摘要
针对传统神经网络收敛速度慢、容易陷入局部极值的问题,文中提出一种改进型小波神经网络以实现网络全局最优化。首先,将小波神经网络与随机矢量函数连接型网络相融合构建一种新型小波链神经网络(NW-FLNN);其次,以小波基函数作为NW-FLNN的隐含层的传递函数,并利用梯度修正法训练该模型各参数;最后,选用澳大利亚新南威尔士州电价数据作为实验数据集,分别对NW-FLNN神经网络、逆传播BP神经网络与小波神经网络进行预测性能比较。实验结果表明:该新型网络预测模型较BP神经网络与小波神经网络性能更优,可明显减少网络迭代次数与隐层神经元数目,且平均百分比误差最大降低至0. 0317,满足实时性要求。
In order to solve the problem that the traditional neural network has slow convergence rate and may easily fall into local extreme,this paper proposes an improved wavelet neural network to realize the global optimization of the network. Firstly,a new wavelet functional link neural network( NW-FLNN) is constructed by combining the wavelet neural network with the random vector functional link net. Secondly,the wavelet basis function is used as the transfer function of the hidden layer of NW-FLNN,and the gradient correction method is conducted to train each parameter of the model. Finally,the electricity price data of New South Wales in Australia is selected as the experimental dataset,and the prediction performances of the NW-FLNN neural network,the back propagation BP neural network and the wavelet neural network are respectively compared. The experimental results show that the new network prediction model has better performance than BP neural network and wavelet neural network,and can significantly reduce the number of network iterations and hidden layer neurons,and the average percentage error is reduced to 0. 0317,which meets the real-time requirements.
引文
[1] Swief R A,Hegazy Y G,Abdel-Salam T S,et al. Support vector machines(SVM)based short term electricity load-price forecasting[C]//PowerTech,2009 IEEE Bucharest. IEEE,2009:1-5.
[2]刘西陲,沈炯,李益国.基于加权双高斯分布的广义自回归条件异方差边际电价预测模型[J].电网技术,2010,34(1):139-144.Liu Xichui,Shen Jiong,Li Yiguo. A Generalized Auto-Regressive Conditional Heterosked asticity Model for System Marginal Price Forecating Based on Weighted Double Gaussian Distribution[J]. 2010,34(1):139-144.
[3] Nargale K K,Patil S B. Day ahead price forecasting in deregulated electricity market using Artificial Neural Network[C]//International Conference on Energy Efficient Technologies for Sustainability. IEEE,2016:527-532.
[4] Gooijer J G D,Hyndman R J. 25 years of time series forecasting[J]. International Journal of Forecasting,2006,22(3):443-473.
[5] Chen Y,Yang B,Dong J. Time-series prediction using a local linear wavelet neural network[J]. Neurocomputing,2006,69(4-6):449-465.
[6] Pany P K,Ghoshal S P. Dynamic electricity price forecasting using local linear wavelet neural network[J]. Neural Computing&Applications,2015,26(8):2039-2047.
[7] Singh N,Husain S,Mohanty S R. An improved WNN for day-ahead electricity price forecasting[C]//Engineering and Systems. IEEE,2016:1-6.
[8] Mirakyan A,Meyer-Renschhausen M,Koch A. Composite forecasting approach,application for next-day electricity price forecasting[J]. Energy Economics,2017,66.
[9]田中大,高宪文,石彤.用于混沌时间序列预测的组合核函数最小二乘支持向量机[J].物理学报,2014,63(16):66-76.Tian Zhongda,Gao Xianwen,Shi Tong,et al. Combination kernel function least squares support vector machine for chaotic time series prediction[J]. Acta Phys. Sin,2014,63(16):66-76.
[10]Cardaliaguet P,Euvrard G. Approximation of a function and its derivative with a neural network[J]. Neural Networks,1992,5(2):207-220.
[11]Pao Y H,Park G H,Sobajic D J. Learning and generalization characteristics of the random vector functional-link net[J]. Neurocomputing,1994,6(2):163-180.
[12]Cecotti H. Deep Random Vector Functional Link Network for handwritten character recognition[C]//International Joint Conference on Neural Networks. IEEE,2016:3628-3633.