Enhanced sequential search strategies for identifying cost-optimal building designs on the path to zero net energy

Identifying cost-optimal building designs, particularly on the path to zero net energy, requires accounting for complex energy interactions between building measures. BEopt, building optimization software developed by the National Renewable Energy Laboratory, incorporates such interactions as it probes a large, multivariate parameter search space for optimal combinations of measures. Measures include wall constructions, window glazing properties, heating, ventilation, and air conditioning (HVAC) equipment, lighting, solar thermal, and photovoltaics. BEopt utilizes a sequential search optimization methodology. Enhancements to this methodology, both in terms of robustness (ability to generate the true cost-optimal curve) and efficiency (number of required simulations), were developed and tested. Regarding robustness, three deficiencies in the sequential search were investigated: the "invest/divest", "large-step", and "positive interactions" special cases. Solutions to the first two special cases do not require user interaction and were implemented in BEopt. Using a test suite (comprised of small, medium, and large optimizations across six climates), the occurrence of the two special cases were identified. The optimization results were additionally validated against large, but not exhaustive, parametric runs. For search efficiency, eleven strategies were devised to reduce the total number of required simulations. The strategies work by either reducing the number of search iterations or by reducing the number of simulations per iteration. Five such strategies were found to be particularly effective without significantly compromising robustness: (1) skip predicted outliers, (2) skip fine points, (3) option lumping, (4) skip less efficient options, and (5) skip extraneous points. Combinations of these strategies were then assembled into efficiency packages, ranging from conservative to aggressive. The most conservative package achieves a 15% reduction in simulations (without any sacrifice on robustness), while the most aggressive package achieves a 71% reduction in simulations (with a 1.2% maximum deviation, compared to the reference optimization, in the cashflow of any optimal building design over the entire range of designs on the path to zero net energy)