Parallel network system architecture is an important mode of production system, which is widely present in various fields of economic and social development, and the research on this kind of system has important theoretical and practical significance. Among them, the optimization of internal resources is an important topic in the study of parallel network systems. In the process of realizing the future development goals of parallel systems, whether the allocation of human, material and financial resources is reasonable or not is directly relates to the scale, economic benefits and development prospects of the enterprise. Therefore, it is necessary to study the internal optimization methods of parallel network systems. Currently, as an important network system optimization method, the parallel network DEA method still has the following shortcomings. On the one hand, due to the use of all subunits to construct the technological frontier of the evaluated decision unit (DMU), the original parallel network DEA method may cause the projection of the DMU to exceed its production frontier, leading to the bias of the efficiency measurement results and the technical infeasibility of the improvement strategy. On the other hand, the existing network DEA methods have not yet systematically discussed issues such as the improvement pressure, cost constraints, target tasks of DMUs, and the priority order of subunit improvements. In order to solve the above problems, this paper firstly introduces the typical parallel network DEA methods and their related problem analysis in detail, and a modified parallel network DEA model is proposed by considering the constraint relationship between DMUs and their subunits. Secondly, on the basis of the modified parallel network DEA model, the parallel network DEA model considering subunit pressure dispersion, subunit optimization order, and cost constraint-objective task is given sequentially according to different adjustment requirements and objectives. In addition, the relevant properties of the proposed models are discussed. Finally, in order to satisfy the situation where the decision maker has multiple decision-making requirements, the comprehensive network DEA models for the internal optimization of parallel systems are further given by combining various factors. In order to further verify the rationality and feasibility of the method proposed in this paper, these models are applied to analyze the efficiency of scientific and technological R&D of university platforms in China. The empirical results show that compared with the original parallel network DEA method, the method of this paper has the following advantages in the efficiency measurement and optimization analysis of the system. Firstly, the method proposed in this paper can provide more reasonable efficiency measurement results and a feasible overall improvement plan. Secondly, the method proposed in this paper can solve the problem of internal resource optimization under various situations according to different real needs. In conclusion, the efficiency measurement and resource optimization analysis under the proposed method in this paper better meets the multiple needs of managers. This has important theoretical significance and application value for further enriching the efficiency measurement of parallel network DEA methods and improving the internal optimization strategy of the system. The method proposed in this paper is only a useful attempt in the evaluation of parallel network systems, and there are still some incomplete considerations. Further meaningful research in the future includes: 1) constructing a heterogeneous parallel network DEA model considering the problem of inconsistent internal structure among parallel systems; 2) extending the proposed methodology to complex chain-parallel network structures considering the case of multiple parallel network structures within the actual system; (3) considering a parallel network DEA model with multiple objective constraints.