基于图神经网络的醇沉过程智能控制方法OA
Intelligent Control Method for Alcohol Precipitation Processes Based on Graph Neural Networks
醇沉过程广泛应用于生物制药和生物工程中,尤其是在提取蛋白质和多糖等生物大分子时表现出一定优势.针对传统优化方法难以处理其非线性和变量间耦合问题,提出了一种结合优先经验回放(PER)和图卷积网络(GCN)的多智能体深度确定性策略梯度算法(MADDPG).算法在价值网络中引入 GCN结构捕捉智能体之间的协作关系,并通过PER机制优先采样关键经验,以加快策略学习并提升训练效率.实验结果表明,所提算法在醇沉过程控制任务中,相较于现有部分方法具有更快的收敛速度和更优的策略表现,验证了其在复杂多参数优化中的优势.
The alcohol precipitation process is widely used in biopharmaceuticals and bioengineering,especially in the extraction of biomacromolecules such as proteins and polysaccharides.In view of the fact that traditional optimization methods are difficult to deal with its nonlinearities and inter-variable coupling,a multi-agent deep deterministic policy gradient(MADDPG)algorithm combining prioritized experience replay(PER)and graph convolutional network(GCN)is proposed.The algorithm introduces the GCN structure in the value network to capture the collaborative relationship between agents,and prioritizes the sampling of key experiences through the PER mechanism to accelerate policy learning and improve training efficiency.Experimental results show that the proposed algorithm has faster convergence speed and better policy performance in the alcohol precipitation process control task than some existing methods,verifying its advantages in complex multi-parameter optimization.
郭鸿斌;王艳;刘相;王团结;戴翠红
江南大学物联网工程学院,江苏 无锡 214122江南大学物联网工程学院,江苏 无锡 214122江南大学物联网工程学院,江苏 无锡 214122江苏康缘药业股份有限公司,江苏 连云港 222043江苏康缘药业股份有限公司,江苏 连云港 222043
信息技术与安全科学
醇沉多智能体系统优先经验回放图卷积网络深度强化学习
alcohol precipitationmulti-agent systemprioritized experience replaygraph convolutional networkdeep reinforcement learning
《信息与控制》 2026 (4)
665-677,13
长三角科技创新共同体联合攻关计划项目(2023CSJGG1700)
评论