Research on Stock Prediction Based on CED-PSO-StockNet Time Series Model

Abstract

In view of the complexity and uncertainty of the stock market, especially the noise interference in the stock data, the traditional single prediction method has been difficult to meet the needs of investors. This paper innovatively proposes the CED-PSO-StockNet time series model to improve the accuracy of stock forecasting. The model first introduces the complete ensemble empirical mode decomposition (CEEMDAN) technology, decomposes the original stock data, estimates the frequency of each component through the extreme point method, and recombines it, so as to effectively remove the noise. Then, the model uses the Encoder-Decoder framework which integrates the attention mechanism to accurately predict the reconstructed components, so as to better extract and use the data features. In addition, this paper also uses the improved particle swarm optimization algorithm to optimize the model parameters. Through five groups of comparative experiments, the effectiveness of each part of CED-PSO-StockNet model is verified, showing its significant advantages in stock forecasting.