泛化能力改善的神经网络方法在空调负荷预测中的应用  

Application of Artificial Neural Network Model to Improve the Generalization Ability in Air Conditioning Load Prediction

作　　者：李海军[1] 何大四[1] 

机构地区：[1]中原工学院,郑州450007

出　　处：《建筑科学》2009年第6期90-94,共5页Building Science

摘　　要：本文针对神经网络方法进行负荷预测时存在的泛化能力较差的缺点,提出了对历史数据进行相关性分析和输入数据主成分分析以提高神经网络泛化能力。实例分析结果表明:历史数据的相关性分析为空调负荷预测神经网络模型输入参数的合理选择提供了依据,主成分分析方法在降低神经网络模型输入维数、提高该方法泛化能力方面有较好的作用。Because existing artificial neural network （ANN） for predicting the thermal load has poor generalization ability, two methods to enhance the generalization ability of ANN were introduced in this paper, which were correlation analysis of the historical data and principal component analysis of input data. ANN input items could be determined reasonably using correlation analysis of the historical data, and the dimension of ANN model would be reduced using principal component analysis. With these the two methods, ANN performance could be improved.

关 键 词：神经网络 泛化能力 相关性分析 主成分分析 负荷预测 

分 类 号：TU831.2[建筑科学—供热、供燃气、通风及空调工程] TP183[自动化与计算机技术—控制理论与控制工程]