A novel AC turning on behavior model based on survival analysis

Occupant control behavior is a key factor affecting the energy consumption of building air-conditioners(ACs).The operating behavior of ACs and their models in office buildings have been investigated extensively.However,although the thermal sensation of occupants is affected by their previous thermal experience,few researchers have attempted to incorporate this effect quantitatively in models of AC turning on behavior.Not considering the cumulative effect may result in inaccurate predictions.Therefore,in this study,a survival model is proposed to describe AC turning on behavior in office buildings under the cumulative dimension of time.Based on a dataset containing environmental parameters and occupant behavior information,as well as considering occupants entering a room as the starting event and turning on an air-conditioner as the end event,the endurance time before an AC is turned on is investigated,and a survival model is used to predict the probability of the AC turning on due to environmental factors.Based on a switch curve,confusion matrix,and tolerance–time curve,the prediction results of the survival model are analyzed and validated.The results show that a tolerance temperature of 29℃and a tolerance duration setting of 1 h can effectively model the turning on behavior of the AC.In addition,based on comparison results of different models,the survival model presents a more stable switching curve,a higher F1 score,and a tolerance curve that is more similar to reality.Different tolerance durations,as well as static and dynamic tolerance temperature settings,are considered to optimize the model.Furthermore,the AC energy consumption is calculated under the survival model and the traditional Weibull model.Simulation results were compared with measurement,and the survival model verified the improvement effect of prediction accuracy by 8%than the Weibull model.By considering the time-transformed accumulation of physical environmental factors,the accuracy of AC turning on models can be improved,thus providing an effective reference for future building energy consumption simulations.