Modeling and Prediction of Condu Modeling and Prediction of Conducted Emission Using Machine Learning with Algorithm Support Vector Machine (SVM)

Modeling and Prediction of Conducted Emission Using Machine Learning with Algorithm Support Vector Machine (SVM)

Abstract

Electronic devices used can produce unwanted emissions, both radiated and conducted. Some techniques that have been used to reduce emissions include filters, shielding, switching modification, and converter design. One way to find out these emissions is to predict emissions that have excess can reduce development costs and time. In this study the prediction of EMI conducted emission power level generated by the LED driver uses several periodic modulating signals (sine signals, triangular signals, and box signals) and non periodic modulating signals (noise signals) with frequencies from 150 KHz - 30MHz and amplitude 0 - 1.2 V. Characteristics and EMI conducted emission obtained is used as training data in machine learning. Using machine learning can find out how accurately the EMI conducted emission power level is predicted by the SVM algorithm (Support Vector Machine). Machine learning will display the results of a comparison between training data and testing data to determine theaccuracy of the EMI power level conducted emission generated by the LED driver. From the data the prediction results are obtained, where the SVM (Support Vector Machine) algorithm results in accuracy obtained on periodic signals (sine signals have an average of 97.257%, triangle signals have an average of 95.262%, and box signals have an average of 92,02%) and non periodic signals (signal noise has an average of 91.022%).