基于小波变换的模糊神经网络短期负荷预测方法  

Wavelet Transform Integrated Fuzzy Neural Network for Short Term Load Forecast

作　　者：郭崇[1,2] 王征[3] 纪建伟[2] 

机构地区：[1]辽宁工业大学管理学院,辽宁锦州121001 [2]沈阳农业大学信电学院,辽宁沈阳110161 [3]国网辽宁省电力有限公司发策部,辽宁沈阳110004

出　　处：《湘潭大学自然科学学报》2017年第1期109-113,共5页Natural Science Journal of Xiangtan University

基　　金：辽宁省自然科学基金项目(201202191)

摘　　要：为了解决传统神经网络的预测精度取决于输入变量和测试样本的缺陷,采用二阶Daubechies小波作为母小波,通过离散小波变换和逆变换的多分辨率把负荷序列分解为4个小波分量,不但把握了负荷序列的规律性,而且减轻了神经网络的学习压力.采用自适应遗传算法对模糊规则和权重进行修正,优化模糊神经网络,提出GNN-W-GAF模型.该模型既发挥了模糊算法的特点,又使得各种知识点在神经网络中相互融合,避免了初始值设定的随意性.仿真结果表明,该方法能显著提高预测精度和预测性能.To overcome the drawbacks of ANN which the prediction effect depends on the input variable and the test samples,the paper applied second-order Daubechies wavelets as mother wavelet and divided load sequence into four wavelet components by multi-resolution wavelet transform and its inverse transform.It not only grasped the importance of the regularity of the load sequence,but also reduced the neural network learning pressure.An adaptive genetic algorithm was used to amend fuzzy rules and network＇s weights,and optimize fuzzy neural network.GNN-W-GAF is proposed.Fuzzy neural network only played the characteristics of fuzzy algorithm but also mixed all kinds of knowledge together avoiding the arbitrariness of the initial value setting.The simulation results show that the proposed prediction method improves prediction accuracy and prediction performance significantly.

关 键 词：神经网络 遗传算法 自适应 小波变换 模糊算法 

分 类 号：O224[理学—运筹学与控制论]