Optimization problems are wide spread in many practical fields, such as production scheduling, transportation, information processing, financial management, etc. Solving these problems effectively not only can produce huge economic benefit, but also has important social significance. Optimization, based on mathematics, aims to find the optimal solution of the real life optimization problem. Traditional optimization methods, such as iterative method, Newton method, conjugate gradient method, are based on mathematics, which have a strict requirement on the model to the problem and usually require constraints and objective function is continuously differentiable. With advances in technology, the optimization problems in real life become more and more complex. Traditional optimization methods can not adapt to these problems. Therefore, seeking the optimum method to solve the problem of complex becomes crucial. Evolutionary algorithms are derived from observation and simulation of nature and biological phenomenon Compared with traditional optimization methods, evolutionary computation as one of the famous bionic intelligent optimization methods usually has a strong adaptability, robustness and parallel processing, which is widely used in many fields, such as scientific research and industrial production. Recent years, hybrid algorithms can overcome the limitations of a single algorithm, which becomes a new research field in evolutionary algorithms. The research of this paper mainly focus on the particle swarm optimization algorithm and the algorithm of center of gravity, the research can be summarized as follows:a). Aiming at solving the problem of low convergence rate and reduction of population diversity, which leads to poor global search ability, this paper proposes a new hybrid algorithm CF-PSO, mixing the particle swarm algorithm and center of gravity algorithm. By taking advantage of the fast convergence speed of particle swarm optimization and global searching ability of center of gravity algorithm, the efficiency of this algorithm is improved. To the maximum the advantages of two algorithms, the probability p is self-adaptive, which enables the algorithm to choose particle swarm update strategy and center of gravity update strategy adaptively in the iteration process.b). When entering local search phase, evolutionary algorithms often fails to find the global optimal point because the step is not easy to control. Traditional optimization methods have great advantages in local search Aiming at this problem. This paper, by using difference quotient instead of gradient, and DFP in iteration process to realize local search, can finds the current rapid and efficient search area of local optimal, and then finds the global optimal solution.c). For late iteration algorithm, the particles usually trap in local optimal solution, but canâ€™t jump from the current search area of the problem, new algorithm are also not exceptional. In this paper, through analyzing the characteristics of Gaussian mutation and Cauchy mutation in the late iteration algorithm, Cauchy mutation is used in late iteration to escape from the local optimum and continue to explore the global optimal solution in other areas.In order to test the performance of the new algorithm, we test it in the CEC2005 test function set. The results show that our algorithm is stable, and more efficient. At the same time, the improved algorithm is applied to the TSP problem, which also shows that the new algorithm can effectively solve this kind of practical problems. |