Research And Application Of Time Series Model Based On Combination Of Attention Mechanism And Neural Network

The prediction of time series data has long attracted attention from all walks of life.Its demand has become more critical in many fields,such as finance,meteorology,integration,and energy.It infers and predicts the state of the next time unit by mining the time series's trend law.The advent of massive data has transformed time-series data from a single sequence to multiple sequences and from low-dimensional to high-dimensional features.The time series forecasting problem's research method has also changed from the initial statistical regression model to the nonlinear modelling of machine learning.For more than a decade,the principal method are the deep learning models,and the typical method is Recurrent Neural Network.Aiming at stock time-series on financial market,we design an environmental attention mechanism based on the two-stage recurrent neural network model.This kind of environmental attention mechanism can adaptively pay attention to the stock price changes of other companies of the similar type in the surrounding environment when predicting the price of a single stock.The network model with environmental attention mechanism is verified by Ping An Bank's stock price forecast and compared with the benchmark model.The results show that the new attention mechanism has a progressive effect on the prediction accuracy when more environmental information is observed.Aiming at power time-series in energy consumption,we design a novel neural network sandwich structure,an improved attention mechanism was inserted into the Double-layer Long short-term memory network(LSTM)network.Such an attention mechanism can adaptively focus on the information of different features in different time units and mine the time information in the sequence through the LSTM network.This novel network structure is compared with the benchmark model on the household energy consumption data set.The results indicate that this novel network has apparent advantages in this data prediction.