<p>In the context of the vigorous development of the global digital economy, e-commerce platforms, as the core hub connecting production and consumption, have increasingly prominent energy consumption issues in their supply chains. According to statistics, the energy consumption of China’s e-commerce logistics and warehousing sector accounts for 3.2% of the total energy consumption of the whole society, with an average annual growth rate of 8.5%. The traditional extensive management model has been difficult to adapt to the requirements of the "double carbon" strategy. In view of the multi-link, strong coupling, and dynamic characteristics of e-commerce platform supply chain, this study proposes an energy efficiency optimization framework that integrates deep reinforcement learning and digital twin technology. By constructing a three-dimensional simulation model including order allocation, warehousing temperature control and transportation route, the equipment power, environmental parameters and order data are collected in real time, the dynamic load is predicted by LSTM network, and the improved DDPG algorithm is used to dynamically adjust the equipment running state. Experiments show that this method can improve the comprehensive energy efficiency of the supply chain by 19.7%, reduce the carbon emission intensity by 14.3%, and increase the inventory turnover efficiency by 12% in the pilot application of a leading e-commerce platform. The research results provide theoretical support and technical paths for the intelligent low-carbon transformation of complex supply chain systems.</p>