Bad data detection and multi-level classification of events in synchrophasor measurements using AXGBoost algorithm

The power grid operators collect the data on a smart grid via the phasor measurement unit (PMU), which is placed at several points on the bus. However, the data at the PMU is prone to bad data; to classify that data, a detection scheme must be executed in the server system. The events in PMU data were subjected to several transmission conditions in the distribution system, and detecting those conditions is the major need for research. Hence to cope with that, a multi-level classification of synchrophasor’s data is proposed in this paper. An agglomerative clustering-based extreme gradient boosting (AXGBoost) classifier is proposed for a multi-level classification. The puzzle optimization method updates the probability of a data line point to the agglomerative cluster detector. The model is analyzed on IEEE 14 and 39 bus system PMU data for classification and evaluation of a model. As a result, the classification accuracy of 99.49% is obtained for IEEE 14 bus, while the accuracy for IEEE 39 bus is 99.74%, which is comparably higher than the 14 bus system. The obtained result on both systems is compared with state-of-the-art techniques such as isolation forest classifier, weighting and imbalance bagging and PMUNET.