Paper Conference

ASHRAE & IBPSA-USA SimBuild 2016: Building Performance Modeling Conference


Optimizing the Use of Reduced Weather Sets in Building Energy Simulations

Yu Joe Huang
White Box Technologies, Moraga, CA

Abstract: In almost all cases, building energy simulations are done with an entire year of weather data. As building models become larger and simulation programs become more complex, simulation runtimes have become an issue, despite the overall improvements in computer speed. One obvious way to reduce runtimes is to not run the simulation for an entire year, but for a subset of days extracted from the year. The trade-off for the quicker runtimes is of course a loss in accuracy. Previous efforts using this technique have simply taken a random sampling of the time series, such as doing the simulation for four arbitrary 15-day periods spaced equally from a "typical year" weather file. This paper investigates using the same technique developed for selecting the most representative months making up a "typical year" weather files but applies it to select a 7-day time series, i.e., a “typical week”, that best matches the average long-term climatic conditions of a month. If the source weather data is a “typical year” weather file, this technique analyzes the cumulative frequency distributions (CFD) of temperature, solar, wind, etc., for each of the 22-25 7-day series within that month, compare them to the CFD for the full month, and then selects the 7-day series with the smallest deviation in the CFDs as the “typical week”. The twelve “typical weeks” are then concatenated to produce a reduced “typical year” weather file containing only 12, rather than 52, weeks of data. For simplicity, the date stamp for each 7-day series can be set to the middle of each month. There are many simple ways to make such reduced weather files compatible with existing simulation program, although relatively modest changes would be needed to simplify their use and speed up runtimes by eliminating re-initialization between the “typical weeks”. Once that is done, runtimes should be reduced to 1/4 with little loss in accuracy compared to running the full year. If the source weather data is the historical time-series, that should be used instead for selecting the “typical week”, which would greatly expand the number of candidate 7-day series by the number of years in the time-series. In such instances, it has been found that the reduced weather data set has an equal or at times even better fidelity than the “typical year” file to the average long-term climate conditions.
Pages: 447 - 453