Assisted Calibration in Building Simulation–algorithm Description and Case Studies

Building simulation must be calibrated to fit the customer’s bill before applying energy saving measures. However, existing calibration methods are usually too complex to be included in building simulation software. The author has developed and implemented in DOE2.1E based building simulation software a calibration method which assists the software user in the calibration process using built-in engineering rules as well as optimization algorithms based on Marquardt-Levenberg non linear least square method. The article presents an overview of the calibration module’s functionalities and two case studies. For the first case study, the calibration process reduced the maximum error on the monthly electrical bill from 143% to 14%. For the second case, the maximum error was reduced from 40% to 11%. For this latter study case, the availability of hourly consumption data shows a need for schedule adjustment prior to calibration. INTRODUCTION Energetic building simulation is now widely used for retrofit evaluation. However, building simulation softwares usually require a large number of entries leading to a high risk of error. Calibration of the model is therefore essential. There are two different views on calibration procedure. The first, as described by Waltz (1992) and Yoon & Lee (1999), relies mainly on the user’s knowledge and experience. It therefore cannot efficiently be used by less or inexperienced users. The other class of calibration procedures is based on a mathematical approach. Carroll and Hitchcock (1993) presented an interesting calibration method. However, such a technique, as mentioned by Sun & Reddy (2006), requires a huge amount of simulations and can hardly be implemented within complex simulation software. The author has developed a calibration method that allies both approaches. The resulting calibration module is now integrated inside a user friendly interface for DOE2.1E simulation software. The module helps the user to adjust parameters so that the monthly electrical consumption and peak power demand fits that of the utility bill. This calibration procedure still requires knowledge of building services and simulations and should not be used by completely inexperienced users. However, the calibration module includes a pre-calibration algorithm that will help in the identification of parameters that could have an erroneous value. For non-expert users that are familiar with building services and simulations, this could significantly reduces the number of parameters to calibrate leading therefore to a fewer simulation runs. The calibration requires four simulation runs per parameter and each simulation takes approximately 9 to 16 seconds to be completed. A previous article (Lavigne & Millette, 2008) indicated that, in some cases, the precalibration could reduce the number of runs up to 55%. Nevertheless, the calibration module shows some limitations. First, as mentioned earlier, it can only be used to adjust monthly consumption and peak power demand. It is also limited to electricity for now and schedules cannot be adjusted with the tool. Finally, the calibration procedure can be simplified as a simple mathematical optimization problem with multiple solutions. Since no exhaustive search of the minimum is done (e.g. Monte Carlo or Latin Hypercube designs), one could end up with a solution that has no real physical meaning. However, this is mitigated by the pre-calibration and by ranges of values imposed to the parameters. The present article presents an overview of the calibration procedure as well as two case studies. METHODOLOGY The calibration process is realised using monthly electrical energy consumption and peak demand and is not intended for schedule adjustments. Other energy sources are not considered for now. The process requires local weather data corresponding to the period of measurements. Standardized weather files are not used. The calibration module is divided into two algorithms: pre-calibration and calibration. Pre-calibration When the user selects the calibration module from a tool menu, the pre-calibration algorithm is first launched. This algorithm does not require any further Eleventh International IBPSA Conference Glasgow, Scotland July 27-30, 2009