基于强化学习的混合元启发式暂态电压稳定特征选择方法及可解释性研究

Reinforcement Learning-based Hybrid Element Heuristic Transient Voltage Stability Feature Selection and Its Interpretability

新型电力系统发展背景下,使用有效的特征选择方法来提取与暂态电压稳定强相关的关键响应特征,对研究暂态电压失稳机理与系统潜在安全隐患具有重要意义。为此,提出一种基于改进过滤法与混合元启发式包装法的复合框架进行特征选择的新方法。基于对称不确定性值改进的最大相关最小冗余性准则进行特征粗筛;将Q学习强化学习融合至元启发式优化算法中,并采用开发探索折衷策略以增强特征细选能力,获取最优关键响应特征子集。在此基础上,采用沙普利值加性解释归因理论综合分析各筛选特征对暂态电压稳定的影响与系统薄弱环节。新型电力系统算例验证了所提方法的有效性。

Under the development of new power systems,it is of great significance to extract the key response features strongly related to the stability of transient voltage with an effective feature selection for the studies of the mechanism of transient voltage instability and the potential security risks of the system.Therefore,a new feature selection method is proposed based on the composite framework of the improved filtering method and the hybrid element heuristic packaging method.The improved Max-Relevance and Min-Redundancy criterion of symmetric uncertainty value is firstly used to have a coarse screen of the features.Then the Q-learning reinforcement learning is integrated into the meta-heuristic optimization algorithm,and the exploitation and exploration compromise strategy is used to enhance the feature fine selection ability to obtain the optimal critical response feature subset.On this basis,the Shapley additive explanation is applied to comprehensively analyze the influences of each of the screening features on the transient voltage stability and the weak links of the system.The effectiveness of the proposed method is verified by an example of a new power system.