Predicting user behavior using transition probability.
Master's thesis, University of Twente.
Full text available as:
As energy becomes more expensive and the environmental effects of fossil fuels become more apparent, a greater interest arises in reducing our energy needs. Projects intended to find new ways of making our everyday lives more energy efficient have become an integral part of the struggle to sustain our current quality of living. These projects target not only large, industrial processes, but also processes on a domestic scale which have a more direct approach to changing the way we consume energy.
The Go-Green project is one of these projects focusing on making our homes a “greener” place. Inhabitants are monitored and information is acquired about the usage of appliances, the behaviour of people, and the state of the surroundings. Energy is saved by learning user habits and reacting to inefficient energy consumption. Inefficiencies can take on the form of appliances that are left running unintentionally, or a central heating system that is warming up an empty house. However, dealing with wasting energy and controlling the way people use their home, should at no point compromise their living comfort. In fact, user comfort is to be increased by applying the same knowledge of user habits to automatize the interaction with devices.
Learning and predicting user habits require a suitable machine learning technique. Research has shown that there are several candidates, all with their own advantages and disadvantages. The specific requirements for this project made it difficult to find a single technique that is suited for our needs. Especially the behavior prediction aspect seemed insuperable for techniques that are primarily applied for behavior classification.
In this thesis we design a new machine learning technique, specifically aimed at predicting user behavior. “Transition probability”, as this technique is called, is explained and an overview is given about its origin and its possible future. We discuss the theory behind this technique and provide a practical validation of our theory by means of simulation. Several key aspects are evaluated to obtain a clear view of performance and applicability to our problem.
|Item Type:||Master's Thesis|
|Research Group:||EWI-PS: Pervasive Systems|
|Research Program:||CTIT-WiSe: Wireless and Sensor Systems|
|Research Project:||Go Green: Greener House Through A Self-learning, Privacy-aware User-centric Energy-aware Wireless Monitoring And Control System|
|Deposited On:||18 June 2012|
Export this item as:
To correct this item please ask your editor
Repository Staff Only: edit this item