Anda di halaman 1dari 12

Energy and Buildings 38 (2006) 220231 www.elsevier.

com/locate/enbuild

HVAC system optimization for energy management by evolutionary programming


K.F. Fong a,*, V.I. Hanby b, T.T. Chow a
a

Division of Building Science and Technology, City University of Hong Kong, Tat Chee Avenue, Kowloon Tong, Kowloon, Hong Kong, China b Institute of Energy and Sustainable Development, De Montfort University, The Gateway, Leicester LE1 9BH, UK

Received 30 December 2004; received in revised form 5 May 2005; accepted 10 May 2005

Abstract Energy management of heating, ventilating and air-conditioning (HVAC) systems is a primary concern in building projects, since the energy consumption in electricity has the highest percentage in HVAC among all building services installations and electric appliances. Without sacrice of thermal comfort, to reset the suitable operating parameters, such as the chilled water temperature and supply air temperature, would have energy saving with immediate effect. For the typical commercial building projects, it is not difcult to acquire the reference settings for efcient operation. However, for some special projects, due to the specic design and control of the HVAC system, conventional settings may not be necessarily energy-efcient in daily operation. In this paper, the simulation-optimization approach was proposed for the effective energy management of HVAC system. Due to the complicated interrelationship of the entire HVAC system, which commonly includes the water side and air side systems, it is necessary to suggest optimum settings for different operations in response to the dynamic cooling loads and changing weather conditions throughout a year. A metaheuristic simulationEP (evolutionary programming) coupling approach was developed using evolutionary programming, which can effectively handle the discrete, non-linear and highly constrained optimization problems, such as those related to HVAC systems. The effectiveness of this simulationEP coupling suite was demonstrated through the establishment of a monthly optimum reset scheme for both the chilled water and supply air temperatures of the HVAC installations of a local project. This reset scheme would have a saving potential of about 7% as compared to the existing operational settings, without any extra cost. # 2005 Elsevier B.V. All rights reserved.
Keywords: Evolutionary programming; Evolutionary algorithm; Optimization; Energy management; HVAC system

1. Introduction Hong Kong is situated in the subtropical region, and the monthly average temperature ranges from 15 to 27 8C throughout a year. According to the statistics from the Hong Kong SAR government, 32% electricity was used for space air-conditioning in 2000 [1]. Currently, the Hong Kong SAR government has launched a set of energy efciency codes to promote the energy efciency requirements and address local sustainability, including those for air-conditioning [2,3]. However, these codes are mainly used to provide the
* Corresponding author. Tel.: +852 2788 8724; fax: +852 2788 9716. E-mail address: bssquare@cityu.edu.hk (K.F. Fong). 0378-7788/$ see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.enbuild.2005.05.008

minimum design requirements for the newly designed buildings. For existing installations of heating, ventilating and air-conditioning (HVAC) systems, little local references about energy efciency are available. In this case the required operating strategies and optimum conditions of different HVAC equipment and sub-systems, international standards and guidelines for operation and energy management may be used. With regard to the strategic facilities management of HVAC system, one of the major foci is on effective and efcient energy management. It is essential to identify the suitable energy management opportunities among a variety of measures in order to achieve the energy saving targets. These measures may cost no extra investment and

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

221

Nomenclature COP coefcient of performance COPnom coefcient of performance at cooling capacity COPrate coefcient of performance at rated cooling capacity COPratio ratio of COPnom to COPrate Cpa specic heat capacity of moist air at constant pressure (kJ/(kg K)) Cpcw specic heat capacity of sea water at constant pressure (kJ/(kg K)) Cpw specic heat capacity of chilled water at constant pressure (kJ/(kg K)) Cs average slope of saturation air enthalpy versus temperature (kJ/(kg K)) f p fraction of full-load power G cooling capacity of chiller (kJ/kg) Grate rated cooling capacity of chiller (kJ/kg) Gratio ratio of G to Grate ha specic enthalpy of moist air (kJ/kg) hfg latent heat of vapourization for water (kJ/kg) ho specic enthalpy of outdoor air (kJ/kg) hs specic enthalpy of saturated air (kJ/kg) hspace specic enthalpy of space air (kJ/kg) ma mass ow rate of dry air (kg/s) mcw mass ow rate of cooling water (kg/s) mw mass ow rate of chilled water (kg/s) m* ratio of air to water effective capacitance rate for wet analysis NTU overall number of transfer units P chiller power (kW) Pnom nominal chiller power (kW) Qcc total cooling coil heat transfer rate (kW) Ql latent heat transfer rate of cooling coil (kW) Qload cooling load (kJ/kg) Qmet cooling load met by chiller (kJ/kg) Qrej heat rejection rate (kW) Qs sensible heat transfer rate of cooling coil (kW) Qwet heat transfer rate of completely wet coil (kW) rpl part load ratio Ta air temperature (8C) Tcw cooling water temperature (8C) Ts surface temperature (8C) Tsa supply air temperature (8C) Tchw chilled water temperature (8C) UA overall heat conductance (kW/K) ewet effectiveness of completely wet coil Subscripts a air stream conditions e effective i inlet or inside conditions o outlet or outside conditions

