Performance Evaluation for Smart Electricity Meters Using Machine Learning

The performance evaluation for Smart Electricity Meters (SEMs) is essential for the device design and electric energy metering, especially in the typical natural environment such as dry heat. Nevertheless, the real performance evaluation is often affected by insufficient and unsuitable feature information. To address this problem, this research provides a performance evaluation model for SEMs by using the machine learning method. First, the Spearman Correlation Coefficient (SCC) is used to assign the different feature weights. These features include run time and corresponding environmental features including temperature and humidity. Next, a Compound Kernel Support Vector Regression (CKSVR) is proposed to fuse these weighted feature information and then predict the metering error change for SEMs. Examples from the Xinjiang high dry heat region demonstrate that the proposed SEMs evaluation model has higher metering error prediction performance and can provide the specific influence degree of different feature information, which can be used for the health management and standard validation of SEMs under typical natural environment.