Evaluation of classification methodologies and Features selection from smart meter data

Abstract The choice of the classification algorithm to map the feature vector to a known labelled database signature is an important step toward loads identification in non-intrusive load monitoring NILM. In this paper, we investigate the quality of load recognition when using various smart features and the commonly used classification algorithms. A low error rate is observed when using classification tree DT, k-NN and support vector machine SVM classifier, the error rate ranges between 20 % and 29 %. Among the smart meter features, the current waveform, the active/reactive power and the transient features have higher interesting recognition results when associated with a specific classifier.