s sp w wet

surface conditions set point water stream conditions wet surface

installation, particularly using the appropriate reset schemes in the operation of the existing equipment of the HVAC systems. Although the automatic control provisions and even the computerized building management systems have been provided for both the full and part load operations in typical HVAC installations, control commissioning mainly covers full load settings, but not necessarily part load situations. The operating capacities of the major equipment would be overprovided in many occasions, and energy management cannot be achieved effectively. Even if the operation engineering staff tries to implement control reset in different seasons and part load conditions, the appropriate settings would inevitably be developed through the trial-and-error approach throughout seasons and years. It is necessary to have a proper predictive method to provide the optimum information of the related parameters, and the reset control can be really implemented as a primary energy management opportunity for effective facilities management in HVAC, as well as the other building services systems. It is useful to consider the simulation-optimization approach in response to the problem of effective energy management. In the context of the building services and HVAC systems, the simulation-optimization approach has been developed in the recent decade. A variety of objective functions have been developed according to the nature of problems, like the year-round energy consumption, lifecycle cost, thermal comfort, compromise among parameters, plant scheduling, design parameters, routing and distribution, as reected from the works of Wright [4], KintnerMeyer [5], Huh [6], deWit [7], Taylor [8], and Fong et al. [9], respectively. In addition, there are increasing applications of the plant simulation models for decision making purpose, such as the works of Dasgupta [10] and Wright et al. [11], a multi-criterion optimization approach would be adopted and a number of possible scenarios instead of a single solution would be provided for decision making. This paper describes a novel simulation-optimization approach that was applied to devise a reset scheme of chilled water and supply air temperatures for the HVAC system in a local subway station. The results from the proposed simulation-optimization approach were used to evaluate the existing operational settings, and suggest the possible saving potential from the newly optimized information.

2. Plant simulation for entire HVAC system In the previous research works, the component-based plant model is seldom applied in full scale for all the

222

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

Fig. 1. HVAC schematic diagram and simulation model under study.

water and sea water pumps. Differential pressure bypass was applied as a hydronic coupling during the part load operation. For the air side system, there were a number of AHUs and fans in the actual installations. In order to simplify the model and hence save simulation time, particularly in the optimization process, it was assumed that the same kind of air side equipment ran in the same operating mode together. Therefore, it was lumped into one AHU, one outdoor air fan, one return air fan and one free cooling fan. The power rating of different equipment is summarized in Table 1. For the local practice of a typical commercial building project, the percentage allocation of power consumption between water side system and air side system is usually about 8020% for the similar type of chillers, air side system and sea water cooling provision. However, from Table 1, the percentage allocation is about 6040%, and it is worthwhile to pay more attention to the operational and control effect of the air side system on the overall energy consumption. 2.2. Non-linearity and highly constrained nature of HVAC system For the HVAC system, the inherent nature of nonlinearity can be shown by the non-linear algebraic equations involved in the mathematical descriptions. In this study, the simulation program TRNSYS was used to develop the HVAC system model. Among a variety of simulation components already adopted, those of the chiller and cooling coil are the most complex. It is important to understand the details of these components as represented in the simulation software. 2.2.1. Mathematical descriptions of water-cooled chiller For the water-cooled chiller component, it modelled the thermal performance and power requirements of a vapour compression refrigeration machine [12]. The power was a function of the chilled water supply temperature set point and cooling water entering temperature. Two external data les were required to determine the chiller performance. The rst data le consisted of the ratios of the capacity to the

