摘要
股票价格形成的时序数据具有非线性和非平稳性,传统的模型难以处理这些特征,为此提出一种处理非平稳数据的变结构时态神经网络预测模型。采用基于多分辨率分析的Mallat算法对股票数据进行预处理,将时序数据分为高频和低频序列。针对不同序列下的数据特征,将其转化为时态型数据集,采用粒子群算法寻找不同序列下的神经网络结构,建立变结构预测模型。实验结果表明,与未改进的神经网络和SVM方法相比,该预测模型具有更低的预测误差。
Aiming at the nonlinearity and non-stationarity of time series data formed by stock price,traditional models are difficult to solve these features.A variable structure temporal neural network prediction model for dealing with non-stationary data was proposed.The Mallat algorithm based on multi-resolution analysis was used to preprocess the stock data,and the time series data were divided into high frequency and low frequency sequences.According to the data characteristics under different sequences,they were transformed into a temporal data set.The particle swarm optimization algorithm was used to find the neural network structure under different sequences,and the variable structure prediction model was established.Experimental results show that the prediction model has lower prediction error than the original neural network and SVM method.
作者
孟志青
朱涵琪
MENG Zhi-qing;ZHU Han-qi(School of Management,Zhejiang University of Technology,Hangzhou 310023,China)
出处
《计算机工程与设计》
北大核心
2020年第6期1733-1741,共9页
Computer Engineering and Design
基金
浙江省自然科学基金项目(MLY18A010031)。
关键词
时态数据
小波分析
粒子群算法
神经网络
股票预测
temporal data
wavelet analysis
particle swarm optimization
neural network
stock forecasting