Multi-agent Optimization Framework (MAOP) for Large Scale Process System Engineering Optimization Problems
Berhane H. Gebreslassie and Urmila M. Diwekar
Center for Uncertain Systems, Tools for Optimization and Management (CUSTOM):Vishwamitra Research Institute, Crystal Lake, IL 60012 – USA
The multi-agent optimization (MAOP) framework provides a way of combining various algorithms in one platform and exploits the strength possessed by each algorithm. Most of large scale process system engineering problems include nonlinear and non-convex problems. Contrary to the standalone optimization algorithm framework, the MAOP strategy avoids the problem of getting stuck in local optima as well as improving the computational efficiency. In this work, we propose a multi-agent optimization framework for solving complex large scale process system engineering problems. The framework uses a variety of different algorithmic agents which include the gradient based local optimizers and metaheuristic algorithms (efficient simulated annealing, efficient genetic algorithm and efficient ant colony optimization algorithms). Each agent encapsulates a particular problem-solving procedure. We investigate the effect of cooperation among agents of the multi-agent system working in parallel and combined into a framework designed to solve large scale combinatorial optimization problems. Computational experiments are carried out using benchmark problems and real world case study. The proposed methodology enables to improve the quality of solutions and the computational efficiency in comparison with non-cooperative multi-agent framework and standalone agents. We have also examined solving the optimization problems with cooperative homogenous and heterogeneous agent systems and the results indicate that the computational efficiencies are improved when the agents are heterogeneous. Moreover, the analysis of the intra- and inter agent cooperation shows that depending on the complexity of the problem the inter- and intra-agent collaboration has a significant impact on system performance. The MAOP framework which includes the five major parts of the algorithm (figure 1) is given in the figure 2.
Fig. 1. The major parts of the MAOP algorithm and the information flow direction.
Fig. 2. The basic flow diagram of the heterogeneous MAOP algorithm
1. Siirola JD, Steinar Hauan S, and Westerberg AW, 2003. Toward agent-based process systems engineering: proposed framework and application to non-convex optimization. Computers and Chemical Engineering 27: 1801-1811.
2. Gebreslassie BH, and Diwekar UM, 2015. Efficient ant colony optimization for computer aided molecular design: Case study solvent selection problem. Computers and Chemical Engineering 78: 1-9.
3. Diwekar UM, Xu W, 2005. Improved genetic algorithms for deterministic optimization and optimization under uncertainty. part i. algorithms development. Industrial and Engineering Chemistry Research 44(18) 7132-7137.
See more of this Group/Topical: Computing and Systems Technology Division