**1. Introduction**

Air conditioning, lighting, and ventilation systems constitute approximately 50% of the total power consumption of a building. In particular, air conditioning systems have a considerable e ffect on energy consumption; hence, energy conservation control mechanisms for air conditioning systems are crucial [1]. Air conditioning systems in a mass building are typically centralized and comprise fan coil units (FCUs) in each area of the building and a chilled water system in the mechanical room [2]. The chilled water system provides chilled water to the cooling coils of the FCUs, and the water in the coils is subsequently converted to cold air that is distributed to each area of the building. In general, because of the change of seasons, chilled water systems are designed on the basis of the maximum load in summer. Therefore, such systems often operate under a partial load ratio, leading to energy waste due to poor operating e fficiency. The coe fficient of performance of a chilled water system is a critical parameter in evaluating the performance of chilled water systems. A higher coe fficient signifies higher chilled water system performance. Chow et al. [3] proposed using a neural network and genetic algorithm optimization to enhance the operating e fficiency of a chilled water system to achieve energy conservation. Browne and Bansal [4] used a regression model to predict the performance of a chilled water system to optimize the control of the system. Numerous studies have focused on e fficiency optimization control strategies for chilled water systems and cooling towers. For example, Yu and Chan [5] used load-based speed control to determine the optimal control frequency for cooling towers and chilled water pumps to optimize the operating e fficiency of chilled water systems. In addition to achieving optimal e fficiency for chilled water systems (by controlling their operation), e ffectively controlling chilled water systems (by using indoor heat demand responses) is another means of achieving cost reduction. Yoon et al. [6] proposed the use of a dynamic demand response controller to change the temperature settings of a chilled water system and control its loading during periods of peak power consumption.

Wang et al. [7] revealed that thermal comfort is a crucial factor influencing the control of heating, ventilation, and air conditioning (HVAC) systems because a user's thermal comfort demand directly affects HVAC system control. Therefore, e ffective control of indoor thermal comfort temperature contributes to the maintenance of environmental comfort and the reduction of FCU energy waste. Specifically, an optimization control strategy to balance between HVAC energy consumption and indoor thermal comfort is valuable. The most commonly cited experiments on the human perception of thermal comfort have been performed by Fanger (1982). His analysis indicated the sensation of human thermal comfort by six factors: Four physical variables, including air temperature, mean radiant temperature, air velocity, and air humidity, and two personal variables, including metabolic rate and clothing insulation. Fanger and Toftum [8] proposed the PMV model, which agrees well with high-quality field studies in buildings with HVAC systems, situated in cold, temperature, and warm climates, studied during both summer and winter. Model predictive control (MPC) is comprehensively used to determine indoor HVAC temperature settings and comfortable temperature values by employing optimal control methods to reduce energy consumption and maintain indoor thermal comfort. Castilla et al. [9] proposed a hierarchical control method that entails the use of a non-linear MPC strategy to predict control. The purpose of this method is to maintain indoor comfort while preventing an HVAC system from entering high-load operation and to optimize e fficiency during low-load operation. Jazizadeh et al. [10] applied a fuzzy control system for predicting indoor thermal comfort to e ffectively reduce the daily mean flow of an HVAC system. Chen et al. [11,12] has proposed two MPC systems for determining the optimal solution to balance indoor thermal comfort and chilled water system control; moreover, a model integrating MPC and dynamic thermal sensation could maintain indoor thermal comfort while consuming energy at a rate lower than that of the conventional MPC–PMV model. Human perception of thermal comfort generally varies with climate. Oldewurtel et al. [13] applied a stochastic MPC system for predicting the change in climate to calculate the minimum energy required to achieve indoor thermal comfort. Fong [14] indicated that indoor thermal comfort is not only a perceived response of the human body to environmental comfort but also a demand response (DR) to the cooling ability of HVAC systems. Lin et al. [15] applied an algorithm to solve optimal chiller loading (OCL) problems. ASHRAE [16] has developed an industry consensus standard to describe comfort requirements in buildings. The standard is known as ASHRAE Standard 55-2004 Thermal Environmental Conditions for Human Occupancy. The purpose of this standard is to specify the combinations of indoor thermal environmental factors and personal factors that will produce thermal environmental conditions acceptable to a majority of the occupants within the space. Brager and Dear [17] have proposed the distinction between thermal comfort response in air-conditioned and naturally ventilated buildings and suggested that behavioral adaptation incorporated in conventional heat balance models could only partially explain these di fferences and that comfort was significantly influenced by people's expectations of the thermal environment. Brager and Dear [18] have presented the relationship between optimum temperatures and prevailing indoor/outdoor temperatures in the study demonstrated that adaptation is at work in buildings with centralized HVAC. The results were interpreted to indicate that occupants of HVAC

buildings had become finely tuned to the very narrow range of indoor temperatures being presented by current HVAC practice. However, there is potentially a very high energy cost to maintaining those narrowly defined comfortable thermal conditions. Humphreys and Nicol [19] presented an adaptive model for thermal comfort, and used a fuzzy logic system to fine tune the room temperature. Dear et al. [20] explained all fields of applied research; it is important that the thermal environment of simulations are regularly 'ground-truthed' with real comfort assessments from human subjects in either chamber studies or 'real' building occupants in field studies. Thermal comfort evaluations by human subjects are a superior contribution to knowledge, having longer-lasting value to the research community than simulated comfort evaluations coming out of a comfort model. Alfano et al. [21] pointed out that HVAC engineers are, in practice, interested in the air temperature value required to ensure thermal comfort rather than in a comfort index.

Many studies explore the use of PID controllers to control the energy-saving control of HVAC systems with optimized algorithms or to use climatic factors to predict the RT requirements of HVAC systems [22]. A few papers combine HVAC system control with indoor comfort requirements, Kampelis et al. [23] proposed the Daily Discomfort Score (DDS) to assess demand response HVAC control and thermal comfort in a university building. Most of the abovementioned documents are optimized for ice water systems, or use the simulation method to estimate the comfort of the indoor ice-water system energy consumption comparison and prediction, but do not combine the indoor real heat demand and HVAC power consumption. This paper aims to solve the optimal control strategy of the chilled water system of HVAC buildings and approach the request of the occupants for indoor thermal comfort. Occupants' adjustments of temperature settings are used as feedback to determine the actual refrigeration ton (RT) demand for a chilled water system and a fuzzy logic system continually to determine the cooling air demand for a chilled water system. Finally, the operating point associated with minimum energy consumption per unit RT is derived using a genetic algorithm. The e ffectiveness of this energy conservation algorithm was tested through a field experiment.
