Advanced Customer Data Analytics for the Future Smart Grid

Abstract

The penetration of renewable energy generation is expected to keep increasing for the years to come. Power grid needs to become more intelligent to accommodate such renewable sources. One of enabling approaches is through smarter and more reliable demand-side management. To successfully promote and deploy a broad range of demand-side management measures, we will need to make full use of and extract more actionable insights from the massive volume of under-utilised smart grid data. Therefore, approaches towards such advanced data intelligence are the main focus of this thesis. Appliance-level load models are crucial to many demand-side management measures. First, an extensible framework for residential load disaggregation problems is proposed. The approach examines both the modelling of home appliances as well as the solvers. A segmented integer quadratic constraint programming (SIQCP) approach is developed to disaggregate a household power profile into the appliance level. The scalability and computation efficiency of the NILM solving framework is also visited. A bootstrap filtering based solver is built to address the scalability issue, while a hybrid solver that combines the constraint programming and the proposed SIQCP solver is developed to improve the computation efficiency significantly. Moreover, a hierarchical hidden Markov model framework to model home appliances is proposed to provide better representation and estimates than the conventional hidden Markov model in the NILM problems. For single-meter load forecasting, a framework relying on LSTM recurrent neural network is proposed. It is also demonstrated that the forecasting accuracy can be further improved by learning the appliance-level load profiles of the major appliances. At last, demand-side enhancing applications that make use of all the proposed works to are discussed in details, including an appliance recommending system.