As an efficient optimization tool, meta-heuristic algorithms have been widely used in the field of task scheduling, owing to their advantages of not relying on specific problem models and efficiently exploring optimal or near-optimal solutions—a key reason they are favored for solving complex scheduling tasks. However, with the rapid development of computing power networks, two critical challenges have become prominent: the increasing complexity of heterogeneous nodes and the exponential growth of data volume. These changes have seriously challenged the performance of existing meta-heuristic algorithms: many struggle with local optima, slow convergence in large-scale scenarios, or failure to balance latency and resource utilization, making it hard to meet increasingly complex application requirements. To address this dilemma, many scholars have focused on developing and improving meta-heuristic algorithms, often by optimizing search mechanisms or enhancing parameter adaptability to boost their applicability and efficiency in complex environments. Among emerging algorithms, the Artificial Lemming Algorithm stands out due to its excellent global search ability and strong adaptability to dynamic tasks, attracting wide attention. Still, it has limitations, such as insufficient convergence precision in late iterations. Fortunately, this paper proposes an Improved Artificial Lemming Algorithm. Through a series of comparative experiments, the improved algorithm shows excellent task scheduling ability: it not only overcomes existing shortcomings but also achieves efficient, stable optimization in complex networks—reducing task latency while improving resource utilization—providing a new solution for task scheduling problems.