基于ADP的复合燃料电池发电系统控制策略

摘要

An approximate dynamic programming (ADP) method is used for fuel cell energy management based on studying on structure of hybrid fuel cell/supercapacitor.The principle and formation of action depended heuristic dynamic programming (ADHDP) are introduced in detail.The action network,which is used to generate control signal constituted by fuzzy-neural network instead of conventional multiple layer perceptron neural network,makes physical interpretation between input and output more clear.The layout of increment output is more satisfied by strong dynamic constraint condition caused by limits of fuel utilization.The structure and parameters adjusting method of critic network and fuzzy-neural network are described in this paper in terms of state and control variables of the system.Finally,the simulation verifies the correctness of the proposed control strategy and its good performance on load following.The fuel utilization is usual in the constraint range even in transient state that ensures the safety and efficiency of the system.Compared with conventional PI regulation,the ADP method has characteristics of fast response and small overshoot.%在研究燃料电池/超级电容器复合电源结构基础上,引入了一种基于近似动态规划(Approximate Dynamic Programming,ADP)的方法,用于燃料电池能量管理.详细介绍了ADP方法中执行依赖的启发式动态规划(Action-Depended Heuristic Dynamic Programming,ADHDP)的原理及构成.采用模糊神经网络产生控制信号,取代了传统由多层感知器所构成的执行网络,使得输入与输出之间的物理意义更为明确；采用增量输出的方式更适用于由于燃料利用率上下限所造成的强动态约束环境.针对复合发电系统的状态变量与控制信号,本文详尽阐述了评价网络与模糊神经网络的结构及其参数的调整方法.最后,仿真分析验证了控制策略的正确性以及良好的负载跟踪特性,并且在暂态过程中,燃料利用率仍在约束的范围内,保证了燃料电池的稳定运行与较高的发电效率.与传统的PI调节方式比较,本文采用的ADP方法具有响应快、超调小的特点.