Chemical engineering related problems, such as production scheduling or process synthesis, are often solved by mathematical programming methods like mixed integer linear programming (MILP) or mixed integer non linear programming (MINLP). These problems are often highly combinatorial and might involve nonlinear functions making them very time consuming to solve by these methods and prone becoming trapped at local optimums....[ Read more ]

Chemical engineering related problems, such as production scheduling or process synthesis, are often solved by mathematical programming methods like mixed integer linear programming (MILP) or mixed integer non linear programming (MINLP). These problems are often highly combinatorial and might involve nonlinear functions making them very time consuming to solve by these methods and prone becoming trapped at local optimums.

In contrast, evolutionary approaches such as Genetic Algorithm, that search a wider spectrum of the solution space, allow the searching process to overcome local optimums if sufficient searching time is provided. These approaches are often proposed for global optimization. However, the searching time for a global optimum is still a lengthy task and often unaffordable.

To improve the solution quality and to shorten the solution time, parallel evolutionary approaches are being developed and evaluated in this research. These new approaches incorporate different decomposition and solution techniques in a parallel computing environment. Benchmark problems such as batch scheduling, optimization under uncertainties and heat exchanger network synthesis are used to evaluate these new approaches.

In the proposed approaches, a problem is first formulated in a way that can be solved by a conventional sequential evolutionary method. Then the formulated problem is decomposed into a number of sub-problems that are solved in parallel. Different decomposition approaches are tested and compared using the benchmark problem. Results show that, with parallel computing, the solution time required is significantly shorter, or, in other words, we can obtain a better solution in a much shorter computing time.