8th International Conference on Advanced Power System Automation and Protection (APAP2019)

705 / 2019-04-23 20:20:36

Non-intrusive Load Decomposition of Stacked Hybrid Recurrent Neural Networks

non-intrusive load decomposition, stacked hybrid recurrent neural networks, deep learning

Non-intrusive Load Decomposition (NILD) is the process of extracting the power consumptions multiple appliances from the total consumption signal of the building. For the problems of weak time characteristic, high misjudgment rate and low accuracy of load decomposition of traditional algorithm for multi-state equipment, this paper, based on deep learning, proposes a stacked hybrid recurrent neural networks (SHRNNs) method to decompose power load. Firstly, the NILD model of SHRNNs is constructed by combining gated recurrent unit neurons, long short-term memory neurons and convolution unit neurons, which often used to solve the time series problem. Then, based on the UK-DALE-2017 dataset，different input sequence lengths are set according to different electrical equipment, and the total load sequence is decomposed by using the built model. Finally, the validity of the model is verified by comparing with the traditional load decomposition algorithm.