Royal Holloway: Deep learning for countering energy theft

Table Of Contents

• Energy theft is considered a crime in many countries, and it usually refers to the intentional avoidance of paying legitimate bills by energy users.
• Energy theft not only disrupts a fair market for electricity consumption, but also causes significant financial losses to utility companies.
• One of the reasons for the rapid growth of advanced metering infrastructure (AMI) over the past few years is to reduce the non-technical loss (NTL) caused by electricity theft.
• Due to the complexity of power theft methods, it is difficult to achieve effectiveness with manual detection alone. Machine learning-based energy theft detection (ETD) models are also being deployed in smart grids and can be used to assist in the automatic detection and identification of electricity thieves.
• Research in the field of machine learning in ETD can be based on classical machine learning (non-deep learning) and deep learning. Deep learning can automatically extract features and optimise their weights by building models – the weight is the parameter that is varied to adjust the expected computational outcome, and the training process is essentially an update of the weights.
• Current ETD models in deep learning can be based on different branches such as multilayer perceptron (MLP), convolutional neural network (CNN), and long short-term memory (LSTM).
• The goal of this project is to build a novel ETD model based on convolutional long short-term memory (ConvLSTM) and to verify its accuracy and robustness in identifying electricity thieves, and to show that it outperforms the MLP and CNNLSTM models.