A Novel Optimization Framework Integrating Multiple Initialization, Automatic Topologization and MINLP Reduction to Accelerate Large-Scale Heat Exchanger Network Synthesis

Ensuring the solution feasibility and achieving an acceptable solution time for the non-deterministic polynomial-time hard (NP-hard) problem of heat exchanger network (HEN) synthesis is still challenging. In this study, we developed a novel optimization framework for efficiently solving large-scale HEN synthesis problems to obtain near global optimal solutions, simultaneously addressing the solution's infeasibility and time. The optimization framework consists of a multiple initial HEN design model (M-INI model), a new distance-based auto structure transformer (A-STR transformer), and a reduced non-linear programming model (R-NLP model). The M-INI model generates various HEN designs based on pinch analysis by varying the minimum approach temperature and the modes of stream splitting and mixing. The A-STR transformer is first proposed to automatically convert various HEN designs into associated superstructure-based topologies as multiple initial starting points for accelerating the solution of the R-NLP model. A strategy for approximating the temperature difference calculation is further applied to reduce the computational loads. Four case studies were conducted using the proposed method to minimize the total annual cost (TAC). The solutions of general cases 1 and 2 are almost equal to the best results in the literature, while the solution time is reduced by 98.7% and 99.8%, respectively, which demonstrates the performance of the proposed method. Especially, the method was applied to two practical industrial HEN cases: a large-scale low-grade heat recovery system and a crude distillation unit. The results of large-scale case 3 show that the optimal solution near the global optimal solution only requires 251.7 s. Case 4 includes pre-splitting in scenario A and free splitting in scenario B. Compared to scenario A, the TAC and solution time in scenario B were reduced by 33.9% and 93.1%, indicating that the pre-splitting strategy excludes the optimal from the search space. The proposed novel optimization framework demonstrates the efficiency and applicability of handling large-scale HEN synthesis problems. It can also guide engineers in the design of a large-scale HEN with lower costs and effort.