Use this URL to cite or link to this record in EThOS:
Title: Smart meter data analytics
Author: Xu, Minghao
Awarding Body: University of Bath
Current Institution: University of Bath
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
In recent years smart meters and advanced metering infrastructures (AMI) have been rolled out to a substantial number of domestic consumers across the world. An unprecedented amount of fine-grained load data have since been generated. Meanwhile, owing to the rapid and ongoing digitalization of the society, a variety of information about the consumers which was unattainable has streamed in and becomes available, such as the consumer's age, education level, income, etc. However, traditional analytics used in power systems are unable to handle the smart meter data efficiently and effectively due to the high volatility, large volume and fast generating speed of domestic load data. The added dimensions to the smart meter data from other sources increase the variety of data and further complicate the processing of the data. This thesis proposes a range of methods to address the challenges from two key aspects: 1) Uncovering the underlying patterns of the smart metering data. This thesis proposes a novel load forecasting methodology that leverages the knowledge learned from one forecasting task to achieve more efficient and accurate forecasting for another by utilizing transfer learning along with deep learning. The adoption of deep neural networks enables the effective modelling of highly complex and nonlinear relationships within smart metering data and has endowed us with strong predictive power. Additionally, transfer learning would further improve the predictive performance and significantly reduce the required amount of data, computational power, and the efforts for hyperparameter optimization. 2) Revealing the interconnection/correlation between data from other sources and smart meter data. Based on the sources, other available data could be classified into two groups, i.e., social-economic/demographic data of consumers and data from the power system. For the social-economic data, this thesis first proposes an ensemble learning framework that could not only predict the social-economic status accurately form smart metering data, but also provide insights into the correlation between the two sets of data due to the model's interpretability. Conversely, a deep Convolutional Neural Network (CNN) based model is proposed to infer the load characteristics from the social-economic data of the consumers. It leverages the convolutional kernel and deep architecture to overcome the hurdle brought by mixed types of data and infers multiple load characteristics simultaneously. It is validated on real data and demonstrates an improvement in both the learning efficiency and the prediction accuracy compared to predicting each characteristic separately. As for the data from the power system, this research preliminarily focuses on the phase connectivity of a consumer. The phase connectivity is not commonly available, however, keeps gaining increasing attention due to the critical need for Low Carbon Technologies integration and network balancing. A novel Spectral and Saliency Analysis (SSA) method is developed to accurately identify the phases of consumers using their smart metering data.
Supervisor: Li, Ran ; Evans, Adrian ; Li, Furong Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available