基于深度强化学习的均衡SaaS软件运行负载的方法

针对软件运维时需要保持SaaS软件运行负载均衡的挑战性问题，研究基于深度强化学习的均衡软件运行负载方法。研究内容包括：均衡运行负载的框架、监控与判定运行负载均衡状况的方法、均衡运行负载的方法。采用软件监控技术，对大型SaaS软件及其运行平台进行一体化精准监控，实现对SaaS软件运行负载状况和部署状况的有效感知和准确判定。采用深度强化学习技术，学习均衡SaaS软件运行负载的策略，实现负载均衡控制过程的自动化。通过蒙特卡洛树搜索技术，集成深度强化学习系统内部的多个学习模型和学习智能体，实现运行负载均衡解决方案的智能化。项目在把深度强化学习技术引入到SaaS软件运行负载均衡的研究方面，具有研究内容上的创新性；在融合软件一体化监控、深度强化学习和蒙特卡洛树搜索方面，具有技术路线上的创新性。项目研究工作对于促进深度强化学习的知识和应用创新具有研究意义；研究成果对于提高软件运维管理能力具有应用价值。

Aiming at the challenging problems of maintaining SaaS software's load balancing in software operation and maintenance, the approach of balancing software operating load is studied based on deep reinforcement learning. Our research includes the framework of load balancing, the method of monitoring and determining the status of load balancing, and the method of balancing software load. By applying techniques of monitoring software, the large-scale SaaS software and its supporting platform of operating are accurately monitored in an integrational mode, so as to achieve effective perception and accurate determination of SaaS software's operating load and deployment status. By adopting deep reinforcement learning techniques, the policy of balancing the operating load of SaaS software is learned, so as to realize the automation of the load balancing control process. By leveraging the Monte Carlo tree search technique, multiple learning models and learning agents inside the deep reinforcement learning system are integrated together, so as to realize the intelligence of the solution for balancing SaaS software's operating load. The novelty of this project lies in two aspects: in terms of innovations in research content, deep reinforcement learning techniques are introduced to the research of SaaS software's load balancing; in terms of innovations in technical route, software monitoring in integrational mode, deep reinforcement learning and Monte Carlo tree search techniques are integrated together effectively. The research work of this project has research significance for promoting the knowledge and application innovation of deep reinforcement learning, and the research results have application value for improving the capabilities of software operation and maintenance management.