分批补料发酵过程多目标优化的分布式强化学习策略  

Distributed reinforcement learning strategy for multi-objective optimization of fed-batch fermentation process

作　　者：李大字[1] 宋天恒[1] 靳其兵[1] 谭天伟[2] 

机构地区：[1]北京化工大学信息科学与技术学院 [2]北京化工大学生命科学与技术学院,北京100029

出　　处：《化工学报》2011年第8期2243-2247,共5页CIESC Journal

基　　金：国家重点基础研究发展计划项目(2007CB714300);国家高技术研究发展计划项目(2008AA04Z131);北京市优秀人才资助项目~~

摘　　要：发酵过程优化问题通常包含有互相冲突的多重优化目标,另外反应本身具有诸多复杂性。提出一种基于Pareto的分布式Q学习多目标策略,用以求解赖氨酸分批补料发酵过程流加速率轨迹的Pareto最优解。该策略中,Q学习算法和Pareto排序法将结合来产生非支配解集,并使之逼近真实的Pareto前沿,利用奖赏机制来描述多重目标之间的关系,并同时使用多组含有随机初始值的agent共同作用改善搜索能力。将所提出的方法应用于赖氨酸分批补料发酵过程的优化中,并与粒子群优化进行了对比,验证策略的性能。Fermentation processes optimization involves multiple and conflicting objectives and the features of these processes contain many complexities.In this paper,a design of a Pareto-based distributed Q-learning（PDQL）optimization strategy was presented to solve Pareto optimal flow rate trajectories for the lysine fed-batch fermentation process.Q-learning algorithm and Pareto sorting method were combined to generate the nondominated solution set and to make it approximate the actual Pareto front.The strategy described the relation of multi-objectives with the help of rewards strategy.For enhancing searching capability,multiple randomly initialized groups of agents were used.The result of PDQL optimization was compared to PSO with the aggregated function method to test its performance.

关 键 词：Q学习算法 多目标优化 赖氨酸分批补料发酵 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]