Innovation-saturated Koopman Kalman filter for distribution system dynamic state estimation against measurement outliers
Innovation-saturated Koopman Kalman filter for distribution system dynamic state estimation against measurement outliers
复制标题
DOI：
10.1016/j.egyr.2023.05.171
复制
发表时间：
2023-10
期刊：
影响因子：
5.2
通讯作者：
Kai Wang;Min Liu;Yanlu Man;Chaowen Zuo;Wang He
中科院分区：
文献类型：
--
作者：
Kai Wang;Min Liu;Yanlu Man;Chaowen Zuo;Wang He
传感器误差、模型不确定性、周围环境变化、数据丢失或网络恶意攻击都可能导致离群值的产生，从而污染配电系统的测量过程。针对这一问题，本文提出了一种基于创新饱和库普曼卡尔曼滤波（IS-KKF）的配电系统动态状态估计方法。库普曼卡尔曼滤波器（KKF）是一种无模型和数据驱动的滤波器，它通过传统的线性卡尔曼滤波器来估计非线性动态。为了使库曼卡尔曼滤波对测量异常值具有鲁棒性，创新饱和机制对其应用了饱和函数。为了校正状态估计，将该机制应用于滤波过程。当出现异常值时，对畸变创新进行饱和处理，防止状态估计结果被破坏。饱和边界的自适应调整是该机制的一个特点。在IEEE 118总线测试系统上进行了大量仿真，验证了该方法的有效性和鲁棒性。仿真结果表明，该方法能够有效地抑制不同幅度、类型和连续时间的异常值，具有显著的计算效率，且不需要额外的测量冗余。
Sensor errors, model uncertainty, changes in the surrounding environment, data loss, or malicious network attacks can all result in outliers, which can pollute the measurement process of the distribution system. In this study, to address this problem, a dynamic state estimation method for distribution systems based on innovation-saturated Koopman Kalman filter (IS-KKF) was proposed. The Koopman Kalman filter (KKF) is model-free and data-driven, and estimates nonlinear dynamics through the conventional linear Kalman filter. To make the Koopman Kalman filter robust to measurement outliers, the innovation saturation mechanism applies a saturation function to it. To correct the state estimation, this mechanism is applied to the filtering process. When outliers occur, the distorted innovation is saturated to prevent the state estimation results from being destroyed. Adaptive adjustment of the saturated boundary is a feature of this mechanism. Extensive simulations were carried out on the IEEE 118-bus test system to verify the effectiveness and robustness of the proposed method. The simulation results demonstrate that the proposed method can effectively reject outliers with different amplitudes, types, and continuous times, has significant computational efficiency, and does not require additional measurement redundancy.