Energy and Buildings 37 (2005) 11–22 HVAC system optimization—in-building section Lu Lu, Wenjian Cai ∗ , Lihua Xie, Shujiang Li, Yeng Chai Soh School of Electrical and Electronic Engineering, Nanyang Technological University, Nanyang Avenue, Singapore 639798, Singapore Received 7 June 2003; received in revised form 25 September 2003; accepted 15 December 2003 Abstract This paper presents a practical method to optimize in-building section of centralized Heating, Ventilation and Air-conditioning (HVAC) systems which consist of indoor air loops and chilled water loops. First, through component characteristic analysis, mathematical models associated with cooling loads and energy consumption for heat exchangers and energy consuming devices are established. By considering variation of cooling load of each end user, adaptive neuro-fuzzy inference system (ANFIS) is employed to model duct and pipe networks and obtain optimal differential pressure (DP) set points based on limited sensor information. A mix-integer nonlinear constraint optimization of system energy is formulated and solved by a modified genetic algorithm. The main feature of our paper is a systematic approach in optimizing the overall system energy consumption rather than that of individual component. A simulation study for a typical centralized HVAC system is provided to compare the proposed optimization method with traditional ones. The results show that the proposed method indeed improves the system performance significantly. © 2004 Elsevier B.V. All rights reserved. Keywords: HVAC system; Optimization; Energy conservation; Simulation 1. Introduction A typical centralized HVAC system is shown as in Fig. 1, which comprises two sections: in-building section and out-building section. It can be further divided into five loops: indoor air loops, chilled water loops, refrigerant loops, condenser water loops and outdoor air loops. The in-building section consists of indoor air loop, chilled water loop and part of refrigerant loop. Indoor air loop includes terminal units, cooling coils, dampers, fans, ducts, and controls. Chilled water loop includes cooling coils, chiller evaporators, pumps, pipes, valves, and controls [1]. In terms of energy consumption, the components of in-building sec- tion account for large portion of total energy used in HVAC systems. A small increase in operating efficiency can result in substantial energy savings. In practice, however, optimal operation for such a system is not an easy task as there are thousands of rooms and hundreds of cooling coils in a large-scale HVAC system and all these components are closely coupled. Targeted at energy conservation, there have been many research works reported either for individual component ef- ficiencies or part of system efficiencies. For the cooling ∗ Corresponding author. Tel.: +65-67906862; fax: +65-67905471. E-mail address: ewjcai@ntu.edu.sg (W. Cai). coil model, Stoecker [2] provided a model with many em- pirical parameters under the assumptions of constant air- flow and water flow. Unfortunately, these assumptions are no longer valid in modern HVAC systems. Braun [3] and Rabehl [4] gave their cooling coil models through detailed analysis, unfortunately, both models are somewhat compli- cated and iterative computations are required. The energy saving potential using variable speed drive (VSD) pumps in chilled water loop is an attractive subject which has at- tracted many researchers’ interests [5–7]. Note that these works only considered the individual element without link- ing it to the whole system energy consumption. The effi- ciencies of pumps and fans were also studied in [8,9], where efficiencies of pumps and fans and required pump heads are assumed to be constants which are approximations for con- stant speed fans and pumps and fixed DP controls, respec- tively. If VSD pumps/fans and variable pump/fan pressure set points are used, the total pump/fan efficiencies may vary from 80 to 40%. For the duct and pipe networks, some researchers only considered simple systems and some considered all the cool- ing coils with the same cooling loads simultaneously, which is not true in practice. House and Smith [10] studied opti- mization of two-zone variable air volume (VAV) heating sys- tem by traditional derivative-based methods, which would become very complicated if the number of zones is more than two. Assuming all the cooling loads of coils were 0378-7788/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.enbuild.2003.12.007