基于LSTM的铝电解槽况预测

Prediction of aluminum pot conditions based on LSTM

铝电解是一个非线性、多变量耦合和大延迟的工业过程体系。在铝电解正常的生产过程中,铝电解槽长期处于高温条件下,因此加大了对铝电解槽况研究的难度。近几年,计算机控制和传感技术在铝电解生产过程中也有了非常广泛的应用,在整个铝电解生产过程中会产生大量的生产数据,旨在找出潜藏在这些数据中的有用信息,分析其与槽况变化的联系及进一步发现槽况变化的规律,为专业人员准确判断和预测其槽况提供重要依据。LSTM算法继承了RNN的特性,从经验中学习来实现预测,不仅学习长期的依赖性还解决了梯度消失问题,所以利用LSTM算法对槽况进行预测,便于及时发现槽况变化,做出相应举措,从而减少企业损失。

Aluminum electrolysis is a non-linear, multi-variable coupling and large delay industrial process system. In the normal production process of aluminum electrolysis, the aluminum pot is exposed to high temperature for a long time, so it is more difficult to study the aluminum pot conditions. Recent years have witnessed the extensive application of computer control and sensing technology to the aluminum electrolysis process. A large amount of production data is generated during the entire aluminum electrolysis process, from which the useful information will be identified to analyze its connection with the change of pot conditions and further discover the change rules of pot conditions. The accurate judgment and prediction of the pot conditions is an important basis for the professional decision. LSTM algorithm inherits the characteristics of RNN, which learns from experience to realize the prediction, not only learns the long-term dependence, but also solves the problem of gradient disappearance. LSTM algorithm is used to predict the pot conditions, which makes it easy to find changes of pot conditions in time and take corresponding measures to reduce corporate losses.