L egender神经网络建模及股票预测

Modeling and Stock Prediction Based on Legender Neural Networks

基于多项式逼近理论,将一组Legender正交多项式做为隐含层神经元的传递函数,再以其加权和函数做为神经网络输出,从而构成一种新型的三层多输入Legender神经网络模型;采用BP学习算法,通过对历史观测样本数据的训练,调整该神经网络的权值,建立非线性时间序列辨识模型,以此预测股票价格的变化。仿真实验表明,Legender神经网络具有优良的逼近任意非线性系统的特性,且学习收敛速度很快;深发展A股预测结果为:训练次数200,最大相对误差5.41%;深证成指预测结果为:训练次数120,最大相对误差4.17%。

This paper presents a new - type three - layer multi - input Legender neural network model based on polynomial approximation theory, which applies the Legender orthogonal polynomial as transfer function of hidden layer neural cell, uses weigh sum as its output. The nonlinear identification model on the time series is proposed to predict the change of stock by introducing the BP learning algorithm, training the data of former sample and adjusting the weights of network. The simulated results show that the Legender neural network has excellent characteristics of approaching any nonlinear system, and the network convergence speed is quite high. The forecasted results of Shenzhen - Development - Bank;s A shares are ： train degrees 200 and the most relative error 5.41%. The forecasted results of the composition stock index of Shenzhen securities exchange market are： train degrees 120 and the most relative error 4.17%.