Reinforcement Learning Agent-Driven Transportation Emission Reduction Forecasting and Autonomous Carbon Credit Trading Framework

Key Points

• The research aims to integrate forecasting of carbon emissions in transportation with automated trading of carbon credits using reinforcement learning.
• Designed a decision-making structure based on reinforcement learning using multivariate time series transportation data.
• Utilized parameters such as vehicle distance, speed, load, traffic conditions, and weather factors as state space inputs.
• Conducted comparative experiments applying PPO, SAC, and DDPG algorithms to evaluate performance.
• Integrated blockchain-based smart contracts for transaction verification and to enhance security.
• Achieved a prediction accuracy of 93.2% and a profitability index of 118% with the PPO model.
• Demonstrated a 9% and 18% improvement in performance over traditional manual trading methods.
• Reduced transaction verification time by approximately 23% through blockchain technology.

Abstract