融合SDE算法和双路加权门控循环神经网络的股价走势预测模型

Stock Price Trend Prediction Model Based on SDE Algorithm and Bidrectional Weighted Gated Recurrent Neural Network

关于股票价格走势的预测,传统的操作方法多是通过统计分析工具或者是单一的机器学习算法进行预测,很难准确把握股价这种时间序列数据的非线性和非平稳性等特征,从而使预测精度受限.融合SDE算法与加权BiGRU网络的优化预测模型,先使用SDE全局寻优网络的结构参数,求得最优初始权值、阈值以及权重系数,再将优化的参数应用到改良的加权BiGRU网络模型中进行预测.优化的预测模型能够有选择的考虑过去和未来时间点对当前时刻数据的影响,而且能有效避免局部最优值以及网络的长程依赖问题.实验结果表明,优化的预测模型与其他传统神经网络预测模型相比较,预测误差得到显著降低,预测准确度得到明显增强.

As for the prediction of stock price,the traditional operation method employed mostly a statistical analysis tool or a single machine learning algorithm,which makes it difficult to accurately grasp the non-linear and non-stationary characteristics of time series data such as stock price,thus limiting the prediction accuracy.The optimized prediction model Based on the SDE algorithm and the weighted BiGRU network,uses the structural parameters of the SDE global optimization network to obtain the optimal initial weight,threshold and weight coefficient firstly,and then applies the improved parameters to the optimized weighted BiGRU network model for prediction.The optimized prediction model can selectively consider the influence of past and future time on the current moment data,and effectively avoid the local optimal value and the long-range dependence of the network.The experimental results show that,compared with the other traditional neural network prediction model,the optimaized prediction model can significantly reduce the prediction error and increase the prediction accuracy.