ABSTRACT
Optimization of nonlinear systems has been an important issue for many problems across various disciplines. Many strategies based on gradient-based search techniques have been developed so far for optimization of nonlinear problems; however, in general, they have some drawbacks. First, the classical quantitative gradient-based methods are subject to numerical problems such as convergence to local minima or divergence when the problem to be optimized is highly nonlinear or ill-defined. Second, the accuracy of the optimization result will strongly depend on the fidelity of the models used, which often cannot warrant precise predictions due to the complexity and multimodality of the system or process. If a portion of the model is not precise, it will affect the entire optimization accuracy due to the inversion of Jaconbian matrix. Third, only quantitative models can be used, thereby ruling out the possibility of using heuristic rules or empirical data, which, in some cases, are the only ways to describe the problem. Last, computing time for optimization is very long. Due to these reasons, there exists a need to develop an intelligent optimization scheme that is flexible such that different models can be easily incorporated into it and it is more efficient in optimal solution searching.
