Research On The Exploration And Exploitation Strategies Controled Intelligent Optimization Approaches And Its Application

Many problems arising in science and engineering fields can be transformed into mathematical models and the solving of such models, as a result, optimization methods, for a long period, are always hot topics in theory and application areas. Intelligent optimization approaches(IOAs) inspired from nature and characterized as being simple, efficient and applicable have provided novel ideas and tools for the solving of complex optimization problems. Rational exploration and exploitation strategies which adjust the global and local search behaviors of IOAs, play key roles in the design of high-performance IOAs. This paper investigates the control issue of exploration and exploitation strategies for IOAs from three aspects, namely, variable relationship, effective parameter configuration and appropriate search strategy selection. Six novel IOAs with controlled exploration and exploitation strategies are proposed and applied to theoretical and practical optimization problems.(1) Variable relationship based strategies are proposed for enhancing the exploration and exploitation capabilities of IOAsVariable relationships of optimization problems under the optimal condition are studied. The abstracted variable relationships are used to reduce the complexity of optimization problems, thus strengthen the exploration and exploitation capabilities of IOAs. First, a variable reduction strategy is presented according to the variable relationships obtained from the optimal condition of unconstrained and first-order derivative optimization problems. Second, an equality constraint and variable reduction strategy is proposed on the basis of the variable relationships mined from the equations expressing the equality constraints of constrained optimization problems(COPs). Third, a local search strategy is designed by using the variable relationships hided in the active inequality constraints of COPs. Each strategy is combined with one or several IOAs and verified by a serial of benchmark optimization problems. Experimental results show that the proposed strategies can significantly improve the performance of IOAs.(2) The equality constraint and variable reduction strategy is integrated with IOAs and applied to practical optimization problems in power systemsThe transmission pricing problem, static economic load dispatch problem and dynamic economic load dispatch problem in power systems are constrained optimization problems with complex objective functions and constraints. By utilizing the equality constraints in these optimization problems, specialized equality constraint and variable reduction strategies are proposed to support IOAs to solve these problems. Experimental results demonstrate that the proposed strategy can remarkably improve the efficiency of IOAs in solving the aforementioned optimization problems in power systems.(3) A parameter adapted across neighborhood search algorithm is put forward for numerical optimization problemsThis paper firstly put forward a novel across neighborhood search(ANS) algorithm, in which a group of individual can search the solution space by crossing neighborhoods of multiple high-quality solutions. Theoretical analyses and experimental simulations show that ANS meets following three standards: simple principles and connivent application, unique optimization mechanisms, and high performance in solving optimization problems of different types. The exploration and exploitation capabilities of ANS highly depend on the parameter setting. Therefore, to enhance the performance of ANS, a parameter adaptation strategy studied. In the parameter adapted across neighborhood search(PAANS) algorithm,the parameter values are dynamically self-adapted by learning from their former performance in generating promising solutions. PAANS is applied to benchmark numerical problems and the obtained experimental results demonstrate that PAANS outperforms ANS and several other PSO(Particle Swarm Optimization) and DE(Deferential Evolution) versions.(4) A parameter adapted simulated annealing algorithm is proposed for the coordinated planning problem of earth observation resourcesThis paper studies the coordinated planning problems of heterogeneous earth observation resources including satellites, UAVs(unmanned aerial vehicles) and airships. The coordinated planning problem is transformed into a task assignment problem. A parameter adapted simulated annealing(PASA) algorithm is designed to solve this problem. In PASA, a tabu list and a dynamic temperature setting strategy are incorporated in order to control the exploration and exploitation abilities of PASA. Experimental results show that PASA can solve the coordinated planning problem efficiently.(5) A dynamic selection of search strategies based particle swarm optimization is presented for numerical optimization problemsTo enhance the exploration capability of particle swarm optimization(PSO), a superior solution based comprehensive learning strategy and an individual level based mutation strategy are designed. In addition, to strengthen the exploitation capability of PSO, four local search strategies are incorporated into PSO, including gradient based BFGS and DFP, as well as derivative-free Pattern Search and Nelder-Mead simplex Search. The search strategies are dynamically selected in terms of the search states of algorithms, namely, by determining whether the PSO gets trapped in local optima or evolves in its later stage. The dynamic selection mechanism of search strategies realizes the rational control of exploration and exploitation of PSO. Extensive experiments verify the high performance of the proposed dynamic selection of search strategies based PSO.(6) A dynamic selection of search strategies based ant colony optimization is proposed for satellite observation scheduling problemThis paper proposes a hybrid ant colony optimization mixed with local search strategy(ACO-LS) for solving satellite observation scheduling problem(SOSP). The scheduling procedure is partitioned into two phases, namely task clustering phase and task scheduling phase. In the task clustering phase, a task clustering graph model and an improved clique partition algorithm are designed in order to generate clustering tasks. In task scheduling phase, an acyclic directed graph model is constructed and hybrid ACO algorithm is presented to obtain optimal or near optimal solutions. As ACO has relatively strong exploration capability, to enhance its exploitation capability, an local search strategy is employed to further optimize the solution produced by ants in each iteration. Experimental results show the superior performance of ACO-LS in solving different SOSP instances, and the task clustering strategy can improve the scheduling efficiency for SOSP.