遞歸神經網路之冰水機故障預測

Chiller Fault Prediction Using Recurrent Neural Networking Method

林于翔(Yu-Hsiang Lin)

指導教授 ： 陳俊勳

國立交通大學/工學院/機械工程系所/碩士(2018年)

摘要

本論文應用深度學習演算法於冰水機的故障模式進行分類及其嚴重程度預測，所使用的數據來自於美國冷凍空調學會(ASHRAE)的實驗計畫 [3, 4]。此研究採用遞歸神經網路(Recursive Neural Network, RNN)方法做為深度學習演算法，以序列的方式建立模型，以監督式學習(Supervised learning)訓練模型；也同時進行RNN、LSTM(Long Short-Term Memory)及GRU(Gated Recurrent Unit)三種神經網路架構的比較。由於神經網路模型的建置必須人為設定一些超參數(Hyperparameters)，為了使模型最佳化，本論文建立12種情境進行交叉驗證(Cross Validation)，根據驗證結果選擇適合超參數及架構來建置模型。在故障模式八類別的分類下，測試資料集預測準確度可達99%以上，在嚴重程度迴歸問題下的MAPE(Mean Absolute Percentage Error)也都在2.5%以下，表示本文所發展出的模型之預測性能極佳，可適用於時間序列的數據分析。

關鍵字

冰水機 ； 遞歸神經網路 ； LSTM ； GRU ； tensorflow

並列摘要

This paper uses the deep learning algorithm to classify the fault mode of the chiller and predicts it's faulty level. The data used in this paper are from ASHRAE experimental program[3, 4]. This paper uses a deep learning model, recursive neural network(RNN), to sequentially build a model that uses supervised learning for training. At the same time, the architecture of RNN, LSTM(Long Short-Term Memory) and GRU(Gated Recurrent Unit) was compared. Since the neural network model must be artificially configured with some hyperparameters, 12 scenarios were established for cross validation, and the suitable hyperparameters、architecture were selected for the model based on verification results. Under the eight-category classification of failure modes, the accuracy of test data set prediction can reach more than 99%, and the severity’s MAPE (Mean Absolute Percentage Error) is also below 2.5%. It indicates that the model developed in this paper has excellent predictive performance and can be applied to time series data analysis.

並列關鍵字

Chiller ； RNN ； LSTM ； GRU ； tensorflow

參考文獻

1. J. Manyika, M. Chui, B. Brown, J. Bughin, R. Dobbs, C. Roxburgh, and A. H. Byers, “Big data: The next frontier for innovation, competition, and productivity”, McKinsey Global Institute.(2011)

2. K. Holmberg, E. Jantunen, A. Adgar, J. Mascolo, A. Arnaiz, and S. Mekid, “Information and Communication Technologies Within E-maintenance”, spring science & business media.(2010)

3. M. C. Comstock and J. E. Braun, “Development of analysis tools for the evaluation of fault detection and diagnostics for chillers”, ASHRAE Research Project 1043 HL 99e20, Report # 4036-3. (1999)

4. M. C. Comstock and J. E. Braun, “Experimental data from fault detection and diagnostic studies on a centrifugal chiller”, ASHRAE Research Project 1043 HL 99e18, Report # 4036-1. (1999)

5. Xuewu Dai and Zhiwei Gao, “From model，signal to knowledge: A data-driven perspective of fault detection and diagnosis”, IEEE Transactions on industrial informatics, pp.2226-2238.(2013)