A Study On Variable Time Step Models For Real Time Building Occupancy Prediction

Predictions of occupancy and occupant behavior are important to optimize the control of building services systems and provide a basis for building energy benchmarking.Existing occupancy prediction models primarily simulate the occupancy behavior in fixed time steps,which cause difficulties to satisfy prediction accuracy and simulation efficiency simultaneously.Specifically,if the time step size is too small,there may exist redundant simulations and thus the simulation efficiency is reduced,especially in periods where is no occupancy change;if the time step size is too large,some important information may be ignored and further the prediction accuracy is reduced,especially in periods where the occupancy change occurs frequently.Meanwhile,the impact of the occupant behavior on building energy use was seldom taken into consideration in existing building energy benchmarking stud ies.This may result in an inaccurate benchmarking value and cause a poor building energy performance evaluation.To address above issues,two novel methods were proposed.One of two methods is a novel occupancy prediction method with variable time steps.In this method,variable-step simulation nodes were firstly selected based on the door opening and closing events.Then a joint probability model was established that combines simulation nodes with indoor CO₂ concentration to predict the number of occupants.The other method is a novel building energy benchmarking methodology based on an agent-based model.The primary objective of the agent-based model is to generate occupant behavior profiles under the premise of satisfying comfort and achieving a significant energy saving to improve building performance.Then,the simulated occupancy behavior profiles were converted to energy consumption data,which was utilized as the benchmarking value to further evaluate the building's energy-saving potential.To validate the proposed occupancy prediction method,it was applied to the measured data and compared with the IHMM,ANN,SVM and CART models.Meanwhile,the proposed building energy benchmarking methodology is applied for a typical household selected from the Japanese residential building database.The main conclusions are summarized as follows:?1?The proposed occupancy model performs better than the IHMM,ANN,SVM and CART models,as its prediction accuracy is higher than other models around 11%and the number of simulation runs is reduced from 1440 to 363.This indicates that the model can increase the prediction accuracy while improve the simulation ef ficiency.?2?The proposed occupancy prediction method can accurately predict occupants'first arrival time and last departure time which are important to the operation and control optimization of HVAC system.?3?Based on the proposed novel building energy benchmarking methodology,the calculated energy-saving potential of HVAC and lighting systems are 4.08 kWh and0.14 kWh,respectively.It is suggested that future research should put more effort into two factors:First,incorporate other sensors?i.e.PIR sensors?to improve the model prediction accuracy in occupancy state.Second,develop a more systematic method with the interaction between different occupants being considered to provide a more accurate building energy benchmarking value.