Today, with continuously growing energy costs and decreasing reserves of fossil fuels, efficient energy consumption becomes more of importance. Providing feedback and adequate information for the user can inspire positive consumer behaviour, and lead to conscious energy consumption. A detailed breakdown of the monthly energy bill provides useful feedback about the energy consumption of the individual appliances in the household. This information can help to identify and decrease the waste of energy. The detailed (appliance level) electrical consumption data can supply useful information on the customs and behaviour of the consumer for the energy providers as well.
The outlined problem can be solved with intrusive metering, deploying a meter to each appliance and collect the measurements to a central hub. However, great expenses are associated with this method. A more promising and practical approach is non-intrusive metering (referred as Non-intrusive Appliance Load Monitoring” or „NIALM” in the literature). Non-intrusive metering is a set of techniques used to obtain estimates of the electrical consumption of individual appliances from measurements of voltage and/or current taken at a limited number of locations of the power distribution system in a building.
In my thesis I concentrated on the non-intrusive approach, and implemented a system based on it. The task is to disaggregate the premises-level electrical consumption into the consumption of the individual appliances. For this task I created a dynamic Bayesian network. With this model (Bayesian network) the main focus is to find the right parameters (transition and emission probabilities). For this I used hierarchical clustering and relative frequency. I tested the implemented system on real life data containing five appliances (washing machine, dishwasher, microwave, refrigerator and television).