基于迟滞神经网络的风速时间序列预测

为了改善风速时间序列的预测性能,提出了一种基于迟滞神经网络的预测方法.通过改变神经元激励函数的方式将迟滞特性引入神经网络中,以增强历史输入对当前响应的影响,从而提高有用信息的利用率,提高风速时间序列的预测性能;借助于相空间重构理论构造风速预测训练样本,采用梯度下降法对网络权值进行训练,利用遗传算法对迟滞参数进行优化.仿真结果表明:与传统神经网络及ARMA模型等方法相比,迟滞神经网络能够有效减小风速时间序列的预测误差,提高预测性能.

Prediction of wind speed time series based on hysteretic neural network

In order to improve the prediction performance of the wind speed time series,a new prediction method based on hysteretic neural network is proposed.Hysteretic characteristic which can make history input change the current response of the neural network is brought into the neural network by changing activation function.Therefore,the utilization rate of useful information is enhanced,and the prediction performance of the wind speed time series can be improved.The training samples are reconstructed by the phase space reconstruction theory,and the connection weights of the network are trained by gradient descent method.And the hysteretic parameters are optimized by genetic algorithm.Simulation results show that the method can get better prediction performances than conventional neural network and ARMA model,and the prediction error can be reduced validly.