【作者】 杨文伟；

【导师】 彭显刚；

【作者基本信息】 广东工业大学 ， 电气工程， 2025， 硕士

【摘要】 在新能源渗透率不断提高的背景下,主动配电网作为未来智能电网的重要发展方向,具有提升电网调度灵活性、优化能源配置和增强系统稳定性的优势。针对主动配电网优化调度问题,本文重点研究分布式电源发电设备、无功补偿设备、储能设备和柔性负荷四类可控资源的协调控制策略,建立相应的数学模型,并提出基于联邦强化学习与扩散强化学习的日前-日内协调优化调度策略,以提高配电网的运行经济性、安全性和适应性。首先,本文详细构建了风力发电单元、光伏发电单元、燃气发电单元、静止无功补偿器、有载调压变压器、静止投切电容器组、电池储能控制器及柔性负荷等可控资源的数学模型,分析其运行特性及调度约束,为优化调度策略提供理论支撑。在此基础上,提出计及资源承载力的日前优化调度模型,以分布式电源运营商、储能运营商、主动配电网运营商和电力用户的综合收益最大化为目标,构建多主体协同优化架构。为兼顾隐私保护和短期运行需求,采用联邦强化学习方法进行迭代求解,实现多主体分布式优化计算,提高优化调度的安全性和稳定性。其次,基于日前优化调度策略的计划量,进一步提出计及安全承载力的日内优化调度模型,以最大化主动配电网安全承载力和最小化调整成本、运维成本为优化目标,同时引入电压偏移、反向负载率、支路载流量和系统网损等安全评估指标作为约束条件。为提升调度策略对动态环境的快速适应能力,本文采用结合自注意力机制的扩散强化学习方法,对日前-日内协调优化调度策略进行校正更新,并通过强化学习的自适应优化能力,实现模型的短时求解。最后,本文基于IEEE33节点改进算例进行仿真验证,分别分析不同优化调度策略对主动配电网资源承载力和安全承载力、交易收益、系统网损、电压偏移等关键指标的影响。实验结果表明,所提出的方法能够在实现隐私保护性能的前提下,有效提升主动配电网的经济性、运行稳定性和动态适应能力,为主动配电网的优化调度提供了理论支撑和实验指导。更多还原

【Abstract】 In the context of the increasing penetration rate of new energy,active distribution network,as an important development direction of smart grid in the future,has the advantages of improving the flexibility of power grid scheduling,optimizing energy allocation and enhancing system stability.Aiming at the optimal scheduling problem of active distribution network,this thesis focuses on the coordinated control strategy of four controllable resources,namely distributed power generation equipment,reactive power compensation equipment,energy storage equipment and flexible load,establishes the corresponding mathematical model,and proposes the day-ahead and intraday coordinated optimal scheduling strategy based on federated reinforcement learning and diffuse reinforcement learning.To improve the operation economy,safety and adaptability of distribution network.First of all,this thesis constructs the mathematical models of controllable resources such as wind power unit,photovoltaic power unit,gas power unit,static reactive compensator,on-load regulator transformer,static switching capacitor bank,battery energy storage controller and flexible load in detail,analyzes their operating characteristics and scheduling constraints,and provides theoretical support for optimizing scheduling strategies.On this basis,a day-ahead optimal scheduling model considering resource carrying capacity is proposed,and a multi-agent collaborative optimization framework is constructed to maximize the comprehensive benefits of distributed power supply operators,energy storage operators,active distribution network operators and power users.In order to take into account the privacy protection and short-term operation requirements,the federated reinforcement learning method is used for iterative solution to achieve multi-agent distributed optimization calculation and improve the security and stability of optimal scheduling.Secondly,based on the planned quantity of day-ahead optimal scheduling strategy,a day-ahead optimal scheduling model is further proposed,which takes maximizing the safety carrying capacity of active distribution network and minimizing the adjustment cost and operation and maintenance cost as the optimization objective,and introduces the safety evaluation indexes such as voltage deviation,reverse load ratio,branch load and system network loss as the constraint conditions.In order to improve the rapid adaptability of scheduling strategies to dynamic environments,this thesis adopts the diffusion reinforcement learning method combined with self-attention mechanism to correct and update the day-ahead and intraday coordinated optimal scheduling strategies,and realizes the short-term solution of the model through the adaptive optimization ability of reinforcement learning.Finally,based on the IEEE33 node improvement example,this thesis carries out simulation and verification,and analyzes the influence of different optimal scheduling strategies on key indicators such as resource carrying capacity and security carrying capacity,transaction income,system network loss,and voltage offset.The experimental results show that the proposed method can effectively improve the economy,operation stability and dynamic adaptability of the active distribution network under the premise of realizing the privacy protection performance,and provide theoretical support and experimental guidance for the optimal scheduling of the active distribution network.更多还原