【摘要】 建筑性能模拟(Building Performance Simulation)包含对建筑围护结构、室内环境、能源消耗的模拟与计算,以及对运行故障的预测和诊断,对于我国绿色建筑行业的发展和节能减排目标的实现具有重要的推动作用。由于建筑系统的复杂性,建筑性能仿真模型通常难以完全描述真实的建筑运行状况,从而在建筑运行能耗和物理参数上表现出与实际数据之间的差异(Performance Gap)。为了获取更准确可靠的仿真结果,我们常常以实测建筑数据为依据,对模型的输入参数进行修正,即模型校验或校验模拟(Calibrated Simulation)。侧重于回顾当前建筑能耗模型校验采用的不同方法,描述了建筑能耗模型校验的典型步骤,将每个步骤中采用的方法按照设备/系统、单体建筑、城市/区域的能耗层级分别进行阐述;对比分析了不同能耗层级下、不同步骤中校验方法的优劣,为建筑能耗模型校验的实际应用给出了详细指导,提出了未来模型校验领域可能的研究方向。更多还原

【Abstract】 According to statistics, China’s building energy consumption accounts for 18%～20% of the total energy consumption, and shows a rising trend. Therefore, building energy saving is getting more and more attention. The accurate, scientific, and reasonable prediction of building energy is the basis of building energy saving. But due to the great difference between the actual conditions and the design conditions, the actual energy use is usually deviated from the prediction of the model heavily, which is called performance gap.Performance gap is caused by numerical uncertainty and modeling uncertainty. Model uncertainty can only be solved by improving the simulation algorithm, while calibration focuses more on reducing the numerical uncertainty. So, when simulating the energy use of one building, an important work is to calibrate the energy model with the measured historical data. Without calibration, the model cannot accurately predict the energy use of the real building.Calibration is based on the physical model, and then uses the measured data to modify the input parameters, so that the simulated data can be consistent with the measured data. By this way, the advantages of physical models and data-driven models are fully utilized, making the calibrated model not only reflect the physical characteristics, but also accurately match the actual energy consumption. The methodology used in calibrated simulation has four important steps:(1)Data collection and input data quality improvement(2)Input parameters tuning(3)Input parameters selection(4)Error analysis and model evaluationAccording to different space scales, energy model calibration can be divided into three levels:(1)Model calibration on component/system level, which focuses on the component models(such as chiller, boiler and pump) and system models(such as HVAC system and lighting system);(2)Model calibration on single building level, considering the energy use of one single building;(3)Model calibration on city/district level, which can be much more complex because of the need for considering the spatial form and microclimate.To improve the quality of the input parameters, researchers often use spot measured data and long-term monitoring data from sensors. This paper describes in detail how to improve the quality of both steady-state data and that related to such stochastic data as occupant behaviors, and points out the defects when the data are hard to observe or the measured data cannot be trusted.About how to adjust and tune the input parameters of the models, the methods can be broadly divided into manual methods and automatic methods. Manual calibration is to adjust model input in terms of users’ professional knowledge and experience, which is time-consuming and might bring high uncertainty to the model. The automatic methods mainly focus on the optimization methods, including Bayesian calibration, heuristic algorithm and hybrid algorithm of both. This paper summarizes the pro and cons of all the methods above and other methods which is helpful for the automatic process.Researchers often use sensitivity analysis to select the input parameters to be adjusted and tuned. Sensitivity analysis can be divided into local sensitivity analysis and global sensitivity analysis, the latter including regression, screening, variance based, and meta-model. The pro and cons of both local sensitivity analysis and different global sensitivity analysis methods are summarized.The error between the measured data and the simulated data is usually used to judge whether the established model is sufficiently consistent with the actual building or not. The granularity, types and energy levels of the data used in judgement can vary greatly according to the purpose of modeling. The coefficient often used to evaluate error includes CvRMSE and NMBE, and different organizations and countries have different standards for the values of two coefficient above.The details of each calibration step of each energy level are compared and summarized, to help researchers have a deeper understanding about the relationship and matching degree between calibration methods and energy level. Based on sufficient paper reviewing and summary, the possible research direction in calibrated simulation is proposed.更多还原