Assessment of Building Energy Simulation Tools to Predict Heating and Cooling Energy Consumption at Early Design Stages

Recent developments in dynamic energy simulation tools enable the definition of energy performance in buildings at the design stage. However, there are deviations among building energy simulation (BES) tools due to the algorithms, calculation errors, implementation errors, non-identical inputs, and different weather data processing. This study aimed to analyze several building energy simulation tools modeling the same characteristic office cell and comparing the heating and cooling loads on a yearly, monthly, and hourly basis for the climates of Boston, USA, and Madrid, Spain. First, a general classification of tools was provided, from basic online tools with limited modeling capabilities and inputs to more advanced simulation engines. General-purpose engines, such as TRNSYS and IDA ICE, allow users to develop new mathematical models for disruptive materials. Special-purpose tools, such as EnergyPlus, work with predefined standard simulation problems and permit a high calculation speed. The process of reaching a good agreement between all tools required several iterations. After analyzing the differences between the outputs from different software tools, a cross-validation methodology was applied to assess the heating and cooling demand among tools. In this regard, a statistical analysis was used to evaluate the reliability of the simulations, and the deviation thresholds indicated by ASHRAE Guideline 14-2014 were used as a basis to identify results that suggested an acceptable level of disagreement among the outcomes of all models. This study highlighted that comparing only the yearly heating and cooling demand was not enough to find the deviations between the tools. In the annual analysis, the mean percentage error values showed a good agreement among the programs, with deviations ranging from 0.1% to 5.3% among the results from different software and the average values. The monthly load deviations calculated by the studied tools ranged between 12% and 20% in Madrid and 10% and 14% in Boston, which were still considered satisfactory. However, the hourly energy demand analysis showed normalized root mean square error values from 35% to 50%, which were far from acceptable standards