Optimizing Cryptocurrency Portfolio Management: Deep Learning with Diverse Data Sources and Multiple Parameters

Abstract

Portfolio optimization is a key area in financial engineering, aiming to maximize investor returns. In this study, we propose an innovative portfolio management system based on Deep Reinforcement learning. Unlike traditional approaches, our system incorporates five parameters, including trading volume, for a more comprehensive market analysis. Furthermore, we leverage multiple data sources simultaneously, providing a more robust perspective. Our approach also stands out by utilizing specially designed deep convolution to separately address each parameter. Additionally, we introduce a five-dimensional attention gating network to better identify critical moments and high-potential assets. Simulation results demonstrate that our approach significantly outperforms traditional methods, offering investors more substantial returns and reduced short-term risk.