An Improved Particle Swarm Optimization Algorithm And Its Application In Mechanical Design

Particle Swarm Optimization(PSO)algorithm uses the strategy of swarm intelligence,and it belongs to a branch of intelligence optimization algorithms.Compared with some traditional computation methods,PSO has a number of advantages,and it can cope with both easy problems and difficult issues without setting initial values.While Artificial intelligence has risen to national strategy in 2017,intelligent optimization algorithms have also received increasing attention from researchers in various fields.Its related applications and researches are gradually becoming systematic,and the prospects are very broad.The basic PSO was proposed in 1995,and it has some disadvantages: slow convergence speed,low convergence precision,and rapidly weakened optimization effect while increasing the difficulty of the problem.In order to overcome these shortcomings,many advanced strategies have been proposed by the predecessors,and their strategies have achieved better performance improvements in a certain direction.Hence,this paper has done the following research:For one thing,in order to overcome the shortcomings of the basic PSO,especially slow convergence speed,low convergence precision and poor robustness,two improved PSO are presented e.g.A Novel Simplified Particle Swarm Optimization(NSPSO),Selfadapted Simplified Particle Swarm Optimization based on Random-weight(SRSPSO).The former only uses the global best solution to guide the particles for optimization,and it excludes the velocity updating formula.The latter combines locking factor using the mechanism of tangent function and a random inertia weight using self-adapted strategy,and these improvements are based on NSPSO.For another thing,twenty benchmark functions and ten kinds of PSOs are introduced.These ten PSOs and two improvements proposed in this paper are enabled to search for10-dimensional and 100-dimensional benchmark functions,and the results are analyzed with some scientific methods.Wilcoxon rand sum test is used to analyze the difference between optima.The time and space complexity of the algorithm is analyzed using the large O method.Expected precision are set according to the result for solving 100-dimansional problems.The success rates and average iteration times that can be calculate by expected precision can verify the capability of convergence for algorithms.The theoretical results show that NSPSO and SRSPSO have the characteristics of fast convergence,high convergence precision,strong exploration and development capability,strong robustness.These two algorithms' time and space complexity are not changed compared with the basic particle swarm optimization algorithm.In the end,four mechanical design problems are used to testify the performance of NSPSO and SRSPSO,and they are fitting cylinder cam profile,predicting project tremor based on material physical quantity,optimum design of Machine-tool spindle parameters and cutting parameters optimization for Titanium alloy turning machining,respectively.The first two applications are essentially the curve fitting.These problems models include unconstrained problems,constrained problem and multi-objective problem.The application results show that NSPSO and SRSPSO can handle those three kinds of problems well.Meanwhile,SRSPSO performs stronger stability than NSPSO.This paper has 15 pictures,23 tables and 120 references.