Efficiency optimization of enterprise resource planning based on deep reinforcement learning: achieving more efficient business process automation

Enterprise Resource Planning (ERP) systems play a key role in modern enterprise management. Business Process Management (BPM), as a core subsystem of ERP, is responsible for coordinating and automating enterprise workflows. However, traditional BPM systems struggle with efficient task scheduling under dynamic load changes. This often leads to reduced execution efficiency, especially under heavy loads. When the load exceeds the processing capacity of the engine cluster, it can cause severe Service Level Agreement (SLA) violations. To address the above issues, this study optimizes the task scheduling architecture for Business Process Management (BPM) in cloud environments from the perspectives of system modeling and scheduling mechanism optimization. The service-level-aware task scheduling process is formalized as a Markov Decision Process (MDP). On this basis, a Service-Tiered Request Scheduling (SRS) framework is constructed. The Proximal Policy Optimization (PPO) algorithm is adopted as the policy learner to realize adaptive scheduling decisions for tasks of different service levels. In service-tiered load tests, results show that under normal load, the Service Level Agreement (SLA) violation rate of SRS-PPO is 6.6%, slightly higher than that of the traditional Earliest Deadline First (EDF) algorithm. Nevertheless, the Quality of Service (QoS) over-provisioning rate of SRS-PPO is reduced by 74.57% compared with EDF. Under heavy-load conditions, the SLA violation rate of SRS-PPO is 21.33%, which outperforms the comparison methods and is 7.04% lower than that of EDF. Its QoS over-provisioning rate is 10.12%, a decrease of 60.81% compared with EDF (p < 0.05). Experimental results demonstrate that SRS-PPO has application potential in handling SLA violations in high-load environments, improving task scheduling flexibility and resource utilization efficiency.