실시간 시장 학습 기반의 암호화폐 자동 거래 시스템

초록

The cryptocurrency market is experiencing rapid growth, accompanied by significant volatility and risk. To effectively manage these risks and maximize profit potential, automated trading systems are becoming increasingly popular. Unlike traditional algorithmic trading methods, this paper proposes an automated trading system based on reinforcement learning that can adapt to market volatility by learning from real-time conditions. This approach mitigates the overfitting problem with training data and allows for the adaptive updating of the neural network model with the latest market data. The proposed method overcomes the limitations of traditional single training and trading interval approaches by updating the model weights periodically with the closest possible data to the trading instance. This adaptability minimizes overfitting and adjusts to new data patterns. Various simulations have confirmed the performance of the proposed method, showing a 1.71% to 9.74% higher profitability compared to traditional single data usage methods.

키워드