摘要
股票市场具有不确定性和非线性等特点,因此准确地预测股票价格对投资者来说是一项重大挑战。现有的股价预测模型较为单一,预测精度不高。针对这一问题,提出一种基于模糊K线的长短期记忆(LSTM)网络和支持向量回归多阶段混合模型(FCLSTM-vSVR)。研究第一阶段,基于遗传算法对LSTM网络进行参数寻优,找到时间窗口和隐藏层神经元的最佳值,并利用训练好的LSTM进行股票价格初步预测,计算出股票价格的残差值。第二阶段,利用模糊K线将原始价格序列转换为模糊数据,并作为vSVR模型的输入,利用vSVR模型预测残差值。综合前文论述后,再将两阶段的预测值之和作为最终的股票价格预测值。通过对比实验得出,该模型具有更高的预测准确率,在股票价格的一步预测方面优于其他对比模型。
Due to the uncertainty and nonlinearity of stock market,accurately predicting stock prices is always a major challenge for investors.The existing stock price prediction models are relatively single and the prediction accuracy is undesirable.In order to solve this problem,a multi-stage hybrid model of long short-term memory(LSTM)networks and support vector regression based on the fuzzy Candlestick,namely FCLSTM-vSVR,is presented.In the first stage,some parameters of LSTM network are optimized based on Genetic Algorithm.The stock price is preliminarily predicted by LSTM,and the residual values of stock price are calculated.In the second stage,original price series are transformed to ambiguous outputs by applying fuzzy Candlestick.Then the fuzzy outputs are used as the inputs of vSVR model to predict the residual values.Finally,the final forecast values of stock price are taken by summing the predicted values of the two stages.Experimental results show that the model presented has higher prediction accuracy and is superior to the baseline model in one-step prediction of stock price.
作者
刘茜阳
宋燕
张亚萌
LIU Xiyang;SONG Yan;ZHANG Yameng(College of Science,University of Shanghai for Science and Technology,Shanghai 200093,China;School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《智能计算机与应用》
2022年第4期54-60,69,共8页
Intelligent Computer and Applications