基于BNN-IKAE双层多智能体深度强化学习的大电网母线电压智能自动调整方法  

Intelligent automatic adjustment method of bus voltage in large power grid based on BNN-IKAE double-layer multi-agent deep reinforcement learning

作　　者：陈东旭 李岩松[1] 许智光 陈兴雷 陈胜硕 刘君[1] 康世佳 CHEN Dongxu;LI Yansong;XU Zhiguang;CHEN Xinglei;CHEN Shengshuo;LIU Jun;KANG Shijia(School of Electrical and Electronic Engineering,North China Electric Power University,Beijing 102206,China;China Electric Power Research Institute Co.,Ltd.,Beijing 100192,China)

机构地区：[1]华北电力大学电气与电子工程学院,北京102206 [2]中国电力科学研究院有限公司,北京100192

出　　处：《电工电能新技术》2024年第11期68-79,共12页Advanced Technology of Electrical Engineering and Energy

基　　金：国家自然科学基金项目(U1866602);国家重点研发计划项目(2021YFB2400805);国家电网有限公司总部管理科技项目(5100-202355409A-3-2-ZN)。

摘　　要：母线电压调整是保证电能质量和电力系统安全稳定运行的重要措施,但目前自动电压控制系统的数据处理及分析效率和正确性仍难以满足愈加复杂的大型电网要求,据此本文提出了一种二值神经网络知识经验融合(BNN-IKAE)双层多智能体深度强化学习算法,在深度强化学习的基础上进行大电网母线电压调整。本文首先介绍了常规的调整流程,并据此搭建了马尔科夫决策过程(MDP)模型;然后针对大电网可调元件参数复杂的问题设计了双层多智能体结构,通过引入二值神经网络(BNN)降低了网络复杂度,解决了模型计算速度慢的问题,并融合了专家经验的知识经验融合(IKAE)模块,通过专家经验池和存量判定机制提高了模型的收敛性和奖励值。最后,在东北地区电网中对提出的基于BNN-IKAE的双层多智能体深度强化学习模型的母线电压调整能力进行了仿真验证,与常规方法相比其调整时间减少了79.331%,调整的成功率增加了19.23%,结果表明基于BNN-IKAE双层多智能体深度强化学习的大电网母线电压智能自动调整方法能够提高计算速度和成功率。Bus voltage adjustment is an important measure to ensure power quality and safe and stable operation of power systems.However,the data processing and analysis efficiency and correctness of automatic voltage control systems are still difficult to meet the increasingly complex requirements of large power grids.Based on this,this paper proposes a binary neural network knowledge experience fusion(BNN-IKAE)double-layer multi-agent deep reinforcement learning algorithm,On the basis of deep reinforcement learning,the bus voltage of the power grid is adjusted.This article first introduces the conventional adjustment process and builds a Markov Decision Process(MDP)model based on it;Then,a double-layer multi-agent structure is designed to address the complex parameters of adjustable components in the large power grid.By introducing a binary neural network(BNN),the network complexity is reduced and the problem of slow model calculation speed can be solved.The knowledge and experience fusion(IKAE)module based on expert experience is integrated,and the convergence and reward value of the model are improved through expert experience pool and stock judgment mechanism.Finally,the bus voltage adjustment ability of the proposed BNN-IKAE based double-layer multi-agent deep reinforcement learning model is simulated and verified in the Northeast power grid.Compared with conventional methods,the adjustment time can be reduced by 79.331%,and the success rate of adjustment increased by 19.23%.The results showed that the intelligent automatic adjustment method for bus voltage in the large power grid based on BNN-IKAE double-layer multi-agent deep reinforcement learning can improve calculation speed and success rate.

关 键 词：深度强化学习 母线电压 神经网络 越限调整 专家经验 

分 类 号：TM761[电气工程—电力系统及自动化]