Estimation of occupant behavior by GMM-HMM with data measured by HEMS

The importance of energy management, which considers the balance between electricity supply and demand to reduce greenhouse gas (GHG) emissions in the household sector, is becoming increasingly recognized. Consequently, interest in utilizing Home Energy Management Systems (HEMS) for residential energy management is growing. HEMS visualize the energy consumption of household devices and control them to achieve energy saving and peak load reduction without compromising the services received by residents.

In many regions, including Japan, research has primarily focused on verifying the electricity consumption reduction effects of HEMS introduction [1]. However, HEMS data has been mainly used for visualizing energy consumption, with limited progress in data applications and appliance control. It is essential to establish ways to utilize HEMS data to effectively achieve energy conservation in households.

Household energy consumption and the usage of individual household devices depend on personal lifestyle. Therefore, it is essential to consider lifestyle patterns when implementing energy-saving measures. However, it is difficult to grasp these patterns. Although, assuming that the operation of household devices is fundamentally based on human behavior, it is possible to interpret resident behavior from the energy consumption data measured by HEMS and understand household lifestyle patterns. In other words, it is possible to estimate the lifestyle of each household without the need for additional sensors, enabling more comfortable and optimal energy-saving measures.

We have begun analyzing the 30-minute interval power consumption data measured by HEMS installed in households of the “Senri-Maruyama-Hill Smart Community” in Osaka [2]. This study aims to clarify the characteristics of residents’ lifestyles from HEMS data as the first step towards appliance control. Specifically, we extracted the states of rooms and home appliances using Gaussian Mixture Model (GMM) and identified state transitions with Hidden Markov Model (HMM). As a result, we were able to understand the states and frequencies of each room, such as being in a room while watching TV or being in a room without watching TV, as well as the usage times and frequencies of household appliances. Additionally, from these results, we could infer lifestyle patterns such as sleep and outings. The accuracy of these findings was evaluated through surveys conducted with the residents.

The proposed method enables the extraction of lifestyle patterns solely from HEMS data, allowing for household-specific appliance control considering the lifestyle of residents. This contributes to future energy-saving measures in the household sector.