基于影响因素分析和小波神经网络的供热量预测  

Heating load forecast based on influencing factors analysis and wavelet neural network

作　　者：王美萍[1] 张佼[1] 田琦[1] 

机构地区：[1]太原理工大学

出　　处：《暖通空调》2014年第3期113-118,共6页Heating Ventilating & Air Conditioning

基　　金：国家"十二五"科技支撑计划项目(编号:2012BAJ04B02)

摘　　要：基于统计方法分析了实测时间序列中各影响因素与供热量的相关性。应用小波分析有效提取序列中的局部信息,与神经网络相结合,可分析蕴藏在系统中的非线性动态特性。建立了小波神经网络预测模型,把影响供热量的因素分为与其相关性较大(系统循环流量、供水温度和回水压力)和较小(供、回水压力和回水温度)的2组数据,预测结果证实与供热量相关性较大的1组影响因素的拟合程度比相关性小的高。就预测结果的准确性与BP神经网络结构进行了比较。结果表明,基于影响因素分析和梯度修正的小波神经网络供热量预测方法具有良好的动态特性、较强的泛化能力和较高的预测精度,适用于系统供热量的短期预测。Analyses. the correlation between influencing, factors and heating load from the measured values of time series by statistical methods. Nonlinear dynamic characteristics hidden in the central heating system can be analysed through using the wavelet analysis that can effectively extract local information of time series, combined with neural network. Establishes the wavelet neural network forecast model. Divides the influencing factors of heating load into two groups, i.e. the higher correlation group （including circulation flow rate, supply water temperature and return water pressure） and the lower correlation group （including supply and return water pressure and return water temperature）. The forecast results show that the fitting degree of the higher correlation group is higher than that of the lower correlation group. Compares the accuracy of the forecasted value with that by BP neural network. The results show that the wavelet neural network heating load forecast method based on influencing factor analysis and gradient revision has better dynamic characteristics, higher generalization ability and accuracy, and it is suitable for short-term forecast of heating load.

关 键 词：影响因素 小波神经网络 供热量预测 时间序列 动态特性 泛化能力 预测精度 

分 类 号：TU995[建筑科学—供热、供燃气、通风及空调工程]