摘要
提出通过小波分解对各负荷子序列进行特性分析初选影响因素后,采用信息熵法从初选变量中自动筛选出对负荷较重要的因素,然后采用改进的主成分分析法消除重要影响因素间的相关性,采用动态聚类法对各分解序列的样本归类,通过灰色关联分析选择出与预测时刻负荷模式最相似的类作为神经网络训练的典型样本集,采用蚁群优化算法训练各子序列相应神经网络模型,采用小波重构得到最终负荷预测结果。并利用某地区1999年的实际负荷对所提方法进行验证,结果表明了该方法的合理性和有效性。
A hybrid load forecast method is put forward. The character analysis is carried out with wavelet decomposition for each load subsequence and influencing factors are thus determined, from which main factors are selected using the information entropy method and their relativity is eliminated using the improved principal component analysis method. The dynamic clustering analysis is used to divide the historical load data into several categories and the grey relative analysis to pick out one as the typical sample set,which colony optimization algorithm is then used to is most similar to the forecasting load mode. The ant train the corresponding neural network model of each decomposed subsequence and the wavelet reconstruction is used to achieve final forecasts. Actual loads of a district in 1999 are taken for verification,which shows the proposed method is rational and effective.
出处
《电力自动化设备》
EI
CSCD
北大核心
2007年第5期40-44,共5页
Electric Power Automation Equipment
关键词
负荷预测
小波变换
信息熵
主成分分析
动态聚类法
蚁群优化算法
load forecast
wavelet transform
information entropy
principal component analysis
dynamic clustering algorithm
ant colony optimization algorithm
