Research on Anomaly Detection of Smart Meters Data Based on Feature Clustering Method

In recent years, electricity market has developed rapidly, the market trading modes and electricity price system have become more diversified. Because of these, energy metering is facing great revolution and the metering data has a significant influence on the interest of market members and even the development of the whole electricity market. This paper presents an approach to quantity the veracity of metering data. First, the accuracy index of metering data is defined. Then, based on the daily load curve composed of the hourly metering data, K-means clustering algorithm is used to classify the loads into several typical patterns. To ensure the clustering effect, a new clustering validity index is presented to determine the optimal clustering number. Deviation between the test load curve and the typical curve is measured, and the discriminant basis of the metering data veracity is established according to the similarity of the two curves and the smoothness of the test curve. At last, the accuracy index is calculated, and the veracity level is ranked in accordance to the setting standard. Experiments on existing data sets shows that the proposed method is feasible and effective.