LAUSR

The measurement error assessment for smart electricity meters consists of the measurement error prediction and the stress factors evaluation, which can be used for improving equipment quality and saving power grid costs, especially under extreme natural environmental stresses. However, actual measurement error assessment suffers from the environmental noise and insufficient feature information. To tackle this problem, in this article, an optimized kernel density estimation (OKDE) is first proposed to identify potential outliers, where a modified distance function and adaptive kernel bandwidth are used to obtain the outlier score. Next, a measurement error assessment method, namely the modified double-kernel support vector regression (MKSVR), is proposed to fuse measurement error and multiple stress features using the modified double-kernel function. Combining the OKDE and MKSVR, actual dataset from the high dry heat region shows that the proposed assessment framework has better evaluation performance. Compared with some classical prediction methods, the OKDE–MKSVR framework has profound outlier detection and measurement error assessment performance under the small sample conditions.