Research On The Learnable Intelligent Optimization Approaches And Its Applications

The optimization techniques have been employed to a wide range of science, engineer-ing fields and so on. These techniques have attracted a lot of attentions from the researchesin both theory and engineering fields. The optimization theory and algorithms have beena hot research issue in both theory and application areas. The intelligent optimization ap-proaches have inspired by nature, which have provided novel ideas and new methods forsolving complex problems in engineering fields. It is concluded from no free lunch theoremin optimization field that the hybridization on optimization algorithms is a very effective wayto improve the performance of algorithms. It is a significant issue by reasonably combiningthe advantages among different algorithms.Based on these existing intelligent optimization approaches, this paper proposes a basicarchitecture of learnable intelligent optimization approaches. It employs the integrated mod-elingideawhichcombinestheintelligentoptimizationmodelwiththeknowledgemodel. Theintelligent optimization model searches the feasible space of optimization problems accord-ing to the mechanism of neighborhood search. The knowledge model discovers some avail-able knowledge from the previous optimization process, and employs the obtained knowl-edge to guide the subsequent optimization. This architecture legitimately combines the intel-ligent optimization model with the knowledge model, and largely pursues the complemen-tary advantages of these two models. The architecture of learnable intelligent optimizationapproaches will provide a useful reference for the improvement of existing optimization ap-proaches.This paper proposes four different forms of knowledge: elitist knowledge, componentknowledge, performance knowledge of operators and performance knowledge of parametercombinations. Thesebasicformsofknowledgeprovideanimportantfoundationforlearnableintelligent optimization approaches. This paper constructs eight different kinds of knowl-edge for the learnable intelligent optimization approaches. These categories of knowledgecan support learnable intelligent optimization approaches to be efficient for solving complexoptimization problems.This paper designs and implements a learnable genetic algorithm to the function opti-mization problem. Twenty-one benchmark functions were applied to compare the perfor-mance of different approaches. Experimental results suggest that the learnable genetic algo-rithm outperforms three recent published approaches. This paper designs and implements five kinds of learnable intelligent optimization ap-proaches to three classical discrete optimization problems. Experimental results suggestthat the learnable intelligent optimization approach outperforms many recent published ap-proaches.To the practical engineering optimization problems, the learnable genetic algorithm andthe learnable ant colony optimization were applied to the system-of-systems simulation opti-mization problem, the scheduling problem of satellite ground station system and the missionplanning of multiple satellites. Experimental results suggest that these approaches can obtainthe satisfactory results.