The rapid evolution of emerging technologies, including the Internet of Things, cloud computing, and edge computing, has resulted in geometric growth in network traffic. Traditional network architectures are limited in their ability to provide a global perspective and real-time responsiveness, which in turn leads to network congestion and load imbalances that degrade service performance. To address these critical challenges, we propose a two-stage load-balancing routing optimization algorithm (LBROA) within the software-defined network (SDN) paradigm, which uses the global view and centralized control to increase network efficiency. First, an improved shortest path first algorithm (ISPFA) is proposed, which uses network status and traffic changes in real time, and integrates pruning techniques based on network Quality of Service (QoS) metrics to identify multiple feasible paths between sources and destinations. These paths then are conceptualized as a virtual pool of network resources, and their performance is evaluated using the comprehensive smoothness metric. Finally, a resource preallocation strategy is introduced to simulate and evaluate the adaptability smoothness of the network under diverse path allocation schemes. A greedy strategy is applied to dynamically select the optimal adaptive path from the virtual resource pool as the data forwarding route. This process facilitates the dissemination of forwarding rules, thereby guaranteeing efficient data transmission. Experimental evaluations conducted on the Mininet simulation platform demonstrate that, compared with three established algorithms, the proposed LBROA significantly improves overall network performance.