基于改进强化学习的新型电力系统低频减载控制方法研究

Research on an Under-Frequency Load Shedding Control Method for New-Type Power Systems Based on Improved Reinforcement Learning

中文关键词:低频减载  新型电力系统  深度强化学习  物理-数据驱动

keywords:under-frequency load shedding  new-type power system  deep reinforcement learning  combined physical-data driven

中文摘要:随着新能源的大规模并网，系统惯量显著减少，导致系统频率稳定性面临前所未有的挑战。低频减载（under-frequency load shedding，UFLS）作为维持系统频率稳定的有效措施，其在新型电力系统中的应用显得尤为重要。在新型电力系统的背景下，传统的UFLS方法与实际功率缺额存在不匹配的问题，亟需开发新的UFLS策略。利用深度确定性策略梯度（deep deterministic policy gradient，DDPG）算法，提出一种新型的UFLS方法。基于此方法对DDPG算法进行了改进，以适应UFLS问题的特殊需求。由于新型电力系统的复杂性和多样性，引入了双经验池结构，通过独特的训练方式增强了算法训练的效率和稳定性。为了进一步提高模型的精度，将模型驱动与数据驱动进行融合，增强了模型的准确性和鲁棒性。在改进的IEEE 39节点系统上实验，验证了该方法的有效性。

Abstract:With the large-scale integration of renewable energy into the grid，the system inertia has significantly decreased，posing unprecedented challenges to the frequency stability of the power system. Under-frequency load shedding （UFLS），as an effective measure to maintain system frequency stability，has become particularly important in the context of new-type power systems. Against the backdrop of new-type power systems，traditional UFLS methods exhibit a mismatch with the actual power deficit，necessitating the development of new UFLS strategies. A novel UFLS method is proposed based on the deep deterministic policy gradient （DDPG） algorithm. Within this method，the DDPG algorithm is further improved to meet the specific requirements of the UFLS problem. Due to the complexity and diversity of the new-type power system，the dual empirical pool structure is introduced，which overcomes the challenges of the algorithm in training efficiency and stability through a unique training approach. To further improve the model’s accuracy，the model-driven and data-driven are fused to enhance the accuracy and robustness of the model. The effectiveness of the method is verified through validation on an improved IEEE 39 bus system.