First, we fill in the missing values with last-day prices in the data pre-processing section. Data mining is then performed, and relationships are used for model building through visualization and correlation analysis. The results show that the product is intrinsically correlated. Next, we designed an Encoder-Decoder Long Short Memory Network (LSTM) price prediction model architecture. Compared to the traditional LSTM model (MSE: 0.00256), our model significantly improved accuracy (MSE: 0.0000517). Finally, we traded a portfolio using the Markowitz Portfolio Optimization Model based on the predicted results, which had a cumulative return of 18,625. 460%.
Research on the Application of Time Series Forecasting Model Based on Encoder-Decoder LSTM Model
2022-10-12
1197439 byte
Conference paper
Electronic Resource
English
Probabilistic Vehicle Trajectory Prediction Based on LSTM Encoder-Decoder and Attention Mechanism
SAE Technical Papers | 2022
|