부산대학교 도서관

The short-term memory cellular neural network (LSTM) provides a new idea for stock price prediction and other complex sequential data. the characteristics of LSTM cyclic neural network and stock market are integrated, and the samples are preprocessed, such as interpolation, wavelet denoising and normalization. the constructed multi-layer LSTM neural network and the multi-layer LSTM neural network based on the same layer LSTM are trained and verified. The least squares layer and least squares layer of the optimal least squares wavelet packet are found by comparing the evaluation index and the prediction results of the model. Results: The prediction accuracy of this method can be increased by about 30%. Experimental results show that this method has low computation cost. The accuracy of forecasts has improved. It can not only help investors to judge the development trend of the securities market effectively before investing in securities, but also help investors to better understand the real situation of the securities market and make reasonable decisions on securities investment.