At Southampton, we have developed an intelligent agent that sits in a home, and by learning its thermal properties, controls the heater so as to minimise the cost or the carbon emitted from the heating load. It uses multi-output Gaussian process regression to predict on-site external temperature as well as carbon intensity of the grid.
Below are the links to a demo of the work and the paper detailing the model and algorithms.
Demo: http://vimeo.com/35586957/
Paper: http://eprints.ecs.soton.ac.uk/22235/