12] S. Li, W. Li, C. Cook, C. Zhu, and Y. Gao, “Independently Recurrent Neural Network (IndRNN):
Building A Longer and Deeper RNN,” Computing Research Repository, vol. abs/1803.04831,
2018.
[13] M. Inoue, S. Inoue, and T. Nishida, “Deep Recurrent Neural Network for Mobile Human Activity
Recognition with High Throughput,” Artificial Life and Robotics, vol. 23, pp. 173–185, Jun
2018.
[14] K. Li, H. Xu, Y. Wang, D. Povey, and S. Khudanpur, “Recurrent Neural Network Language
Model Adaptation for Conversational Speech Recognition,” 19th Annual Conference of the International
Speech Communication Association, pp. 3373–3377, 2018.
[15] J. Su, S. Wu, D. Xiong, Y. Lu, X. Han, and B. Zhang, “Variational Recurrent Neural Machine
Translation,” Thirty-Second AAAI Conference on Artificial Intelligence, pp. 5488–5495, 2018.
[16] Y. Zhang, R. Xiong, H. He, and M. G. Pecht, “Long Short-Term Memory Recurrent Neural
Network for Remaining Useful Life Prediction of Lithium-Ion Batteries,” IEEE Trans. Vehicular
Technology, vol. 67, no. 7, pp. 5695–5705, 2018.
[17] S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol. 9,
no. 8, pp. 1735–1780, 1997.
[18] F. A. Gers, J. Schmidhuber, and F. A. Cummins, “Learning to Forget: Continual Prediction with
LSTM,” Neural Computation, vol. 12, no. 10, pp. 2451–2471, 2000.
[19] I. H. Witten, E. Frank, M. A. Hall, and C. J. Pal, Data Mining: Practical Machine Learning
Tools and Techniques. Morgan Kaufmann, 2016.
[20] M. J. Zaki and W. M. Jr., Data Mining and Analysis: Fundamental Concepts and Algorithms.
Cambridge University Press, 2014.
[21] A. Kazantzidis, P. Tzoumanikas, A. Bais, S. Fotopoulos, and G. Economou, “Cloud Detection
and Classification with the Use of Whole-sky Ground-based Images,” Atmospheric Research,
vol. 113, pp. 80–88, 2012.
[22] O. Vinyals, A. Toshev, S. Bengio, and D. Erhan, “Show and Tell: A Neural Image Caption
Generator,” IEEE Conference on Computer Vision and Pattern Recognition, pp. 3156–3164,
2015.
[23] K. Xu, J. Ba, R. Kiros, K. Cho, A. C. Courville, R. Salakhutdinov, R. S. Zemel, and Y. Bengio,
“Show, Attend and Tell: Neural Image Caption Generation with Visual Attention,” 32nd
International Conference on Machine Learning, vol. 37, pp. 2048–2057, 2015.
[24] A. Agrawal, J. Lu, S. Antol, M. Mitchell, C. L. Zitnick, D. Parikh, and D. Batra, “VQA: Visual
Question Answering,” International Journal of Computer Vision, vol. 123, no. 1, pp. 4–31, 2017.
[25] J. Donahue, L. A. Hendricks, M. Rohrbach, S. Venugopalan, S. Guadarrama, K. Saenko, and
T. Darrell, “Long-Term Recurrent Convolutional Networks for Visual Recognition and Description,”
IEEE Trans. Pattern Anal. Mach. Intell., vol. 39, no. 4, pp. 677–691, 2017.
[26] K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” in IEEE
Conference on Computer Vision and Pattern Recognition, pp. 770–778, IEEE Computer Society,
2016.