摘要
针对小波分解预测方法中高频随机序列预测精度低的问题,采用一种自行开发的小波神经网络工具箱进行预测,该方法既有小波函数信号特征提取能力,又有BP神经网络工具箱快速、方便、鲁棒性强等特点,同时给出了神经网络初始化方法。该方法的显著特点是简捷、快速、实用性强,可以克服非平稳信号预测及大批量数据的训练问题,对推广小波神经网络的应用具有重要意义。对短期电力负荷预测的仿真实验结果表明,该方法绝对平均精度达到1%以上,并具有较高的预测效率。
For the low accuracy prediction problem of high frequency random series in wavelet decomposing method, a self-devel oped wavelet neural network toolbox is adopted to predict, which not only has the extraction ability of wavelet function signal, but also possesses features of high-speed, convenience and robust of BP neural network toolbox, a method for neural network initialization is provided. The most obvious characteristic of the method mentioned is simplicity, high-speed and utility, which is capable of overcoming problems of nonstationary signal prediction and mass data training. It has great significance for the promotion of applying wavelet neural network. The simulation experiments of short-term load prediction indicates that absolute average accuracy of this method reach more than 1 % and has higher prediction efficiency.
出处
《计算机工程与设计》
CSCD
北大核心
2013年第6期2262-2266,2276,共6页
Computer Engineering and Design
基金
国家安全生产监督管理总局科研计划基金项目(06-472)
河北省教育厅科学技术研究基金项目(Z2006439)
关键词
初始参数
小波变换
小波神经网络
工具箱
短期负荷
预测
initial parameter
wavelet transform
wavelet network
toolbox
short-term electric load
prediction