Appliance Classification using BiLSTM Neural Networks and Feature Extraction

One significant challenge in Non-Intrusive Load Monitoring (NILM) is to identify and classify active appliances used in a building. This research focuses on the classifying process, exploring different approaches for the feature extraction of the appliances' power load to improve the classification accuracy. In this paper, we present a new method - Spectral Entropy and Instantaneous Frequency-based Bidirectional Long Short Term Memory (SE-IF BiLSTM). It uses feature extraction from the power load to obtain information, such as instant frequency, spectral entropy, spectrogram, Mel spectrogram and signal variation, to feed BiLSTM Neural Network. We also test different options for the BiLSTM to decide the most optimal settings. This method improves the classification performance, achieving up to 98.57% classification accuracy.