The Research Of Swarm Intelligence Optimization Algorithms Based On Surrogate Model

Optimization problems, both in the field of engineering and in everyday life in a wide range ofapplications. Over the years, people committed to solve the optimization problem and the development ofa number of related methods, traditional, classical optimization methods including differential extremummethod, conjugate gradient method, quasi-Newton method, the common characteristics of these methodsare only as local optimization method single peak, low-dimensional simple function to achieve betteroptimization results and practical problems are usually high-dimensional, multi-extreme, non-linear, thetraditional method is difficult to effective.Swarm intelligence optimization algorithm is a class of complex problems developed in recentdecades optimization (solving) method. Different from the conventional method, such methods tend tomimic the nature of certain groups behavior, by a simple, low-wise a plurality of individual respectivesearch, and then repeat the process of the information interaction, so that the groups showed someintelligent behavior, adaptive guidance of individuals moving toward optimal solution and, ultimately,better global optimal solution. Typical of such algorithms evolutionary algorithms, particle swarmoptimization, ant colony algorithm, such extraction, such as biometric algorithm for its good robustness,excellent ability of global optimization massive data mining, large-scale combinatorial optimization Thetraditional algorithm powerlessness field achieved good results, and thus become the hotspot of research inrecent years with applications.However, the swarm intelligence algorithms do further analysis is not difficult to find this kind ofalgorithm structure there is a high degree of non-linearity, dynamic and a certain randomness. Relatedscholars have done a lot of research, but the running mechanism of the algorithm, the convergence ofcritical theory has yet to be improved, in particular, the evaluation of the merits of the individual (alsoknown as adapt degrees) to rely on no uniform evaluation criteria specific issues. Algorithm design oftenrely on specific issues and empirical knowledge, algorithm performance is heavily dependent on theexperts in the field; On the other hand, the fitness too costly for many engineering applications, makinginefficient algorithms run, which to some extent also hindered extensive use of the algorithm.Depth analysis of the structural characteristics of the swarm intelligence optimization algorithm andsearch process, and combined with the theories and methods of machine learning, analysis of thecharacteristics of the distribution of the fitness function, as well as the pros and cons of the fitness modelhas been proposed. Summarizes the existing fitness model based on the proposed one able tosimultaneously reduce the fitness computation cost and improve the accuracy of the fitness modelalgorithm-fitness Agent regression model based on the manifold, and further analysis of the algorithmtheory base and computation complexity. Next, we realize this agent model agent model using particleswarm algorithm Riemannian manifold learning and polynomial regression techniques. Finally, comparedto the standard test set of functions test our algorithm, the standard PSO and other proxy model algorithmand practical engineering examples to test our algorithm achieved a good result of the operation and tosave computing time the other two algorithms, proved the feasibility and effectiveness of the algorithm. Inaddition, the system introduces the minimalist approach of the current mainstream data, given the principle of the various dimensionality reduction method, and pointed out that the linear dimensionalityreduction and manifold advantages and disadvantages of each dimension reduction. In experiments as wellas the outlook part, given the behavior of swarm intelligence should pay attention to the angle, and furtherimprovements of the agent model.