A group from Brasilia University jointly with their colleagues from Sao Paulo University and Bahia State University presented an innovative control system for household electricity consumption called MELISSA (Modern Energy LLM-IoE Smart Solution for Automation). This is a smart assistant for using in every home. It is based on the combination of Internet of Things and Artificial Intelligence: MELISSA collects data from home meters, analyses the use of electricity and provides personal recommendations adjusted to the life style of a specific family.
The scientists started their work with the architecture design and divided the system into two independent modules. The analytical module was designed to process data received from smart meters, sockets, thermometers and motion detectors. The language module was designed to interact with users. Then the scientists selected the optimal language model and trained it using the typical energy consumption scenarios: the consumption dynamics depending on time of day, temperature, family users and home appliances used.
Simultaneously they created the data collecting and filtering system: the specialists connected IoT-gadgets in the participants’ homes, organized data transfer in real-time mode and the primary treatment of data — eliminated noises, identified anomalies, built consumption profiles. Then the data was channeled to the language module, which developed recommendations in the format understood by any ordinary user. They took into account not only technical parameters, but behavioral patterns of the participants, as well: if they willing to follow the advice, how often they interact with the system, what style of communication is most comfortable for them.
The scientists paid special attention to the system’s ability to adjust for a specific user — to learn from interaction, to take feedback into account and to adjust its tips. Thanks to such architecture, MELISSA not only analyzed the consumption, but talked to people explaining what should be changed and why: for example, to turn on the washer during the night time, to decrease the temperature in the air conditioner or to readjust the boiler.
The tests of the system went on for one year covering 97 households with different levels of income. The results turned out to be quite convincing: on average, the energy consumption decreased by 5.66%, and in the families with high daily consumption — by 7–8%. Quite simple recommendations worked especially well: e.g., switching off the appliances on standby, adjustment of thermostat or shifting the laundry to the nighttime when the tariffs are low. More costly measures, such as equipment replacement or installing heat insulation were implemented much more rarely.
Interesting enough, MELISSA did not just reduce the consumption, but also increased the engagement of the users: due to understandable explanations and precise recommendations of the system, people were more willing to listed to its advice. As an outcome of the experiment, 78% of the participants appraised the system’s performance at the level higher than 7.5 out of 10, emphasizing that the system was user-friendly and provided for the feeling of control of the situation.
The Brazilian scientists propose to extend the sphere of MELISSA system application by including not only energy consumption control, but also water and gas consumption control, and to integrate it with the home renewable energy devices, such as solar panels.
