Degradation trend evaluation for smart meters under high dry heat natural environments

Smart Meters (SMs) degradation trend evaluation (DTE) is critical for accurate electricity metering and improved energy efficiency, particularly in extreme natural environments. However, environmental noise and insufficient interpretability often limit the performance of DTE. To remedy this problem, an optimized local density-based (OLD) method is first proposed for outlier identification, where a modified distance measurement and double nearest neighbors are used to improve the identification performance. Next, a DTE model, namely multi-kernel twin support vector regression (MTSVR), is presented to combine multiple environmental features using the proposed up-and down-kernel combination method. Integrating the MTSVR and OLD, real-world SMs datasets collected from a high dry heat area in China are utilized for model verification. The results demonstrate that the OLD-MTSVR framework has a superior performance for SMs degradation trend analysis. More importantly, the feature diversity can be analyzed quantitatively through the model interpretability.