基于人工智能算法的铝电解工艺优化求解
2022-06-21赵春红
赵春红



摘要:为提高铝电解工艺参数优化能力,基于近端策略优化(PPO)深度强化学习算法,构建铝电解工艺参数优化模型,以实现铝电解工艺优化求解。对铝电解工艺流程进行介绍,确定其工艺中的重要技术参数;然后将深度与强化学习相结合,建立铝电解参数寻优模型。在原始 PPO 算法基础上加入RMSProp 算法和冲量思想得到M-RMSProp-PPO 算法,将该算法应用到模型中进行铝电解工艺参数寻优。结果表明:改进后算法可实现快速收敛,累计奖赏高达15%,说明改进后的算法性能更优越;与决策树算法进行搭配后,可以得到铝电解工艺优化最优组合。
关键词:铝电解;工艺优化;深度强化学习;M-RMSProp-PPO
中图分类号:TP392
文献标识码:A文章编号:1001-5922(2022)06-0186-05
Optimization of aluminum electrolysis technology based on artificial intelligence algorithm
ZHAO Chunhong
(Sichuan Vocational College of Culture & Communication, Chongzhou 611200, Sichuan China
)
Abstract:In order to improve the parameter optimization ability of the aluminum electrolysis process, the aluminum electrolytic process parameter optimization model is constructed based on the proximal strategy optimization (PPO) deep reinforcement learning algorithm to realize the optimization and solution of the aluminum electrolysis process. First, the aluminum electrolysis process is introduced to determine the important technical parameters in the process; then combine depth and reinforcement learning to establish the aluminum electrolytic parameter optimization model; add RMSProp algorithm and impulse idea based on the original PPO algorithm to obtain M-RMSProp-PPO algorithm and apply it to the model to optimize the aluminum electrolytic process parameters. The experiment show that the improved algorithm can achieve rapid convergence with a cumulative reward of 15%, indicating that the improved algorithm is superior; after matching with the decision tree algorithm, the optimal combination of aluminum electrolysis process could be obtained.
Key words:aluminum electrolysis; process optimization; deep reinforcement learning; M-RMSProp-PPO
我國制造业的不断发展,铝电解的生产需求呈爆炸式增长,将铝与多种金属材料组合后可以得到新的合成金,而合成金广泛应用于工业领域、发动机和电动机等,其具有较好的发展前景。然而,传统的铝电解生产主要受工艺人员经验限制,工艺优化方法主观性太强,从而导致工艺优化效率低、稳定性差,
不能达到当前工业领域的高效铝电解生产需求。因此,提出一种新的工艺优化方法对当前铝电解运用具有重要的研究意义。有学者提出将BP-ANN-MCSA算法应用到电弧增材工艺中,进行参数寻优,其利用BP神经网络的深层学习特性,实现增材工艺参数寻优,且寻优速率提升了10%,具有一定的可行性[1];在近端策略优化算法PPO的基础上,加入RMSProp,并构建一个……
