基于神经网络与多目标优化的建筑围护结构热工参数组合优化设计方法研究
摘要：
针对传统建筑围护结构热工参数优化数值模拟方法耗时长、效率低,难以满足工程时效要求的问题,本文首先利用EnergyPlus能耗模拟数据构建了BP神经网络建筑能耗预测模型,然后利用建筑能耗预测模型与多目标优化模型给出了不同围护结构热工参数推荐取值。研究结果表明:综合屋面、墙面、地面及外窗热工参数的建筑能耗预测模型拟合效果良好,平均决定系数(R 2 )为94.98%,平均均方根误差变异系数(CVRMSE)为4.42%;相较于案例建筑,基于BP神经网络能耗预测模型与多目标优化模型所得最优参数组合能耗降低38.7%,初投资增加25.10万元,增加了15.1%。
Abstract：
Aiming at the issues of long time-consuming and low efficiency in traditional numerical simulation methods for optimizing thermal parameters of building envelopes, which are difficult to meet the timeliness requirements of engineering projects, this paper initially employs EnergyPlus energy consumption simulation data to construct a building energy consumption prediction model on the basis of a BP neural network,and then provides recommended values for various thermal parameters of the envelope using the building energy consumption prediction model combined with a multi-objective optimization model. The research results show that the building energy consumption prediction model, incorporating thermal parameters of roof, wall, floor and window, exhibits a high degree of fit, with an average coefficient of determination (R2 ) of 94.98% and an average coefficient of variation of root mean square error (CVRMSE) of 4.42%. Compared with the case building, the optimal parameter combination, as predicted by the BP neural network energy consumption prediction model and multi-objective optimization model, reduces energy consumption by 38.7%, while the initial investment increases by 251 thousand yuan, representing a 15.1% increase.
Keywords:envelope; thermal parameter; BP neural network; particle swarm optimization (PSO); parameter optimization