In solving optimization problems for building design and control, the cost function is often evaluated using a detailed building simulation program. These programs contain code features that cause the cost function to be discontinuous. Optimization algorithms that require smoothness can fail on such problems. Evaluating the cost function is often so time-consuming that stochastic optimization algorithms are run using only a few simulations, which decreases the probability of getting close to a minimum. To show how applicable direct search, stochastic, and gradient-based optimization algorithms are for solving such optimization problems, we compare the performance of these algorithms in minimizing cost functions with different smoothness. We also explain what causes the large discontinuities in the cost functions.

10acoordinate search10adirect search10agenetic algorithm10ahooke–jeeves10aoptimization10aparticle swarm optimization1 aWetter, Michael1 aWright, Jonathan, A. uhttps://simulationresearch.lbl.gov/publications/comparison-deterministic-and02172nas a2200169 4500008004100000245009100041210006900132260002700201300001400228490000800242520160400250100002001854700002501874700002401899700001601923856006301939 2003 eng d00aComparison of a generalized pattern search and a genetic algorithm optimization method0 aComparison of a generalized pattern search and a genetic algorit aEindhoven, Netherlands a1401-14080 vIII3 aBuilding and HVAC system design can significantly improve if numerical optimization is used. However, if a cost function that is smooth in the design parameter is evaluated by a building energy simulation program, it usually becomes replaced with a numerical approximation that is discontinuous in the design parameter. Moreover, many building simulation programs do not allow obtaining an error bound for the numerical approximations to the cost function. Thus, if a cost function is evaluated by such a program, optimization algorithms that depend on smoothness of the cost function can fail far from a minimum.

For such problems it is unclear how the Hooke-Jeeves Generalized Pattern Search optimization algorithm and the simple Genetic Algorithm perform. The Hooke-Jeeves algorithm depends on smoothness of the cost function, whereas the simple Genetic Algorithm may not even converge if the cost function is smooth. Therefore, we are interested in how these algorithms perform if used in conjunction with a cost function evaluated by a building energy simulation program.

In this paper we show what can be expected from the two algorithms and compare their performance in minimizing the annual primary energy consumption of an office building in three locations. The problem has 13 design parameters and the cost function has large discontinuities. The optimization algorithms reduce the energy consumption by 7% to 32%, depending on the building location. Given the short labor time to set up the optimization problems, such reductions can yield considerable economic gains.

1 aWetter, Michael1 aWright, Jonathan, A.1 aAugenbroe, Godfried1 aHensen, Jan uhttp://www.ibpsa.org/proceedings/BS2003/BS03_1401_1408.pdf00504nas a2200133 4500008004100000245008500041210006900126300001200195490000700207100002500214700001800239700002500257856008800282 2003 eng d00aField Testing Model-Based Condition Monitoring on a HVAC Cooling Coil Sub-System0 aField Testing ModelBased Condition Monitoring on a HVAC Cooling a103-1160 v241 aBuswell, Richard, A.1 aHaves, Philip1 aWright, Jonathan, A. uhttps://simulationresearch.lbl.gov/publications/field-testing-model-based-condition00551nas a2200133 4500008004100000245010500041210006900146260002900215100002500244700001800269700002200287700002500309856008300334 2002 eng d00aNon-Linear Recursive Parameter Estimation Applied to Fault Detection and Diagnosis in Real Buildings0 aNonLinear Recursive Parameter Estimation Applied to Fault Detect aLiège, Belgiumc12/20021 aBuswell, Richard, A.1 aHaves, Philip1 aSalsbury, Tim, I.1 aWright, Jonathan, A. uhttps://simulationresearch.lbl.gov/publications/non-linear-recursive-parameter00468nas a2200121 4500008004100000245007400041210006900115490000800184100001800192700002200210700002500232856008900257 1996 eng d00aCondition Monitoring in HVAC Subsystems using First Principles Models0 aCondition Monitoring in HVAC Subsystems using First Principles M0 v1021 aHaves, Philip1 aSalsbury, Tim, I.1 aWright, Jonathan, A. uhttps://simulationresearch.lbl.gov/publications/condition-monitoring-hvac-subsystems00506nas a2200121 4500008004100000245009300041210006900134260002700203100002200230700001800252700002500270856008900295 1995 eng d00aA Fault Detection and Diagnosis Method Based on First Principles Models and Expert Rules0 aFault Detection and Diagnosis Method Based on First Principles M aBejing, Chinac09/19951 aSalsbury, Tim, I.1 aHaves, Philip1 aWright, Jonathan, A. uhttps://simulationresearch.lbl.gov/publications/fault-detection-and-diagnosis-method00600nas a2200157 4500008004100000245008700041210006900128260002900197100002300226700002300249700002400272700001800296700002200314700002500336856008100361 1994 eng d00aModel-Based Approaches to Fault Detection and Diagnosis in Air-Conditioning System0 aModelBased Approaches to Fault Detection and Diagnosis in AirCon aLiège, Belgiumc12/19941 aBenouarets, Mourad1 aDexter, Arthur, L.1 aFargus, Richard, S.1 aHaves, Philip1 aSalsbury, Tim, I.1 aWright, Jonathan, A. uhttps://simulationresearch.lbl.gov/publications/model-based-approaches-fault