Dynamic scheduling algorithm for container resources based on reinforcement learning

Dynamic scheduling of container resources is a central problem in cloud computing and edge computing. Considering that there is dynamicity and complexity, this work puts forward a reinforcement-learning-based scheduling algorithm. Based on a mathematical model construction, the scheduling problem is converted into a dynamic optimisation problem, and an adaptive scheduling method with the proximal policy optimisation (PPO) algorithm is designed to optimise resource utilisation, load balancing, and task completion time. Experimental results demonstrate that the algorithm is effective in different loads, achieves significant improvements in resource utilisation, reduces load variance effectively, minimizes task completion time significantly, and displays strong robustness under burst loads. It approximates the policy function by a deep neural network and achieves efficient resource allocation by integrating dynamic environmental interaction. It offers a new technical approach to resource management of containerised platforms. This work reinforces the value of reinforcement learning in complicated scheduling environments and offers theoretical support and practical guidance for constructing efficient and stable containerised systems.