Supply chain logistics optimization faces mounting challenges in the era of global commerce, particularly in handling dynamic demand patterns and complex path planning requirements. Traditional optimization methods often fall short in addressing these challenges, especially when dealing with seasonal fluctuations and real-time decision-making needs. This study develops an integrated solution leveraging advanced deep learning technologies to enhance supply chain efficiency and responsiveness.We design a three-module collaborative algorithm: an LSTM-CNN hybrid model for demand prediction, a double deep Q-network for inventory optimization, and a graph attention network for path planning. The system was trained and validated using comprehensive datasets from 2020 to 2024, encompassing over 80 product categories across multiple industry sectors. The LSTM-CNN module incorporates attention mechanisms to handle promotional events, while the path planning module optimizes delivery routes considering real-time conditions.Real-world enterprise application validation demonstrates significant improvements: inventory turnover increased by 25.1%, logistics costs reduced by 23.7%, and on-time delivery rate reached 96.7%. The system showed particular strength in handling seasonal fluctuations, maintaining high service levels even under demand variations of ±30%. This collaborative algorithm effectively solves complex scenarios and fluctuating demand problems that traditional methods struggle to address, providing data-driven decision support for enterprises to achieve precise logistics management.