基于LSTM的铝电解槽生产决策算法研究与应用

Research and application of decision-making algorithm for aluminum reduction cell production based on LSTM

在铝电解槽生产过程中,需要工艺技术人员依据历史生产数据每日进行生产决策,其中出铝量和设定电压调整量是两个比较重要且耦合性强的决策参数,决策的好坏直接影响着电解槽的稳定运行和高效产出铝。本文提出了一种采用海洋捕食者算法来优化双遗忘门LSTM结构模型的决策算法(MPA-dfLSTM),针对数据特点对LSTM结构优化为双遗忘门LSTM(dfLSTM),并通过海洋捕食者算法来进行参数寻优,将训练得到的模型应用到出铝量和设定电压调整量的决策中。MPA-dfLSTM模型相比于其他三种决策模型获得了较高的性能指标,可以为铝电解生产工艺技术人员提供生产决策参考。

In the production process of aluminum reduction cell,process technicians need to make daily production decisions based on historical production data,in which the output of aluminum and the amount of set voltage adjustment are two more important and coupled decision-making parameters,and the quality of decision-making directly affects the stable operation of the aluminum reduction cells as well as the efficient production of aluminum.In this paper,a decision algorithm(MPA-dfLSTM)using the marine predator algorithm to optimize the double forgotten gate LSTM structure model is proposed,the LSTM structure is optimized to the double forgotten gate LSTM(dfLSTM)according to the data characteristics,and the parameter optimization is carried out by the marine predator algorithm,and the trained model is applied to the decision on the output of aluminum and the amount of voltage adjustment.Compared with the other three decision models,the MPA-dfLSTM model obtains higher performance indicators,which can provide production decision reference for aluminum production process technicians.