In Energy Performance Contracts (EPC), savings resulting from building and energy system retrofit are estimated as the difference between the actual consumption an prediction from a baseline model. This paper proposes an automated method to select the most relevant baseline model. In addition to commonly used cooling degree days (CDD), climatic variables mixing temperature and humidity are introduced, such as dew point or moist air enthalpy. In terms of prediction algorithm, 11 algorithms, linear and nonlinear, are compared. The novel methodology enables an energy analyst to easily find the most relevant model for a specific building, with a focus on both precision and seasonal bias. By applying the novel methodology on 11 buildings under real EPCs, the result shows a decrease of prediction error by 23.5% with respect to models referenced in the contracts. Lastly, the proposed method is generalized to address the case where contracts involve multiple buildings of different types or consumption ranges, with the ability to identify a common best model by using new dimensionless indicators, and reach an 11% improvement in prediction error.

在能源绩效合同 (EPC) 中，建筑和能源系统改造产生的节约被估计为实际消耗量与基线模型预测之间的差异。本文提出了一种自动方法来选择最相关的基线模型。除了常用的冷却度日 (CDD) 外，还引入了混合温度和湿度的气候变量，例如露点或湿空气焓。在预测算法方面，比较了线性和非线性11种算法。这种新颖的方法使能源分析师能够轻松找到与特定建筑物最相关的模型，同时关注精度和季节性偏差。通过在真实 EPC 下的 11 座建筑物上应用新方法，结果表明预测误差减少了 23。合同中引用的模型为 5%。最后，所提出的方法被推广以解决合同涉及多个不同类型或消费范围的建筑物的情况，能够通过使用新的无量纲指标来识别通用的最佳模型，并在预测误差方面达到 11% 的改进。