面向新能源电力系统状态估计的伪波动数据清洗

Pseudo-fluctuation Data Cleaning for State Estimation of New Energy Power System

随着新能源并网容量增加,电力系统量测数据波动性增强,脏数据极易伪装成波动数据入侵电网二次侧信息系统,造成脏数据辨识难度加大,电力系统状态估计精度降低。面向新能源电力系统状态估计,提出一种基于数据时空关联特性的量测数据多级清洗辨识方法。以传统鲁棒容积卡尔曼滤波为算法基础,利用新息向量进行1级清洗,辨识突变数据;利用皮尔逊时序相关系数对突变数据进行2级清洗,辨识波动数据与可疑脏数据;利用卷积神经网络对可疑脏数据进行3级清洗,深度辨识波动数据与伪波动脏数据;通过辨识结果动态调整噪声尺度因子,修正脏数据及动态状态估计方向。该方法旨在改进动态状态估计,使其具有较高的伪波动脏数据识别及防御能力,在新能源系统数据波动性较强的情况下仍具有较高的估计精度和准确性。最后,利用省级实网负荷及新能源数据进行实网算例分析,验证所提出方法的可行性和有效性。

A large number of new energy power generation devices connected to the grid may amplify the fluctuation of power system measured data. Bad data can easily be disguised as fluctuation data to invade the secondary information system of power grid and impair the state estimation of power system. In accordance with the state estimation of new energy power system, we propose a multi-level cleaning and identification method of measured data based on the temporal and spatial correlation characteristics of the data. Based on the traditional CKF algorithm, innovation vectors are adopted to perform primary cleaning so as to identify mutation data, the Pearson is adopted to perform secondary cleaning so as to identify suspicious bad data, and the CNN is adopted to perform tertiary cleaning so as to identify pseudo-fluctuation bad data. Then, the model dynamically adjusts the noise scale factor by using the identification results, and realizes the correction of the bad data. The purpose of the method proposed in this paper is to improve the traditional robust state estimation method. The improved method can be utilized to identify and defend against pseudo-fluctuation bad data, and it still has high accuracy in new energy power systems with strong data volatility. The effectiveness and feasibility of the above method are verified by real grid data.