Projects

These are few projects I have completed or still be working on.

• Disaggregating Solar Generation Using Smart Meter Data and Proxy Measurements from Neighbouring Sites:  This paper investigates the problem of disaggregating solar generation from smart meter data when historical disaggregated data from the target home is unavailable,  and deployment characteristics of the PV system are unknown.  The paper has been accepted in e-Energy'21 conference.


• Solar Disaggregation: State of the Art and Open Challenges:  In this notes paper we survey the literature on solar disaggregation and describe datasets that can be used for evaluating disaggregation methods.  We identify limitations and threats to validity of this research,  and discuss existing challenges and how they can possibly be addressed. The note paper has been accepted in the 5th International Workshop on Non-Intrusive Load Monitoring.


• Disaggregation Based Occupancy Estimation:  Accurate room-level occupancy estimation is essential to different building applications for energy saving,  smart building management.  However,  there is often a trade off between sensor budgets and estimation accuracy.  In this paper,  we try to find a balance point in those two metrics and propose a new occupancy estimation method by disaggregating accurate floor-level counts via existing common sensors available at room-level.


• PCGMM:  This project is a project which relates to Probability Graph Model (PGM).  In this project,  we explore different Markov Models,  such as multi-dimension markov chain and multi-layer markov chain,  for Procedure Content Generation in the game The Legend of Zelda.  We compare the performance of the models in terms of playability,  diversity and style match.


• Unsupervised Anomaly Detection by Analyzing Power Consumption of Individual Households:  This project compares the performance of different unsupervised anomaly detection techniques on real world residential power consumption data.  It also investigates how different feature consutrction methods affect the performance of anomaly detection methods.