Research On Forecasting Of New Energy Stocks Price Based On LSTM Hybrid Model

With the rapid development and progress of deep learning related technologies and theories,related algorithms have been applied to stock price prediction.Because of its powerful learning ability and other advantages,deep learning often performs better than traditional prediction methods in stock price prediction.The new energy industry has the characteristics of environmental protection and energy saving,so its development prospects are bright.The government’s policies also promote the development of the new energy industry.By predicting the price trend of new energy stocks,it can help investors make appropriate decisions,reflect the development of the new energy industry to a certain extent,and promote the effective allocation of related resources.This study is carried out under several important assumptions,and 20 stocks involved in various fields of the new energy industry are selected as the data set,which ensures the rationality of the forecast in the context of the industry to a certain extent.The dataset contains20 features commonly found in stock price prediction tasks.Through the optimal selection of the feature data set and the improvement of the prediction model,the prediction accuracy of the new energy stock price is improved.New energy stocks are more popular,and this study has a good prediction effect on short-term stock prices.The specific research contents are as follows:(1)Aiming at the improvement of input features for model training,this paper proposes two feature selection methods Lasso regression and Extremely randomized trees to filter more effective input features.Based on the benchmark model LSTM,the prediction results of the two feature selection methods are compared,and it is found that the prediction results of the two feature selection methods are better than the prediction result of the original data set.The prediction effect of the feature data set generated by the Extremely randomized trees is better,and the RMSE of the prediction result is reduced by 6.18% compared with that of the original data set,which improves the prediction accuracy.(2)Aiming at the improvement of model prediction performance,this paper proposes three hybrid models: LSTM-Attention,DALSTM and CNN-LSTM-Attention,which are based on the benchmark module LSTM and combined with the CNN module and attention mechanism.By comparing the prediction results of different models based on the optimal feature data set,it is found that the prediction results of the three hybrid models are better than the prediction result of LSTM.The CNN-LSTM-Attention model has the best prediction effect,and the RMSE of the prediction result is 2.88% lower than that of LSTM,which improves the prediction accuracy.