subsystems of the entire HVAC system. However, in view of the interrelationship among all these subsystems, a holistic approach should be applied and a plant model of the entire HVAC system is required in order to truly simulate the effects of a variety of operating parameters and factors. 2.1. Component-based plant simulation model of entire HVAC system A component-based model of the entire HVAC system was built up with TRNSYS, together with the required control and operating functions in order to reect the yearround operation of different equipment within the subsystems. This plant simulation model was applied in a local subway station, which needed a suitable reset scheme of the supply air temperature for the AHUs inside the platform and concourse, as well as that of the chilled water supply temperature of the chillers. The schematic diagram of the HVAC system and the respective plant component models of the subway station is shown in Fig. 1. There were two chillers, each with cooling capacity of 1820 kW (518 TR), together with the associated chilled
Table 1 Equipment power rating of the HVAC system Item 1 2 3 4 5 6 7 8 9 10 Equipment Water-cooled chiller 01 Water-cooled chiller 02 Chilled water pump 01 Chilled water pump 02 Sea water pump 01 Sea water pump 02 AHU fan (lumped) Outdoor air fan (lumped) Return air fan (lumped) Free cooling fan (lumped) Total Power consumption (kW) 348.66 348.66 30 30 22 22 222 41.55 163 112.5 1340.37

Relative percentage (%) 26.01 26.01 2.24 2.24 1.64 1.64 16.56 3.10 12.16 8.39 100

Sub-total percentage Total 59.8% for the water side system (52.02% for chillers)

Total 40.2% for the air side system

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

223

rated capacity and the coefcient of performance (COP) to the rated COP, as functions of the chilled water temperature set point and the cooling water entering temperature. The second data le contained the fraction of full load power as a function of the partial load ratio of the chiller. The mathematical expressions to determine the major parameters, such as the chiller power consumption, cooling load met, heat rejection rate, chilled water supply and return temperatures, are summarized in Eqs. (1)(11). COPnom COPrate COPratio G Grate Gratio Pnom G COPnom (1) (2) (3)

Eqs. (12)(22). Qwet ewet; a ma ha; i hs; w; i ewet;a m 1 eNTUwet 1m 1 m eNTUwet 1m ma Cs mw Cpw UAwet ma

(12) (13) (14)

NTUwet Cs

(15) (16) (17) (18) (19) (20) (21) (22)

hs; w; o hs; w; i Tw; o Tw; i

ha; o ha; i ewet ha; i hs; w; i (4) Tchw; o Tchw; i (5) Ta; o Ts; e Ta; i Ts; e eUAo =ma Cpa (6) (7) (8) (9) (10) hs; s; e ha; i ha; o ha; i 1 eUAo =ma Cpa ma UAwet ha; i hs; w; o mw Cpw UAi ma ha; i ha; o mw Cpw

Qload mw Cpw Tchw; i Tchw; sp Qload rpl G Qmet minimumQload ; G P Pnom fflp COP Qmet P

Ts; i Tchw; o

Qrej Qmet P Tchw Tchw;i Tcw;o Tcw;i Qmet mw Cpw Qrej mcw Cpcw

Qcc mw Cpw Tchw; o Tchw; i

The average saturation specic heat, Cs, depends on the outlet water temperature and therefore an iterative method was adopted to nd the outlet chilled water temperature. 2.2.3. Dynamic control and operation algorithm In the development of the HVAC plant simulation model, just linking up all the related components of equipment, pipework and ductwork could not simulate the real plant operation in different loading and climatic conditions throughout a year. Additional operation and control input was necessary, so that the operating quantity and capacity of the equipment, as well as the corresponding ow conditions at different portions of pipework and ductwork, could be effectively converged through the successive iteration processes. In order to provide such essential information to the plant model, a generic component for setting up control expressions was incorporated for different operation scenarios of the equipment on the TRNSYS platform. The whole plant model could then truly reect the HVAC system with complete dynamic operation of different equipment and subsystems for the uctuating cooling demand throughout the yearly operating hours. In the hourly control and operation of different equipment and subsystems, there were four sets of input data les: total and space loads, weather conditions, sea water temperature and that specically for the component model of water-cooled chiller. For the water side system, the total cooling load was used to determine the number of chillers, in turn the numbers

(11)

2.2.2. Mathematical descriptions of cooling coil This component modelled the performance of a dehumidifying cooling coil, using the air effectiveness model outlined by Braun [13] and the efciencies of rectangular-plate ns from Threlkeld [14]. For the air cooled and dehumidied through a coiling coil, the limiting exit state would be that the air were saturated at a temperature equal to that of the incoming chilled water stream. This corresponds to the minimum possible enthalpy of the exit air. The air side heat transfer effectiveness was dened as the ratio of the air enthalpy difference to the maximum possible air enthalpy difference if the exit air were at the minimum possible enthalpy. Fin efciencies were required in order to calculate heat transfer coefcients between the air stream and the coil. For the hot and humid climatic conditions in Hong Kong, usually the coil surface temperature at the air inlet is less than the dewpoint of the incoming air, then the coil is completely wet and dehumidication occurs throughout the coil. Therefore, the wet analysis was applied for simulation and the major mathematical descriptions are summarized in

