Robust-PCA Deep Learning for PQ disturbances classification using Synchrosqueezing Wavelet Transform

Abstract

In this paper a Robust-PCA Deep Learningalgorithm using Synchrosqueezing Wavelet Transform isproposed for PQ disturbances mutli-classification.The algorithmwas implemented and programmed in MATLAB using customcode. This approach avoids white noise, outliers and overfittingphenomena. The Synchrosqueezing Wavelet Transform isperformed and a Robust-PCA mapping is done. External data isnecessary to perform the pretreatment for autoscaling. A DeepFeed Forward Neural Network is implemented with 5 layers, 3 ofthem are hidden layers with more than 1 million parameters to fit.The quality of the solution is validated by the cross validation ofparameters, R2 and Q2.Moreover, mean square error (MSE),theroot of the mean square error (RMSE), the mean absolutepercentage error (MAPE), Akaike information criterion (AIC) andthe Schwarz information criterion (SBC) are estimated. Theadjusted R2 value is 0.989 and the RMSE obtained is 1.789.Thevalue of R2 is 0.995. All these parameters are calculated over thetest set.