(1. 重慶大學 資源與安全學院，煤礦災害動力學與控制國家重點實驗室，重慶 400044；
2. 博眉啟明星鋁業(yè)有限公司，眉山 620010；
3. 貴陽鋁鎂設計研究院有限公司，貴陽 550000；
4. 四川省四維環(huán)保設備有限公司，遂寧 629000)
摘 要: 本文對鋁電解槽陽極效應機理和故障參數進行研究，提出了一種基于深度學習的陽極效應預測方法，能適應不同維度、不同數據特征的槽況參數，直接從海量原始數據中挖掘故障特征信息，大幅縮減效應響應時間，具有很好的魯棒性和抗干擾能力，同時在模型調試優(yōu)化上，采用Batch normalization算法和梯度檢驗，提高了模型收斂速度和穩(wěn)定性。結果表明：該模型效應預測準確率和F1分數分別達到94.65%和0.9317，提前預報時間可達16 min，并通過現(xiàn)場實驗驗證，達到實際生產要求。
關鍵字: 鋁電解；300 kA；陽極效應預測；深度學習；算法優(yōu)化
（1. State Key Laboratory of Coal Mine Disaster Dynamics and Control, College of Resource and Safety Engineering, Chongqing University, Chongqing 400044, China;
2. Bomei Qimingxing Aluminum Co., Ltd., Meishan 620010, China;
3. Guiyang Aluminum Magnesium Design & Research Institute Co., Ltd., Guiyang 550000, China;
4. Sichuan Siwei Environmental Protection Equipment Co., Ltd., Suining 629000, China）
Abstract:The anode effect mechanism and fault parameters of aluminium electrolytic cells were studied, and a deep learning-based anode effect prediction method was proposed. It can adapt to the parameters of tank conditions in different dimensions and different data characteristics, and directly mine fault characteristic information from massive raw data. It greatly reduces the response time of the effect, has good robustness and anti-interference ability. At the same time, in the model debugging optimization, the Batch Normalization algorithm and gradient test are used to improve the model convergence speed and stability. The prediction accuracy and F1 score reach 94.65% and 0.9317, respectively. The prediction time can reach 16 min, and it is verified by field experiments to meet the actual production requirements.
Key words: aluminium electrolysis; 300 kA; anode effect prediction; deep learning; optimization