224

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

of the chilled water and sea water pumps in operation. The power consumption of chillers would be determined by COP, which was determined by the chilled water and sea water entering temperatures, as well as the fraction of full load power. For the air side system, based on the space sensible and latent loads together with the supply air temperature set point, the space air temperature and humidity ratio were determined. This information was useful to determine whether the full speed or half speed operation was applied to the AHU fan, outdoor air fan and return air fan. In addition, enthalpy control was utilized to decide the implementation of free cooling mode, which would call for the free cooling fan in operation, with the same speed of the other operating fans. This dynamic control and operation algorithm was essential to provide the necessary system response to different loading and climatic conditions, although this would increase the degree of the non-linear, constrained and discrete nature of the system performance, in particular to the energy consumption of both the water side and air side systems. The HVAC optimization problem hence requires an effective method that could handle the non-linear, discrete and highly constrained characteristics. 2.3. Review of existing performance of the HVAC system For the existing HVAC system of the subway station project, no energy meter was installed for individual equipment and only the overall energy consumption was logged. Through the establishment of the whole HVAC plant model, it could have better understanding of the system, and perceive the opportunities of energy management. The plant model was validated in a similar application [15]. In the existing operating conditions, the chilled water supply temperature was 7 8C and the supply air temperature was 16 8C. Therefore, the monthly energy consumptions of the entire HVAC system, as well as the water side and air side sub-systems, are illustrated in Fig. 2. The variation of the total energy consumption prole is similar to the typical one for local situation, with the peak within July to August during the hot and humid summer, and

the trough within December to February in the mild cold winter. However, from the separate proles of the water side system and the air side system, the latter plays a crucial role in contributing the energy consumption during both the cold and mid-seasons. This is related to the high relative energy requirement of the air side system as mentioned in Table 1, and further investigation through appropriate optimization methodology is necessary.

3. Evolutionary programming for optimization 3.1. Development of optimization methodologies for HVAC problems The typical optimization problems of HVAC system have discrete, non-linear and highly constrained search space. Since the last two decades, there has been continuous advancement of the optimization methods for a variety of HVAC and building services problems, so that local optima can be avoided and global optimal can be effectively identied. Wright and Hanby [16] applied the direct search method, but there might be inaccuracy when constraints were encountered. Wright [17] used the successive approximation method for improvement. Hanby and Angelov [18] adopted univariate search method, a gradient-based technique, to carry out plant design optimization based on the dynamic performance from plant simulation. In the recent years, there are growing applications of evolutionary algorithm (EA) in handling different optimization problems. Giraud-Moreau and Lafon [19] have highlighted that EA is very suitable to handle the complex mechanical design problems since no derivative information is required. Simpson et al. [20] applied EA for the optimization of pipework networks; Wright [21] used EA for the HVAC optimization studies in sizing; Huang and Lam [22] adopted EA for optimizing the controller performance in HVAC systems; Sakamoto et al. [23] examined the application of EA to optimize the operation schedule for a district heating and cooling plant; Asiedu et al. [24] focused on the application of EA for the duct

Fig. 2. Monthly energy consumption at the existing operation settings of chilled water and supply air temperatures.

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

225

system design. Wright et al. [12] even applied multiobjective genetic algorithm search method to identify the optimal pay-off characteristic between building energy cost and occupant thermal discomfort. 3.2. Evolutionary algorithm at a glance For EA in general, there are three major paradigms evolution strategy (ES), evolutionary programming (EP) and genetic algorithm (GA) [25,26]. Their major differences lie in the data representation, importance of recombination and mutation, and the approach of selection. Generally for GA, the representation of the individuals is binary, recombination (or crossover) is essential, mutation has less importance and selection is stochastic. For EP, the representation is real-valued, mutation is essential, recombination is not included, and the selection is stochastic. For ES, the representation is also real-valued, both mutation and recombination are important, but the selection is deterministic. For those EA examples mentioned before, many optimization problems were handled by GA [12,2024], but the effectiveness of other paradigms in EA, such as EP and ES, has not been thoroughly studied and reported. In this paper, the paradigm of EP was applied as the optimization method linking with the plant simulation model. EP in effect has similar feature in the stochastic selection as that of GA, but it is not necessary to be encoded in binary representation for processing, since EP can directly handle the real-valued individuals and mutate them with a suitable variation operator. 3.3. Algorithm of evolutionary programming There are four main stages in this EP: initialisation, evaluation, selection and variation. The ow chart of the developed EP algorithm is shown in Fig. 3. 3.3.1. Initialisation In this EP optimization run, the population of individuals in each epoch was set to be 10 and the stopping rule was set at the maximum epoch of 50. At the initialisation stage, i.e. the very beginning of the EP optimization, an initial population of individuals would be generated randomly within the corresponding bounds of the optimized parameters involved. Then this population of individuals would be carried forward to the next stage for evaluation. 3.3.2. Evaluation The developed EP was real-valued, and handled oating point and integer variables. Modularity was the major feature, therefore the optimization model was developed in MATLAB which could effectively handle the parametric matrices across different generations. The plant simulation model would provide the tness values of the parents,

