中文摘要：

采用牛津大学Angeliki Xifara使用Ecotect系统模拟的768个不同建筑物数据，尝试将半参数中的部分线性单指标模型（PLSIM）用于住房建筑物负荷的预测研究中。同时采用BP神经网络以及迭代加权最小二乘法分别建立热负荷、冷负荷预测模型，将3种方法所得结果进行比较。研究结果表明部分线性单指标模型在建筑物负荷预测中相对误差均在O．00104以内且更直观，可以为国家调整住房结构、节约能源提供有力的模型支持。

英文摘要：

According to the 768 groups of different building data which was simulated by Ecotect by Angeliki Xifara from Oxford University, the article attempted to use partially linear single-index model to predict housing energy by predicting the heat load （HL） and cooling load （CL）. At the same time, this paper also use BP neural network and iteratively reweighted least squares to build models to predict the heat load （HL） and cooling load （CL）. To compare the results of the three methods, it showed that in terms of forecasting building load, the mse of the partially linear single-index model is less than 0.00104 and it is more intuitive. In terms of Country to adjust the energy structure and formulate energy policy, the model can provide strong support.