Skip to Main Content
Wide Area Monitoring (WAM) of power systems is achievable with the help of a large amount of time-stamped synchrophasor data generated by PMUs. On the other hand, massive data from PMUs has also brought challenges to store, analyze and transmit results without causing a bottleneck in the available information processing infrastructure. In addition to the amount of data, the dimensionality of synchrophasor data is bound to increase with more PMUs coming into operation. In order to enable real time surveillance of the grid, high-speed synchrophasor data has to be processed before a new set of data arrives for processing. Machine learning techniques which are being studied for power systems may suffer from the “curse of dimensionality” thus, detrimentally affecting their performance. In this paper, we introduce an online technique to reduce dimensionality of synchrophasor data on the fly. This method will extract correlations between synchrophasor measurements such as voltage, current, frequency etc. and represent it with their principal components without the loss of too much information. The proposed method can be used as a pre-processor before storing or transmission of synchrophasor data where trends are more important than exact data.