Research On Stock Price Prediction Method Based On RNN-CNN Mode

The stock market is an important part of the financial sector.Deep learning can adapt to the nonlinear,multi-scale and high noise characteristics of stock data,so it has become a research focus of stock price prediction in recent years.The existing deep learning model has some problems in stock prediction,such as time delay,large series fluctuation and low fitting,and does not fully consider the influence of different factors on stock prices.In order to solve the above problems,this paper proposes a stock forecasting structure based on RNN-CNN architecture.And fully consider the stock technical indicators,as well as public opinion,news information on the stock impact.The main work and innovation points are as follows:(1)Aiming at the problem of only considering the historical trading data of the stock market,this paper analyzes the influence of macroeconomy on the stock market and introduces the stock technical index.The deep learning model is easy to be affected by feature dimension and noise,which leads to the reduction of stock prediction accuracy.The RNN-CNN prediction architecture model is proposed.It is noted that there are multiple variants of RNN.After consulting,no performance comparison of RNN and its variants in stock prediction is found.Therefore,in the framework of RNN-CNN,the accuracy of different variants of RNN in stock prediction is compared and analyzed.(2)Aiming at the influence of public opinion and news on stocks,a stock prediction model based on multi-source feature fusion is proposed.Based on Bert huge corpus,transfer learning is used to mine stock news features.Use Attention to give different weights to different news on the same date.(3)In view of the influence of RNN-CNN architecture model parameters on model prediction results,particle swarm optimization algorithm(PSO)was used to optimize the selection of model hyperparameter.Considering that the value of inertia factor and learning factor in PSO is fixed,it cannot satisfy the problem of global and local search ability of the balance algorithm,so we try to optimize the inertia factor and learning factor in PSO.The optimized PSO is used to select the super parameters of the RNN-CNN architecture model to improve the prediction accuracy of the model.Through agriculture,forestry,husbandry and fishery plate and part of the stock data for empirical research.The results show that:(1)Compared with the existing models,the RNNCNN architecture model proposed in this paper has the best performance in multiple evaluation indexes.And the introduction of technical index can improve the prediction accuracy of the model obviously.(2)The fusion of multi-source features affecting stock prediction not only solves the problem of imperfect feature extraction under multi-source input.Moreover,the unstructured stock data is processed reasonably,and the structured index is integrated to enrich the feature extraction and improve the prediction accuracy.(3)Improved PSO to optimize RNN-CNN model hyperparameters,reduce human-caused errors,and improve the prediction accuracy.And through the stability test,it proves that the forecast model proposed in this paper has high stability,which provides valuable reference for the stock price forecast.