Stock Prediction based on BP Neural Network

Abstract

In this paper, using Wind database, the stock code 600276 is selected as the main research object, and the daily K data from 15 October 2018 to 14 October 2022 is selected. The variables in the article are Open Price, Close Price, High Price, Low Price, Volume, and Amount. Firstly, the data is normalized and then a BP neural network model is used for training. In the model after several training sessions, the reciprocal 20 pieces of data in the "opening price" variable are selected as the data for the prediction set to observe the training. Finally, the results of the test set of real stock price tests are introduced and the predicted results are visualized. It is concluded that when the hidden layer nodes of the neural network are fewer, the structure of the neural network is too simple, then its learning ability and classification ability will be reduced, but if the hidden layer nodes are too much, the structure of the neural network is too complex, the network is overloaded, and the efficiency will be reduced, and the ability of the promotion will be deteriorated. Therefore, neural network training should ensure the classification ability of the neural network on the one hand and the promotion ability of the neural network on the other hand.