Aiming at industrial parks and business parks equipped with photovoltaic (PV) power plants and vanadium redox battery energy storage devices, this work studies the collaborative scheduling optimisation problem of real-time response of two-area active distribution system to the random peak shaving demand of large power grids. Firstly, considering the randomness of source and load, the stochastic dynamic changes of PV output, various load demands and the grid peak shaving demand are described as Gauss-Markov processes. Secondly, the hierarchical dynamic control mode is used to transform the collaborative dynamic scheduling problem of two-area active distribution system into a two-layer scheduling optimisation model. The upper layer considers the total cost of operation as the optimal goal and resolves problems related to the task assignment of peak shaving demand for active distribution systems in each area. Meanwhile, the lower-layer areas are optimised to complete the peak shaving task assigned by the upper layer and realise the economic operation of the active distribution system in each area. This study proposes a corresponding model-independent double-layer Q learning algorithm to optimise the hierarchical scheduling strategy. A simulation is conducted to verify the effectiveness of this algorithm. These results indicate that the hierarchical scheduling optimisation mechanism and double-layer Q learning algorithm can effectively solve the collaborative scheduling problem of two-area active distribution systems considering the peak shaving demand of the power grid.

针对配备光伏电站和钒氧化还原电池储能装置的工业园区和商业园区，本研究研究了两区域有源配电系统对随机削峰需求的实时响应的协同调度优化问题。大型电网。首先，考虑源和负荷的随机性，将光伏输出的随机动态变化，各种负荷需求和电网调峰需求描述为高斯-马尔可夫过程。其次，采用分层动态控制模式将两区域主动配电系统的协同动态调度问题转化为两层调度优化模型。上层将总运营成本视为最佳目标，并解决与每个区域中主动配电系统的调峰需求任务分配相关的问题。同时，对下层区域进行优化，以完成上层分配的调峰任务，并实现每个区域中有源配电系统的经济运行。该研究提出了一种与模型无关的双层Q学习算法，以优化分层调度策略。进行仿真以验证该算法的有效性。