A Novel Multisource Feature Fusion Framework for Measurement Error Prediction of Smart Electricity Meters

The precise measurement error prediction for smart electricity meters (SEMs) under extreme natural environments is necessary for improving electrical energy efficiency and promoting smart grid development. Nevertheless, insufficient feature information and model fusion performance often limit practical measurement error analysis. To remedy this problem, a novel multisource feature fusion framework is proposed for measurement error prediction utilizing improved kernel support vector regression (IKSVR) and optimized adaptive genetic algorithm (OAGA). First, the Pearson correlation analysis and Z-score normalization are carried out for SEMs data preprocessing, which can be used for selecting and weighting the input feature. Next, IKSVR is employed to model the measurement error under a highly cold environment, where an improved kernel fusion structure is proposed to extract different feature information including time and environmental factors. To solve the multiparameter optimization problem in IKSVR, the OAGA is further presented for parameter setting. Using the actual SEMs dataset from the highly cold region in China, different experiment results demonstrate that our framework has a superior prediction performance compared with some popular data-driven approaches under small samples.