This study proposes a hierarchical optimization strategy based on intelligent algorithms to address the problems of high module coupling and cross-system data fragmentation (38% duplication reporting rate) in the scientific research management system of China Electric Power Research Institute. Methodology: First, the modular service architecture (MSA) is used to reconstruct the system, and the original 32 functional modules are divided into 5 types of reusable service units based on business characteristics. Secondly, a two-stage optimization model is constructed. In the first stage, the service deployment plan is generated by an improved genetic algorithm (module coupling weight α = 0.62, response time weight β = 0.28), and the Pareto optimal solution set is selected after 500 iterations. In the second stage, the dynamic threshold feedback mechanism is used to adjust the resource allocation strategy in real time. At the same time, a lightweight data center is designed, and the feature alignment algorithm is used to realize the automatic mapping of 9 types of core data fields with financial, material and other systems, and the event-driven architecture is used to improve the efficiency of data flow. The actual test shows that the reuse rate of system modules has increased to 78%, the function iteration cycle has been shortened to 11 days, and the cross-system data synchronization accuracy has reached 96.3%. This study verifies that the hierarchical intelligent optimization method can effectively solve the resource allocation problem in enterprise-level system reconstruction while reducing the complexity of the algorithm.