Analysis of Artificial Intelligence based Methods For Sensorless Disaggregation of the Residential Electric Power Signals

Seeking sustainable solutions for ever-increasing energy demand is one of the key priorities today. Energy conservative behavior of domestic users can enable with appliance-wise consumption statistics. Non-Intrusive load monitoring (NILM) is a popular technic to calculate appliance-wise energy consumption without using individual sensors. The availability of energy data and the use of smart meters increase the popularity of NILM in many application areas in recent years. This study presents a comprehensive analysis of the application of two key deep Artificial Intelligence (AI) families, Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) for NILM. An unbiased comparison performs in a common platform between CNN and Gated Recurrent Unit (GRU) a member of RNN family in a different dimension of NILM. Regression-based sequence to point input to output mapping was considered for comparison along with the soft association method. Both NN models were modified and improved prior to compassion. Thereafter, a rigorous comparison was performed in disaggregation performances, time taken by each NN, and computational complexity. Performance indices including, accuracy, precision, recall, and Fmeasure has been used for the comparison. Results provided important facts related to each NN and the relationship between architectures, appliances, activation signatures, and available volume of training data with respect to disaggregation performances.