Fig. 3. Algorithm of evolutionary programming.

then direct the results to the EP algorithm for evaluation, selection and variation. The offspring would be then produced for the plant simulation model, and the loop continued until the required stopping rule was satised. 3.3.3. Selection In this study, the stochastic approach was mainly applied based on the paradigm of EP. The population was selected by the roulette wheel, which is a proportionate selection approach, and the probability of selection is directly proportional to the relative tness of the individual within the population. However, in order to maintain the potentially good solution, the elitist individual that had the best tness was selected and carried forward to the population in the next generation. 3.3.4. Variation In this application of EP, a variation factor ni was applied to mutate each element of an individual. This incorporated a Gaussian-distributed random number wi with a mean of zero and unit standard deviation, which has proven to be effective in perturbations and mutation [27]. Therefore, for each mutated element xi, mut from the parent element xi of the selected individual, it was determined in Eqs. (23) and (24) as follows: xi; mut xi ni and ni i eg1=c (24) (23)

where g was the epoch and c was a decay constant, generally set to 15% of the maximum epoch. As a whole, the variation

226

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

factor ni provides a step length of search, which was gradually reduced as the search progressed, causing the population as a whole to converge.

hours throughout a year. Etotal and Etotal; i EWCC; i ECHP; i ESWP; i EAHU; i EOAF; i ERAF; i EFCF; i (26) where N, 6935 h, based on 19-h daily operation of the HVAC plant for 365 days a year. For the optimization in Scenario 2, each individual of a population would include both Tchw, sp and Tsa, sp, for the effect of both set points would be studied. The tness function was Eqs. (25) and (26) again, in which year-round energy consumption would be evaluated. On the other hand, for EP optimization in Scenario 3, each individual would contain both Tchw, sp and Tsa, sp, but the tness function would be based on monthly evaluation, as shown in Eqs. (26) and (27). The energy consumption was found from the operating equipment rst, then the monthly energy consumption was determined for the corresponding monthly operating hours. Etotal
k X Etotal; i i1 N X Etotal; i i1

(25)

4. Problem formulation 4.1. Features of HVAC system under study To illustrate the optimization opportunities for energy management of the HVAC system, the focus should be on the existing installed features of the equipment. Since constant speed chilled water pumps and differential pressure bypass were installed, the reset of chilled water supply temperature Tchw, sp may be more benecial for this type of equipment, instead of for the variable speed pumps. Similarly, 2-speed supply air fans for AHUs were installed in the air side system, the reset of supply air temperature Tsa, sp may be more effective in improving energy efciency, compared to that for the variable speed fans. Wulnghoff [28] suggested that there would be noticeable saving potentials to carry out temperature reset for chilled water temperature, supply air temperature or even both. The amount of saving potential would depend on the complexity and conguration of the HVAC system, and there is no clear guideline of the optimum combination of Tchw, sp and Tsa, sp. Therefore, suitable optimization technique is required for the study in this area. 4.2. Formulation for optimization Based on the existing features of the HVAC system under study, the problem parameters to be optimized included:  chilled water supply temperature of chiller;  supply air temperature of AHU. Three optimization scenarios were designed for the HVAC system: (1) Scenario 1: minimization of year-round energy consumption with the optimum Tchw, sp only. (2) Scenario 2: minimization of year-round energy consumption with both the optimum Tchw, sp and Tsa, sp. (3) Scenario 3: minimization of monthly energy consumption with both the optimum Tchw, sp and Tsa, sp. For EP optimization in Scenario 1, each individual of a population would contain Tchw, sp only. The tness function that was used for evaluation is shown in Eqs. (25) and (26). For each operating hour i, the hourly total energy consumption Etotal, i was determined for all the operating equipment in response to the cooling loads and weather conditions in that hour. Finally the year-round energy consumption Etotal was summed up for all the operating

