Konferensartikel

Commonly used deterministic methods are unable to capture the randomness in occupant behavior and its impact on electric power consumption. In this paper, we propose a new data-driven model to capture occupant behavior in a stochastic manner. Unlike existing models and prediction tools, this new model does not require occupant presence data and can learn occupants’ arrival and departure time based on lighting power consumption data, which is more readily available than occupant presence data. We applied this occupant behavior model to lighting power consumption prediction and implemented the entire prediction process in Modelica. We then validated the Modelica model by comparing the predicted daily, weekly and monthly peak lighting power with measurements from two small commercial buildings. The results suggest that the prediction matches the measurement within acceptable deviations of 7%. The results also indicate that the proposed stochastic model performs better for long-term prediction of lighting power (monthly and weekly) than the short-term (daily).

Occupant behavior modeling, occupant presence prediction, lighting power prediction, regression model, stochastic simulation