基于BPT-MLR模型的建筑能耗分析和预测

通过对福建省厦门市某高校8栋公寓楼的房间日平均用电量的分析,提出一种建筑能耗的平衡点温度-多元线性回归(BPT-MLR)模型.使用统计方法识别平衡点温度,并根据该平衡点温度分段对房间日平均用电量进行多元线性回归预测分析;对8个参数进行筛选,最终选4个参数作为模型变量,包括1个数值型变量(室外空气平均温度)和3个定类型变量(性别、节假日指数和晴雨天指数).结果表明：对比3种数据驱动模型,BPT-MLR模型的预测性能最优,其R²值达到了95.29%,比BP神经网络模型和多元线性回归模型的R²值分别高出0.04%和24.64%.

Analysis and Prediction of Building Energy Consumption Based on BPT-MLR Model

A balance point temperature-multiple linear regression(BPT-MLR)model for building energy consumption analysis and prediction is proposed by analyzing the average daily electricity consumption of rooms in 8 apartment buildings in a university in Xiamen City, Fujian Province. A statistical method is used to identify BPT, and a MLR prediction analysis is performed for the average daily room electricity consumption based on this BPT segment. 8 parameters are screened and 4 parameters are finally selected as model variables, including 1 numerical type variable(average outdoor air temperature)and 3 fixed type variables(gender, holiday index and sunny day index). The results show that the BPT-MLR model has the best prediction performance when comparing 3 data-driven models, with R² value 95.29%, which is 0.04% and 24.64% higher than that of the BP neural network model and MLR model respectively.