在工業鐵生產中高爐非常重要。它是鋼鐵持續生產的核心。高爐的使用壽命會直接反映在煉鋼的總成本上。它受多種原因影響，如冷卻板和高爐碳磚的厚度。其中碳磚厚度是最關鍵的。近十年來，中鋼4號高爐碳磚的侵蝕速度異常的快。這將大大縮短 4 號高爐的使用壽命。碳磚的厚度是根據爐床溫度經由特殊公式計算得出。當爐床溫度達到歷史高溫度時，意味著碳磚變薄了。在不影響鋼鐵產量的情況下控制高爐溫度並不容易。如果可以提前知道溫度趨勢。它將能夠隨著即將到來的情況改變操作方法。因此，本研究將提出機器學習模型來預測研究中的爐床溫度。
在本研究中，主成分分析（PCA）和皮爾森相關係數用於驗證數據集的特徵數目。 PCA 可以顯示不同特徵數量數據的複雜性。此外，可以根據皮爾森係數刪除數據集中的高度相關的特徵。
為了證明預測爐膛溫度的可能性。使用隨機森林訓練一個溫度分類器，可以對爐床溫度進行分類。分類器在測試集中達到 95% 的準確率。最後，基於LSTM建立了一個爐床溫度模型，利用高爐運行數據預測爐膛溫度。此外，為了優化溫度預測模型，本研究將提出一個新的訓練方法。此外，找出輸入和輸出長度的最佳組合，以減少模型的損失。該模型的 MSE 為 0.00939，而未經訓練以平均溫度計算的基準線為 0.0615。

The blast furnace plays a very important role in industrial iron production. It is the heart of the consistent production of steel. The service life of the blast furnace is directly reflecting the total cost of steelmaking. It is affected by many reasons, like the cooling plate and the thickness of carbon bricks in the blast furnace. Among the reason carbon brick is the most critical ones. In the past decade, the carbon brick in blast furnace No.4 of China Steel Corporation has been eroded at an abnormal rate. It would dramatically reduce the life span of blast furnace No. 4. The thickness of the carbon bricks is calculated with a custom formula from the hearth temperature. When the hearth temperature hit a historical highest temperature means the carbon brick became thinner. It is not easy to control the temperature of blast furnace without affecting the iron production. If the temperature trend could be known in advance. It would be able to change the operating method with the upcoming situation. Therefore, a machine learning model would be proposed to predict the hearth temperature in the study.
In this study, principal components analysis (PCA) and Pearson correlation coefficient are used to determine the feature number of the dataset. PCA could show the complexity of data with the different number of features. Furthermore, the high correlation features in the dataset could be deleted based on the Pearson coefficient.
To demonstrate the possibility of predicting hearth temperature. A temperature classifier is trained with random forest which can classify the hearth temperature. The classifier achieves 95% accuracy in the test set. At last, a hearth temperature model was developed by LSTM which uses the operation data of blast furnace to predict the hearth temperature. In addition, to optimize the temperature prediction model, a different approach in the training process will be compared with common method in this study. Also, find out the best combination of input and output lengths to reduce the loss of the model. The MSE of the model is 0.00939 and the baseline which calculate with average temperature without training is 0.0615.

[1] T. Yang, "Ironmaking production trend in China: High efficiency," Energy conservation, Environment protection and low cost. Ironmaking, vol. 33, pp. 1-11, 2014.
[2] F. Neumann, H. Schenck, and W. Patterson, "Influence of iron companions on carbon solubility, carbon activity and saturation point in cast iron," Giesserei, vol. 47, pp. 25-32, 1960.
[3] K.-X. Jiao, J.-L. Zhang, Z.-J. Liu, C.-L. Chen, and Y.-X. Liu, "Analysis of blast furnace hearth sidewall erosion and protective layer formation," ISIJ International, vol. 56, no. 11, pp. 1956-1963, 2016.
[4] Z. Guo, J. Zhang, K. Jiao, Y. Zong, and Z. Wang, "Occurrence state and behavior of carbon brick brittle in a large dissected blast furnace hearth," steel research international, vol. 92, no. 11, p. 2100273, 2021.
[5] N. Pai and V. Ilango, "Lstm neural network model with feature selection for financial time series prediction," in 2020 Fourth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(I-SMAC), 2020: IEEE, pp. 672-677.
[6] J. Cai, J. Luo, S. Wang, and S. Yang, "Feature selection in machine learning: A new perspective," Neurocomputing, vol. 300, pp. 70-79, 2018.
[7] S. Khalid, T. Khalil, and S. Nasreen, "A survey of feature selection and feature extraction techniques in machine learning," in 2014 science and information conference, 2014: IEEE, pp. 372-378.
[8] J. Benesty, J. Chen, Y. Huang, and I. Cohen, "Pearson correlation coefficient," in Noise reduction in speech processing: Springer, 2009, pp. 1-4.
[9] H. F. Eid, A. E. Hassanien, T.-h. Kim, and S. Banerjee, "Linear correlation-based feature selection for network intrusion detection model," in International Conference on Security of Information and Communication Networks, 2013: Springer, pp. 240-248.
[10] Y. Liu, Y. Mu, K. Chen, Y. Li, and J. Guo, "Daily activity feature selection in smart homes based on pearson correlation coefficient," Neural Processing Letters, vol. 51, no. 2, pp. 1771-1787, 2020.
[11] A. J. Myles, R. N. Feudale, Y. Liu, N. A. Woody, and S. D. Brown, "An introduction to decision tree modeling," Journal of Chemometrics: A Journal of the Chemometrics Society, vol. 18, no. 6, pp. 275-285, 2004.
[12] L. Breiman, "Random forests," Machine learning, vol. 45, no. 1, pp. 5-32, 2001.
[13] S. Hochreiter and J. Schmidhuber, "Long Short-Term Memory," Neural Computation, vol. 9, no. 8, pp. 1735-1780, 1997, doi: 10.1162/neco.1997.9.8.1735.
[14] N. Ketkar, "Introduction to keras," in Deep learning with Python: Springer, 2017, pp. 97-111.