基于深度Q学习的强鲁棒性智能发电控制器设计  

Design of strong robust smart generation controller based on deep Q learning

作　　者：殷林飞[1] 余涛[1] YIN Linfei;YU Tao(School of Electric Power,South China University of Technology, Guangzhou 510640, China)

机构地区：[1]华南理工大学电力学院,广东广州510640

出　　处：《电力自动化设备》2018年第5期12-19,共8页Electric Power Automation Equipment

基　　金：国家重点基础研究发展计划(973计划)项目(2013CB228205);国家自然科学基金资助项目(51477055)~~

摘　　要：在现代互联大电网背景下,研究了多区域强鲁棒性的智能发电控制策略。在Q学习的架构下,将深度神经网络的预测机制作为强化学习的动作选择机制,提出了一种具有强鲁棒性的深度Q学习算法,设计了基于该算法的智能发电控制器。针对智能电网下的智能发电控制问题,在多智能体系统的框架下采用所提深度Q学习算法进行控制,并与传统的PID、Q学习和Q(λ)算法进行对比。在IEEE标准2区域和以南方电网4区域为背景的仿真模型(采用了23 328种不同模型参数)中进行数值仿真,仿真结果验证了所提深度Q学习算法的可行性和有效性,也验证了所设计控制器的强鲁棒性。Under the background of modern interconnected large power grid,a smart generation control strategy wltn strong robustness in multi-areas is studied. In the framework of Q learning, taking the prediction mechanism of the deep neural network as the action selector of Q learning, a DQL（ Deep Q Learning） algorithm with strong robustness is proposed, and on this basis,a smart generation controller is designed. The proposed DQL algorithm in the multi- agent system is applied for smart generation control in the smart interconnected power grid, and is compared with the traditional PID algorithm, Q learning algorithm and Q （λ） learning algorithm. The simulative results of IEEE standard two-area model and the four-area model based on China Southern Power Grid with 23 328 different parame- ters verify the feasibility and effectiveness of the proposed DQL algorithm and the strong robustness of the designed controller.

关 键 词：深度Q学习 智能发电控制 强鲁棒性 深度神经网络 多智能体系统 

分 类 号：TM761.2[电气工程—电力系统及自动化]