Appliance Identification using Long Short-Term Memory Recurrent Neural Networks

Abstract

This thesis discusses the application of Long Short-Term Memory Recurrent Neural Networks to identify electrical appliances based on the current they draw. An LSTM-model, implemented using Tensorflow, is trained and validated using the PLAID-dataset. This model achieves an average F1-score of 92% on a testing subset of the data, thereby improving on the state of the art. The resulting model is robust to noise, and generalizes well to previously unseen examples, provided the data are pre-processed to the correct format. In conclusion, LSTMs are well-suited for the appliance identification problem, but the amount of data and computing power required restrict their practical applications.