基于机器学习的配电网异常缺失数据动态清洗方法

针对传统配电网数据清洗过程中异常数据判断阈值需要人为设定、缺失数据填补效率不佳的局限性，提出基于机器学习的配电网异常缺失数据一体化动态清洗方法。首先，基于局部异常因子检测算法和高斯混合模型，提出一种异常数据动态检测改进算法，实现配电网异常数据阈值的准确自动选择。其次，基于随机森林算法与最小二乘回归法，提出一种配电网缺失数据动态填补算法。根据缺失数据时间长度自适应优化填补算法，在保证数据填补精度的同时降低计算时间。在此基础上，通过异常数据检测和缺失数据填补共同构建一体化动态清洗架构。采用湖南某地区配电网数据进行实例验证，结果表明所提方法可实现异常辨识阈值准确自动选择，有效检测配电网异常数据，并且实现缺失数据填补精度与速度的平衡，具有较好的工程应用价值。

Dynamic data cleaning method of abnormal and missing data in a distribution network based on machine learning

There is a limitation of manual setting of an abnormal data judgment threshold and there will be inefficient filling of missing data in the traditional process of data cleaning in a distribution network. This paper proposes an integrated dynamic cleaning method for distribution network abnormal and missing data based on machine learning. First, based on a local outlier factor and Gaussian mixture model, an improved dynamic identification algorithm is proposed to realize the automatic selection of threshold of abnormal data. Second, based on the random forest algorithm and least squares regression method, a dynamic filling algorithm for missing data is proposed. Depending on the length of missing data, it adaptively optimizes the filling algorithm to ensure filling accuracy and reduce running time. An integrated dynamic cleaning architecture is built through abnormal data identification and missing data interpolation. The data of the distribution network in a certain area of Hunan are used for example verification. The results show that the proposed method can realize accurate and automatic abnormal data detection and achieve a balance between the filling accuracy and speed of missing data in a distribution network. This has good engineering application value.