Research On Electromagnetism-like Mechanism Algorithm Based On Pattern Search With Its Applications

Many scientific research and engineering application problems could be summarized down to the global optimization problems. The characteristics of the objective functions of these problems usually present high dimensions, non-convex, non-differentiable, existing lots of local minimum and so on. For traditional optimization methods, such as secant method, the steepest descent method, cannot solve these problems or nead a long calculation time. Recently, a new approach called meta-heuristic method has the capbility of finding the near-optimal or the optimal solution at a reasonable time. Thus, more and more attentions have been attracted on the development of this new method and some classic meta-heuristic methods have been exploited, including simulated annealing algorithm, genetic algorithm etc.The main purpose of this dissertation is to explore the effective solution of the optimization problem under different situations based on electromagnetism-like mechanism algorithm (EM), which is a new kind of meta-heuristic algorithm. First, a new pattern searching-electromagnetism-like mechanism algorithm (PSEM) is propesed and an improvement is applied to reduce the disadvantages of the EM algorithm. Then, the constrained optimization problem, multi-objective optimization problem and other ones are solved by using improved algorithm. Finally, a protype of function optimization solver is designed by summarizing the theoretical and practice achievements of the optimization problem solution. The main research work and contributions of the dissertation are introduced as follows:For unconstraint function optimization problems, a modified charge formula is adopted to simply the calculation of the resultant force for enhancing the efficiency of the algorithm. Then, an effective variable step size pattern search algorithm is combined with the EM algorithm to improve the local search abilitity. Meanwhile, a disturbing point is set up in the population to increase the diversity of the algorithm at the moving stage. The numerical results show that the modified algorithm has higher precision compared with the original algorithm.For multi-objective optimization problem, the EM algrithm is improved in order to satisfy the requirements of the cases:A non-dominated set is used in calculation. The corresponding charge formula and calculation of the resultant force are proposed. Fast non-dominated sorting method and elitist strategy in NSGA-II are introduced in calculation. In this situation, the elitist strategy is modified to increase the diversity of population and avoid premature convergence. In the particles moving process, the case of overstepping the boundary has been directly set as the new boundary value. In order to validate the multi-objective EM algorithm, nineteen functions are selected to calculate the corresponding non-dominated set. It is noted that non-dominated sets found by multi-objective EM algorithm have good convergence and diversity.Considering the bad traits of some traditional neural network training algorithms, such as the slow convergence rate and the nature of local convergence, a training algorithm of neural network based on EM is presented and then it is applied to tourism demand forecasting problems successfully. Due to the complexity of the different cases, the various neural network forecasting models have been designed and developed by taking account of the econometric and time series issues. Compared with the other agrithems, the good precision of prediction can be obtained, which also proves the validity of the EM training algorithm in this application.The EM is applied in the constrained optimization problems. A processing constraints method is developed by introducing a kind of feasibility and dominance rules into the EM algorithm and modifing the charge formula of the EM algorithm. It makes the EM can be applicable from the unconstrained optimization to constrained optimization. Further, this EM algorithm for the nonlinear constrained optimization of machining parameters has been investigated for multi-pass milling operations. The results show that the proposed method is better than other algorithms and achieves significant improvement. On the other hand, a flexible cutting strategy also has been put forward to optimize the depth of cut for each pass, cutting speed and feed simultaneously. It is achieved to optimize multi-parameters at the same time.Based on the above theoretical and practical achievements, a protype of the function optimization solver mfcEM1.0has been developed to expand the applications of the EM. A series of function optimization examples has been calculated using this system in order to introduce the implementation of the system, which also shows that this system is available and efficient.In the end of the paper, the researches of the dissertation are summarized and some future research directions have been presented.