Research On Stock Index Prediction Model Based On Three-stage Attention Mechanism

Stock index sequence has the characteristics of low stability,high noise and easy to be disturbed,and there are many problems such as multiple noise and multiple interference among the constituent stocks of its data.These characteristics make the prediction of stock index become the most challenging problem in time series prediction.With the development of deep learning,more and more neural network models are applied to the prediction of financial time series.However,few models can independently select relevant sequences and make effective prediction of stock index based on the temporal dependence and spatial correlation between sequences.In order to predict the stock index effectively,the stock index prediction model combines the time convolution network and introduces a three-stage attention module in the encoder-decoder structure.The driving sequence composed of component stock price data and the target sequence composed of stock index price data constitute the training data set.First model using the time convolution network learning drive sequence,the correlation between the first stage and then through attention module learning drive sequence between the strong and dispersion characteristics of the weight,the second phase attention yourself module in the first stage module,on the basis of combining with the target sequence weight steady learning characteristics,distribution of the weight of driven by learning,network unit can independently extract relevant drive characteristic of the sequence.In the third stage,the attention module pays more attention to the historical value of the target sequence when calculating the weight,so as to learn the time dependence of the sequence itself more deeply.The model can predict the stock index effectively by selecting the historical value of correlation driving sequence and target sequence.The experimental verification was conducted on the NASDAQ 100 stock data setand the CSI 300 stock data set,and the stock index prediction model was compared with the three types of prediction methods.The experimental results show that under the same experimental environment,the stock index prediction model is superior to all comparison methods,and the stock index prediction model is more outstanding in short-term prediction than long-term prediction.