DC Signature Analysis Aided Power Source Identification

This paper proposes a novel design for identification of input power sources in an electronic system fed through a direct current (DC)-bus powered by multiple input sources. A high-sampling power source identification module is designed to record highly precise DC voltage and current measurements. These measurements are passed through a moving average filter to remove high frequency outliers. A filtered support vector machines approach is proposed for classification of the input source, based on load-end rectified DC signatures. Using a variable sampling rate of up to 1000 samples per second, preliminary laboratory tests demonstrate that the proposed design works with > 76% gain in accuracy and > 66% gain in sensitivity over the state-of-the-art support vector machine classifier in 0.5 second of training time and is highly robust to measurement noise.