基于数据驱动和深度强化学习的孤岛多风柴电压控制研究  

Study on Voltage Control of Isolated Island Multi-winds Diesel Hybrid System Based on Data-driven and Deep Reinforcement Learning Method

作　　者：贺柱 柴良明 吴楚 郑仕涛 HE Zhu;CHAI Liangming;WU Chu;ZHENG Shitao(Ruili Power Supply Bureau of Yunnan Power Grid Co.,Ltd.,Yunnan Ruili 678600,China;Yunnan Power Grid Co.,Ltd.,Dehong Power Supply Bureau,Yunnan Dehong 678400,China)

机构地区：[1]云南电网有限责任公司瑞丽供电局,云南瑞丽678600 [2]云南电网有限责任公司德宏供电局,云南德宏678400

出　　处：《电力电容器与无功补偿》2021年第4期1-7,共7页Power Capacitor & Reactive Power Compensation

基　　金：云南电网有限责任公司科技项目(YNKJXM20170842)。

摘　　要：针对风能及负荷波动导致的孤岛多风柴系统电压波动问题,提出用双馈感应发电机(double fed induction generator,DFIG)负荷侧变流器(load side converter,LSC)补偿系统无功的策略。考虑LSC状态量不完全可测及模型参数不确定的情况,本文基于连续性动作空间的深度强化学习算法提出一种数据驱动、无模型的LSC附加控制方法。该方法以非线性的方式获取控制策略来减小电压波动,并且针对模型的不确定性具有更快的响应速度和更强的自适应性。首先建立孤岛多风-柴混合系统的数学模型;其次离线采用深度强化学习在连续性空间中优化LSC的无功电压控制策略,然后在线通过堆叠去噪自动编码器(stacked denoising auto-encoder,SDAE)提取特征,并采用策略网络进行在线控制。数例仿真结果表明,与传统的控制策略相比,本文所提控制方法可以有效抑制电压的波动,且具有更好的系统动态稳定控制效果。In view of voltage fluctuation of isolated island multi-wind diesel hybrid system due to wind energy and load fluctuation,the strategy for compensating system reactive power by the use the load side converter(LSC)of the double fed induction generator(DFIG)is proposed.Considering the situation of not completely measurable variables and uncertain parameters of LSC,a kind of date driven,model-free LSC additional control method based on deep reinforcement learning algorithm of continuous action space is proposed in this paper.The method can obtain control strategy in a non-linear manner to reduce voltage fluctuation and have a faster response speed and stronger adaptability to the uncertainty of the model.Firstly,the mathematical model of the isolated island multi-wind-diesel hybrid system is set up.Then,the reactive voltage control strategy of LSC in the continuous domain is optimized offline on the basis of deep reinforcement learning algorithm.After that,the features are extracted online through stacked denoising auto-encoder(SDAE)and the strategy network is used for online control.The numerical simulation results show that,compared with the traditional control strategy,the control method proposed in this paper can effectively suppress the voltage fluctuation and have a better dynamic stability control effect of system.

关 键 词：孤岛多风柴系统 双馈感应发电机(DFIG) 深度强化学习 附加智能电压控制器 动态稳定性 

分 类 号：TM614[电气工程—电力系统及自动化]