摘要
为提高复杂场景下的青霉素产量,提出一种基于改进TD3算法的发酵控制模型Res-OD-TD3,以实现对青霉素发酵过程的优化控制。首先,将残差网络融合到Actor和Critic网络中,用于提升其收敛性;其次,使用OU噪声代替传统高斯噪声,使动作值能够在连续控制任务中以更平滑的方式进行探索,减少探索过程中的剧烈波动;最后,通过动态调整的Huber损失函数对原有损失函数进行改进,在训练的不同阶段提供更合适的损失度量,使算法在处理异常值时更为鲁棒。仿真实验结果表明,与传统的TD3算法和DDPG算法相比,Res-OD-TD3算法对于温度精度和稳定性的控制效果更好,溶解氧浓度分别提升了4.1%、5.5%,产量分别提升了12.8%、27.8%,证实了其在青霉素发酵优化控制中的显著优势。
To improve the production of penicillin in complex scenarios,a fermentation control model Res-OD-TD3 based on an improved TD3 algorithm is proposed to achieve optimized control of the penicillin fermentation process.Firstly,the residual network is integrated into the Actor and Critic networks to enhance the convergence of the network;Secondly,using OU noise instead of traditional Gaussian noise allows for smoother exploration of action values in continuous control tasks,reducing drastic fluctuations during exploration;Finally,by dynamically adjusting the Huber loss function,the original loss function is improved to provide more suitable loss metrics at different stages of training,making the algorithm more robust in handling outliers.The simulation experiment results show that compared with traditional TD3 algorithm and DDPG algorithm,Res-OD-TD3 algorithm has better control effect on temperature accuracy and stability,with dissolved oxygen concentration increased by 4.1%and 5.5%respectively,and yield increased by 12.8%and 27.8%respectively,confirming its significant advantages in optimizing penicillin fermentation control.
作者
王晓君
徐涛
赵春丽
彭亮亮
杜越
WANG Xiaojun;XU Tao;ZHAO Chunli;PENG Liangliang;DU Yue(School of Information Science and Engineering,Hebei University of Science and Technology,Shijiazhuang 050000,China;Shanxi National Pharmaceutical Group Weiqida Pharmaceutical Co.,Ltd.,Datong 037000,China)
出处
《软件导刊》
2025年第10期104-110,共7页
Software Guide
关键词
深度强化学习
控制策略
TD3算法
残差网络
青霉素发酵
deep reinforcement learning
control strategy
TD3 algorithm
residual network
penicillin fermentation