(27)

where, k = 598 h (31 19) for January, March, May, July, August, October and December; or k = 570 h (30 19) for April, June, September and November; or k = 532 h (28 19) for February only. The constraints on this optimization problem were the inequality expressions as follows:  the feasible range of Tchw, sp, i.e. 5 8C Tchw, sp 8 8C;  the feasible range of Tsa, sp, i.e. 13 8C Tchw, sp 19 8C;  the space air temperature <27 8C.

4.3. Preview of search landscape Before the actual run of the simulation-optimization suite, it would be useful to have some basic understanding about the search landscape. If there were too many optimized parameters, it may not be easy to preview how those parameters inuence the tness value and their interrelationship. In this optimization problem, there were two optimized parameters Tchw, sp and Tsa, sp, so a two-dimensional search surface could be developed with the parametric study within their feasible ranges of operation. For the HVAC system under study, the search landscape of the entire system is plotted in Fig. 4. In order to have a thorough understanding, the parametric surfaces of the water side system and air side system were also developed as shown in Figs. 5 and 6,

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

227

Fig. 4. Search landscape of year-round energy consumption vs. Tchw, sp and Tsa, sp for entire HVAC system.

respectively, but these two surfaces were not directly related to the search process, since only the total energy consumption was included in the tness function. From Fig. 4, it is found that the shape of the search landscape is like a chair, the surface smoothness is fair

and there may not be local minima or maxima. As a minimization problem, the region with Tchw, sp near 5 8C and Tsa, sp near 19 8C would not be the concern, since there are relatively high tness values. With the help of Figs. 5 and 6, there is a better understanding to the resultant landscape of

Fig. 5. Parametric surface of year-round energy consumption vs. Tchw, sp and Tsa, sp for water side system.

228

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

Fig. 6. Parametric surface of year-round energy consumption vs. Tchw, sp and Tsa, sp for air side system.

the entire system as shown in Fig. 4. From Fig. 5, it is advantageous to have higher Tsa, sp in the water side system, and there are two major reasons. The rst reason is that higher chilled water temperature would render to a higher Tsa, sp, in turn the chiller would consume less power due to better COP. The second reason is when Tsa, sp is high during the mid-seasons, there would be more frequent opportunity that the outdoor air enthalpy is less than the space air enthalpy. The air side would then have more chance to operate in free cooling mode, thus the chiller plant and the associated pumps would have more hours out of operation. For the performance of the air side system as shown in Fig. 6, there is a very different effect. As mentioned before, there would be more chance to run in free cooling mode for higher Tsa, sp. However, when the chillers are out of operation, the space air temperature would be less frequently lower than the specied design space temperature, thus the fans would have more opportunity to run in the full speed mode, instead of the half speed mode. Owning to the high relative percent of the energy consumption of the air side equipment as shown in Table 1, there should be a pay-off between the operation of the water side system and air side system in different Tchw, sp and Tsa, sp. This would be around the seating area of the chair-like search landscape in Fig. 4.

epoch. For the results of year-round energy consumption, three searching proles with the label of best, mean and maximum are involved in each graph, which demonstrate the, respectively, best, average and maximum tness value in the population at a certain epoch. 5.1. Prediction of year-round setting of Tchw, sp From Fig. 7, for the optimization of the HVAC system operation at the existing set point of Tsa, sp at 16 8C, the minimized year-round energy consumption is 1.682 107 GJ. The convergence of the best solution starts from epoch 21 and thereafter, and the stopping criterion at epoch 50 has been justied. In this regard, the corresponding optimum Tchw, sp is 6.5 8C. 5.2. Prediction of year-round settings of both Tchw, sp and Tsa, sp For both Tchw, sp and Tsa, sp under optimization, the yearround energy consumption from the optimum pair is 1.636 107 GJ, as shown in Fig. 8. For the best solution, the search also started to converge from epoch 21. In this case, the optimum solutions of Tchw, sp and Tsa, sp are 5.2 and 15.0 8C, respectively. 5.3. Prediction of monthly settings of both Tchw, sp and Tsa, sp For the optimization of monthly settings, the results for all the 12 months are consolidated into one single graph. The

5. Optimization results and analysis For the optimization results, two types of graphs are presented, one is the optimized parameter against the searching epoch, another is the tness value against the

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

229

Fig. 7. Year-round energy consumption vs. epoch for only Tchw, sp under optimization.

