基于数据驱动的地下车站 能耗预测模型对比研究

对比了最小二乘多元线性回归、岭回归、Lasso 回归、随机森林、XGBoost 在地下车站通风空调、垂直 交通能耗预测领域的应用效果。研究发现，对于地下车站通风空调和垂直交通能耗预测，各算法的均方根误差 的变异系数(CV-RMSE)均在 10%以下，可达到工程应用要求的精度。其中，XGBoost 算法在通风空调能耗预测 中的 CV-RMSE 为 5.1%，在垂直交通能耗预测中的 CV-RMSE 为 5.4%，预测效果明显优于其他算法。从计算 成本来看，最小二乘回归、岭回归、Lasso 回归算法计算成本较低，随机森林和 XGBoost 模型调参复杂、计算 成本较高。对比常用的数据驱动算法在地下车站能耗预测中的预测精度和计算成本，为地下车站模型的搭建提 供算法参考。

Comparative Research on Data-driven Energy Prediction Models for Underground Subway Stations

This study compared a variety of commonly used data-driven methods in the field of energy-prediction models of ventilation and air conditioning (VAC) and vertical transport (TRANS) in subway stations, including least-squares multiple linear regression, ridge regression, lasso regression, random forest, and XGBoost. It was found that the CV-RMSE indices of the aforementioned methods were less than 10% for VAC and TRANS energy prediction, the accuracy of which is adequate for engineering applications. Among them, the CV-RMSE of XGBoost is only 5.1% for the VAC model and 5.4% for the TRANS model, which is obviously better than that of the other algorithms. From the perspective of computational cost, the least squares regression, ridge regression, and lasso regression algorithms have comparatively low cost, while the random forest and XGBoost models have high computational costs. In this study, the algorithm reference for the development of a subway station energy prediction model is provided by comparing the prediction accuracy and calculation cost of commonly used data-driven methods.