小波网络应用于空调负荷预测  

Application of Wavelet Neural Network for Air-conditioning Load Prediction

作　　者：陈柳[1] 

机构地区：[1]西安科技大学能源学院,陕西西安710054

出　　处：《建筑科学》2009年第10期70-73,共4页Building Science

基　　金：陕西省教育厅专项科研计划项目(07JK312)

摘　　要：准确预测空调负荷不仅对蓄能空调高效运行意义重大，而且也是冷热电三联产技术发挥优势的关键所在。本文提出一种小波网络应用于空调负荷的预测模型，通过小波分解，把空调负荷序列分解为不同频段的小波系数序列，再将各层的小波系数子序列重构到原尺度上，然后对小波系数序列采用相匹配的BP神经网络模型进行预测，最后合成空调负荷序列的最终预测结果。该预测模型中的低频小波系数a3和中频小波系数d3的神经网络输入变量为前1天小波系数值和对应时刻的温度、相对湿度、风速、总辐射量、天气状况和星期几编码共7个因子，并采用主成分分析法进行输入变量的降维；高频小波系数d2和d1以前几日的小波系数为输入因子。经过对西安市某综合楼的空调负荷进行预测，证明了预测值和实际运行值拟和很好，相对误差为-10％～8％。该预测模型具有预测精度较高、推广能力较强及计算速度较快的优点。Accurate prediction of air-conditioning load is very important not only for the efficient running of thermal storage air- conditioning system, but also for the presentation of the advantages of combined cooling heating and power（CCHP） technology. In this paper, a model of wavelet neural network for predicting the air-conditioning load was proposed. Firstly, through the wavelet transform, air-conditioning load series were decomposed into wavelet coefficient subsequences with different bands, and the wavelet coefficient subsequences were reconstructed into the original scales. Secondly, these wavelet coefficient subsequences were predicted using appropriate BP neural networks. Finally, the final prediction of the air-conditioning load was obtained by integrating the above results. For this predicting model, the input parameters of neural network of low and middle frequency wavelet coefficients included wavelet coefficients of the former day and meteorological factors, such as temperature, relative humidity, air velocity, et al. Also, the dimensionality reduction was carried out with principal component analysis （PCA） for these input parameters. On the other hand, the input parameters of neural network of high frequency wavelet coefficients were selected from wavelet coefficients of a few days ago. The air-conditioning load prediction results of a complex building in Xi＇an indicated that the prediction values in agreement with the operational values, the relative error was of - 10 % - 8 %. And therefore, it was considered that the wavelet and neural network models had good quality in terms of prediction precision, computing speed and generalization.

关 键 词：小波变换 BP神经网络 空调负荷预测 

分 类 号：TU831.2[建筑科学—供热、供燃气、通风及空调工程]