Fig. 8. Year-round energy consumption vs. epoch for both Tchw, sp and Tsa, sp under optimization.

search processes for the minimized monthly energy consumption are shown in Fig. 9. Convergence is found after epoch 26 for nearly all the months, so the stopping criterion at epoch 50 has been also justied. Based on the optimization results in this scenario, the monthly reset schemes of Tchw, sp and Tsa, sp are summarized in Fig. 10. In the summer season, a higher Tchw, sp around 8 8C and a lower Tsa, sp around 13.7 8C would enhance the

COP of the chillers, although the operating period of chillers would be longer. On the other hand, during the mid-seasons, a lower Tchw, sp around 5 8C and a higher Tsa, sp around 15 8C would enhance the opportunity of free cooling and the half speed operation. At this time, the contribution of saving potential of the air side system outweighs the lower COP due to lower Tchw, sp. During the cold season, the optimum Tchw, sp and Tsa, sp are in between those of hot and

Fig. 9. Minimum monthly energy consumption vs. epoch.

230

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

Fig. 10. Monthly reset scheme of Tchw, sp and Tsa, sp for the HVAC system. Table 2 Comparison of year-round energy consumption against the existing settings Scenario Existing settings of Tchw, sp and Tsa, sp Scenario 1 Minimization of year-round energy consumption with optimum Tchw, sp only Year-round energy consumption (GJ) Saving percentage (%) 1.685 10 7 1.682 10 7 0.15 2.86 6.74

Scenario 2 Minimization of year-round energy consumption with both optimum Tchw, sp and Tsa, sp 1.636 10 7 Scenario 3 Minimization of monthly energy consumption with both optimum Tchw, sp and Tsa, sp 1.571107

mid-seasons, around 7 and 14 8C, respectively, and this would facilitate the half speed operation of the air side fans. Among the optimum pairs of Tchw, sp and Tsa, sp, none of them is close to the existing settings of 7 and 16 8C, respectively. As a whole, the inuence of energy consumption from the air side operation is crucial, with particular effect during the cold and mid-seasons. 5.4. Potential energy savings against existing settings The results of the three scenarios under study are summarized in Table 2, and the saving potentials compared. It was found that the optimum setting of Tchw, sp could not help much in energy saving, with a small saving percentage of 0.15%. If both Tchw, sp and Tsa, sp were involved, the saving percentage would be up to 2.86% for the year-round energy consumption. However, if the minimization is broken down at a monthly level, the saving potentials would be greatly increased to 6.74%. This illustrates the effectiveness of optimizing both Tchw, sp and Tsa, sp on the monthly basis. A monthly reset scheme of both Tchw, sp and Tsa, sp would be the most effective way in energy management of the HVAC system.

6. Conclusion With emphasis on efcient energy management of the existing HVAC system, a plant simulation tool is useful for

an in-depth investigation, particularly for non-typical projects and installations. By linking such a program to an appropriate optimization algorithm which can handle the discrete, non-linear and highly constrained nature of the HVAC problems, the required settings, with saving potentials for the energy management opportunities, can be identied. In this study, a simulationEP coupling approach which incorporated the component-based simulation and evolutionary programming optimization was linked for such purpose. The HVAC system under investigation had an air side system that would have signicant effect to the overall energy consumption. From the optimization results, the component-based HVAC model and the evolutionary programming technique worked well together in providing the optimum combination of the chilled water and supply air temperatures for effective energy management throughout a year. Among the different scenarios under study, optimizing both the set points of chilled water and supply air temperatures in a monthly basis would give a saving potential of about 7%, as compared to the one with the existing settings. The monthly reset scheme with both chilled water and supply air temperatures was suggested, and this would be an essential guideline to the operating team of the HVAC system under study, which would prevent any unnecessary estimates for the operating settings. For those plants already controlled and monitored by the building management system, this information would be also useful as a reference for supervisory control. As a

K.F. Fong et al. / Energy and Buildings 38 (2006) 220231

231

whole, the simulationEP coupling approach demonstrated effective potential to provide the required operational settings for existing installations, in order to implement a more strategic facilities management and better energy management, with immediate saving but without extra cost.

