Nonintrusive load monitoring using TT-transform and neural networks

This paper presents nonintrusive load monitoring (NILM) method using the TT-transform feature extraction technique and multilayer perceptron with back-propagation neural network. The low sampling rate of load data were measured by using a smart meter to provide information on the operation of the loads such as air conditioner, personal computer and lighting. The accuracy of load identification with event based method is dependent on the measured signals and the selected features. To improve the identification of various types of loads, the TT-transform feature extraction technique is applied so as to provide additional features for characterizing the loads. Based on the load data obtained from the smart meter, the different features extracted such as real power, reactive power, power factor, apparent power, impedance, event signal (ΔP) and TT-transform features. These features are used as inputs to MLP-BPNN for identifying the various types of loads. The neural network is validated by testing with unknown dataset at different sampling rates. Neural network testing results showed that accuracy load identification can be achieved by using additional input features such as the TT-transform with best feature combination of real power, reactive power and TT-transform.