Energies 2015,8 11004 comfort in a commercial building,the process of inputting the data into these tools needs to be totally automated in order to facilitate the tool's operation. 3.3.Other Techniques In addition to ABM and MAS,some researchers have proposed other models and techniques aimed at simulating occupants'energy-related characteristics.Yamada et al.[97]developed a system that combines neural networks,fuzzy systems,and predictive control in order to control air-condition systems.Their system can predict the number of occupants in order to estimate building performance to achieve energy savings and high comfort levels for indoor conditions.However,neural network-and fuzzy system-based models typically need a training process,and for Yamada et al.'s [97]developed tool,this training process needs a considerable amount of time.Their proposed system therefore needs to be improved in its training level.Yamada et al.[97]also considered only the temperature as an indicator for comfort level.Such works on comfort level may consider other aspects of indoor comfort, such as humidity and air speed.Wang et al.[98]proposed a Markov chain-based model for building-occupancy simulations in commercial buildings;the model can simulate occupants'stochastic movements in order to predict each occupant's location.It can also produce nonsynchronous occupants' location-changes according to the time and distribution of occupants in space;such predictions become inputs for building management processes for energy savings.However,they validated the model by single offices,which is problematic since for such studies,more cases-especially multiple offices-need to be considered to study occupants'stochastic movements.Jazizadeh et al.[82,83]developed a framework that models occupants'thermal preference profiles into HVAC control logic in order to set room conditions at occupants'desired temperatures.They employed a fuzzy based model to put occupants'comfort profiles into the framework.The results from their test bed of a university building showed up to a 40 percent reduction in HVAC daily average airflow.However,similar to Dounis and Caraiscos [93],they did not clearly respond to the balance between thermal comfort and energy conversation,which is important since achieving a level of thermal comfort might lead to increasing total energy consumption of a building.Zhao et al.[99]developed a practical data-mining approach that collects the energy consumption data of various systems and appliances within office spaces to find occupants'passive energy behaviors.The proposed data-mining approach is based on nominal classification(ie.,C4.5 decision tree,locally weighted naive bayes,and support vector machine)and numeric regression algorithms (i.e., linear regression and support vector regression).The approach has the capability to separately find the behaviors of individual occupants and the schedule of an occupant groups and use this information to set various office appliances and systems in order to reduce the energy consumption.However,the validity of their proposed data-mining approach was limited to data that may have included some incorrect outcomes;such data-mining models require a considerable sample of validated data to test the models and show their effectiveness.Hong et al.[18]presented a framework,DNAs,to observe and simulate occupant energy use behaviors in built environments.This framework is developed based on four key components:(a)drivers of occupants'energy-related behaviors;(b)needs of occupants,(c)actions carried out by occupants;and (d)building's systems acted on by occupants.Such occupancy components directly and indirectly influence building's energy consumption,and therefore DNAs provide theEnergies 2015, 8 11004 comfort in a commercial building, the process of inputting the data into these tools needs to be totally automated in order to facilitate the tool’s operation. 3.3. Other Techniques In addition to ABM and MAS, some researchers have proposed other models and techniques aimed at simulating occupants’ energy-related characteristics. Yamada et al. [97] developed a system that combines neural networks, fuzzy systems, and predictive control in order to control air-condition systems. Their system can predict the number of occupants in order to estimate building performance to achieve energy savings and high comfort levels for indoor conditions. However, neural network- and fuzzy system-based models typically need a training process, and for Yamada et al.’s [97] developed tool, this training process needs a considerable amount of time. Their proposed system therefore needs to be improved in its training level. Yamada et al. [97] also considered only the temperature as an indicator for comfort level. Such works on comfort level may consider other aspects of indoor comfort, such as humidity and air speed. Wang et al. [98] proposed a Markov chain-based model for building-occupancy simulations in commercial buildings; the model can simulate occupants’ stochastic movements in order to predict each occupant’s location. It can also produce nonsynchronous occupants’ location-changes according to the time and distribution of occupants in space; such predictions become inputs for building management processes for energy savings. However, they validated the model by single offices, which is problematic since for such studies, more cases—especially multiple offices—need to be considered to study occupants’ stochastic movements. Jazizadeh et al. [82,83] developed a framework that models occupants’ thermal preference profiles into HVAC control logic in order to set room conditions at occupants’ desired temperatures. They employed a fuzzy based model to put occupants’ comfort profiles into the framework. The results from their test bed of a university building showed up to a 40 percent reduction in HVAC daily average airflow. However, similar to Dounis and Caraiscos [93], they did not clearly respond to the balance between thermal comfort and energy conversation, which is important since achieving a level of thermal comfort might lead to increasing total energy consumption of a building. Zhao et al. [99] developed a practical data-mining approach that collects the energy consumption data of various systems and appliances within office spaces to find occupants’ passive energy behaviors. The proposed data-mining approach is based on nominal classification (i.e., C4.5 decision tree, locally weighted naïve bayes, and support vector machine) and numeric regression algorithms (i.e., linear regression and support vector regression). The approach has the capability to separately find the behaviors of individual occupants and the schedule of an occupant groups and use this information to set various office appliances and systems in order to reduce the energy consumption. However, the validity of their proposed data-mining approach was limited to data that may have included some incorrect outcomes; such data-mining models require a considerable sample of validated data to test the models and show their effectiveness. Hong et al. [18] presented a framework, DNAs, to observe and simulate occupant energy use behaviors in built environments. This framework is developed based on four key components: (a) drivers of occupants’ energy-related behaviors; (b) needs of occupants, (c) actions carried out by occupants; and (d) building’s systems acted on by occupants. Such occupancy components directly and indirectly influence building’s energy consumption, and therefore DNAs provide the