References
[1] Hong Kong Energy End-use Data (19902000). EMSD, HKSAR Government, December 2002. [2] Code of Practice for Energy Efciency of Air Conditioning Installations, EMSD, HKSAR Government, 2002. [3] Performance-based Building Energy Code, EMSD, HKSAR Government, 2003. [4] J.A. Wright, The optimised design of HVAC systems. Ph.D. Thesis, Loughborough University of Technology, UK, 1986. [5] M.C.W. Kintner-Meyer, An investigation of optimal sizing control of air-conditioning systems in commercial buildings, Ph.D. Thesis, University of Washington, US, 1994. [6] J. Huh, Optimal air-conditioning system operating strategies for combined temperature and humidity control in buildings, Ph.D. Thesis, University of Colorado, US, 1995. [7] M.S. deWit, Uncertainty analysis in building thermal modeling. Building Simulation 95, IBPSA, Madison, WI pp. 324330, 1995. [8] R.D. Taylor, Development of an integrated building energy simulation with optimal central plant control, Ph.D. Thesis, University of Illinois at Urbana-Champaign, Urbana, IL, USA, 1996 [9] K.F. Fong, T.T. Chow, L.S. Chan, W.L. Ma, C.K. Fong, A preliminary study of optimization of pipe route design of district cooling system, in: Proceedings of the 4th International Conference on Indoor Air Quality, Ventilation and Energy Conservation in Buildings, vol. II, China, 2001, pp. 10311038. [10] D. Dasgupta, Optimal scheduling of thermal power generation using evolutionary algorithms, in: D. Dasgupta, Z. Michalewicz (Eds.), Evolutionary Algorithms in Engineering Applications, Springer, Berlin: New York, 1997, pp. 317328. [11] J.A. Wright, H.A. Loosemore, R. Farmani, Optimization of building thermal design and control by multi-criterion genetic algorithm, Energy and Buildings 34 (2002) 959972.

[12] IISiBat Version 3.0.0.26, TRNSYS 15.3, Solar Energy Laboratory, University of Wisconsin-Madison, Madison. [13] J.E. Braun, Methodologies for the design and control of chilled water systems, Ph.D. Thesis, University of Wisconsin, Madison, 1988. [14] J.L. Threlkeld, Thermal Environmental Engineering, second ed., Prentice-Hall, New York, 1970. [15] K.F. Fong, V.I. Hanby, T.T. Chow, Optimization of MVAC systems for energy management by evolutionary algorithm, Facilities 21 (10) (2003) 223232. [16] J.A. Wright, V.I. Hanby, The formulation, characteristics, and solution of HVAC system optimized design problems, ASHRAE Transaction 93 (pt 2) (1987) 21332145. [17] J.A. Wright, HVAC simulation studies: solution by successive approximation, Building Services Engineering Research Technology 14 (4) (1993) 179182. [18] V.I. Hanby, P.P. Angelov, Application of univariate search methods to the determination of HVAC plant capacity, Proceeding of CIBSE A: Building Services Engineering Research Technology 21 (3) (2000) 161166. [19] L. Giraud-Moreau, P. Lafon, A comparison of evolutionary algorithms for mechanical design components, Engineering Optimization 34 (2002) 307322. [20] A.R. Simpson, G.C. Dandy, L.J. Murphy, Genetic algorithms compared to other techniques for pipe optimization, Journal of Water Resources Planning and Management 120 (4) (1994) 423443. [21] J.A. Wright, HVAC optimisation studies: sizing by genetic algorithm. Building Services Engineering Research Technology, 17(1), the Chartered Institution of Building Services Engineers, UK, pp. 714, 1996. [22] W. Huang, H.N. Lam, Using genetic algorithms to optimize controller parameters for HVAC systems, Energy and Buildings 26 (1997) 277 282. [23] Y. Sakamoto, A. Nagaiwa, S. Kobayasi, T. Shinozaki, ASHRAE Transactions 105 (pt 2) (1999) 111. [24] Y. Asiedu, R.W. Besant, P. Gu, HVAC duct system design using genetic algorithms, HVAC&R Research 6 (April (2)) (2000). [25] T. Back, D.B. Fogel, Z. Michalewicz, Evolutionary Computation 1: Basic Algorithm and Operators, Institute of Physics Publishing, Bristol and Philadelphia, 2000. [26] H. Beyer, The Theory of Evolution Strategies, Springer-Verlag, Berlin Heidelberg, 2001. [27] Z. Michalewicz, D.E. Fogel, How to Solve It: Modern Heuristics, Springer, Heidelberg, 2000. [28] D.R. Wulnghoff, Energy Efciency Manual, Energy Institute Press, Wheaton, Maryland, USA, 1999.

Anda mungkin juga menyukai