BackgroundAlthough cloud computing has been increasingly adopted in university information systems to support teaching, research, and administration, existing cloud service platforms still face significant challenges in two key dimensionsintelligent resource scheduling and comprehensive security protection. ProblemMost current resource allocation strategies rely on static models or heuristic-based rules, which lack adaptive capabilities in dynamic and high-load environments. Simultaneously, existing cloud security mechanisms often focus on traditional encryption methods or basic access control, which are insufficient for protecting sensitive data in open, high concurrency subsystems such as digital libraries. Proposed solutionTo address these gaps, this paper proposes an intelligent architecture optimization and deep security design scheme tailored for university cloud platforms. The core innovation lies in the integrated application of deep reinforcement learning (DRL) and advanced cryptographic frameworks to achieve dynamic, secure, and efficient service delivery. MethodologyDRL is employed to build a system-aware, environment-adaptive resource scheduling algorithm that maps complex system states to optimal resource configurations in real time. For the security framework, this paper introduces a multi-layered encryption and access control strategyattribute-based encryption (ABE) ensures secure data transmission, while the scheme is extended to Multi-Authority Ciphertext-Policy ABE (MABE) based on the Ring-LWE assumption, enabling decentralized identity access management (IAM) with formal security guarantees. In addition, a K-Means-based network log analysis mechanism is developed for real-time detection and classification of user behavior and potential attacks, forming a closed-loop system of threat perception, behavior recognition, and active response. ResultsThe experimental results show that the K-Means + ABE attribute-based encryption algorithm can be applied to the network information security management of LAN in terms of data. Compared with the classical heuristic-based Particle Swarm Optimization (PSO) algorithm and greedy algorithm, the fitness function value of the new algorithm in this paper is 4.4% and 5.6% higher on average, respectively, which can effectively balance the demand for QoS and resource cost. SignificanceTherefore, the optimization scheme is superior to the traditional architecture design in response efficiency, resource utilization, and security performance. This paper provides a feasible theoretical basis and technical path for the construction of cloud platform under the background of university digital transformation.