A Resident-differentiated Model For Prediction Of Occupancy And Occupant Activity In Residential Buildings

Energy-related behavior(especially occupancy status and occupant activity)is one of major factors that significantly influence energy consumption in residential buildings.Therefore,accurately modeling energy-related behaviors can enhance the building simulation performance.In previous studies,due to the lack of considering household diversity,current occupant behavior modeling methods showed relatively poor prediction accuracy and limited applicability.Meanwhile,these studies did not fully consider the relationship among different energy-related behaviors,leading to significant gap between actual and prediction behavior profiles.To address these issues,this study aims at developing suitable methods to realistically predict the occupancy behavior and occupant activity by accounting for household diversity and associations between different behaviors.For occupancy prediction,spearman correlation is firstly applied to identify the relationship between main household features and total occupancy duration.The correlation coefficients are subsequently used weights for features so as to identify typical household groups.Finally,Markov chain(MC)models are developed for identified household groups.For energy-related behavior prediction,a two-stage clustering analysis is firstly performed to identify the main integrated behavior patterns and specific characteristics of nine different energy-related behaviors.Based on the clustered behavioral patterns,a CART decision tree model is applied to reveal the relationship between household characteristics and various behavioral patterns.This model is also used to assign target behavioral pattern labels for new households according to their characteristics.A set of MC models are created for predicting energy-related behavior separately and summed by the created combination rules with the consideration of behavioral associations.To validate the performance of the proposed methods,Time-use-survey(TUS)data in England were used in this study and their performance were compared with conventional MC models.The main conclusions are summarized as follows:(1)The Spearman correlation analysis results showed strong relationship between household diversity and occupancy status.Accordingly,it is necessary to consider such relationship in occupancy behavior modeling.Influencing factors that have relatively high correlation include employment status,income,age and household identity.(2)The influence of household diversity on occupancy was considered comprehensively,and the distinct occupancy profiles pattern were identified.Compared with conventional MC models,the predictive accuracy of the proposed occupancy prediction method improved significantly.The mean absolute error(MAE)and root mean squared error(RMSE)decrease 20.57% and 15.35%,respectively.Such results demonstrated that the importance of considering household diversity and the proposed method can generate more realistic occupancy profile through selecting appropriate household characteristics.(3)The proposed two-stage clustering method can effectively identify typical integrated behavioral patterns and specific characteristics of nine energy-related behaviors,e.g.,total duration of different behavioral states,occurrence probability and time period with peak occurrence probability.Therefore,the proposed method can not only obtain the households with similar integrated behavior pattern but also differentiate the variation characteristics of nine different behaviors.Based on the results of the CART model,different categories of energy-related behavior(such as going outside,sleeping,washing dishes,washing clothes,watching TV,use of video)have strong association with household characteristics and the classification accuracy is greater than 80%.(4)The proposed modeling method for energy-related behavior prediction can reasonably characterize the probability distribution of integrated behavior pattern and accurately capture the peak variation.Compared with conventional MC models,the MAE and RMSE for integrated behavior prediction decrease 24.17% and 22.97%,respectively.At the same time,the proposed method can improve the predictive accuracy of seven energy-related behaviors,except washing dishes and washing clothes.Particularly,the behavior with maximum improvement is going outside,which reaches at 9.2%.While for washing dishes and washing clothes,the prediction accuracy is almost the same with conventional MC models.