Zhang Yan-jun, Zhang Shu-guo, Fu Guang-wei, Li Da, Liu Yin, Bi Wei-hong
Institute of Information Science and Engineering, Hebei Key Laboratary of Especial Optical Fiber and Fiber Sensor, Yanshan University, Qinhuangdao 066004, China.
Guang Pu Xue Yu Guang Pu Fen Xi. 2012 Apr;32(4):915-20.
This paper presents a novel algorithm which blends optimize particle swarm optimization (PSO) algorithm and Levenberg-Marquardt (LM) algorithm according to the probability. This novel algorithm can be used for Pseudo-Voigt type of Brillouin scattering spectrum to improve the degree of fitting and precision of shift extraction. This algorithm uses PSO algorithm as the main frame. First, PSO algorithm is used in global search, after a certain number of optimization every time there generates a random probability rand (0, 1). If rand (0, 1) is less than or equal to the predetermined probability P, the optimal solution obtained by PSO algorithm will be used as the initial value of LM algorithm. Then LM algorithm is used in local depth search and the solution of LM algorithm is used to replace the previous PSO algorithm for optimal solutions. Again the PSO algorithm is used for global search. If rand (0, 1) was greater than P, PSO algorithm is still used in search, waiting the next optimization to generate random probability rand (0, 1) to judge. Two kinds of algorithms are alternatively used to obtain ideal global optimal solution. Simulation analysis and experimental results show that the new algorithm overcomes the shortcomings of single algorithm and improves the degree of fitting and precision of frequency shift extraction in Brillouin scattering spectrum, and fully prove that the new method is practical and feasible.
本文提出了一种新颖的算法，该算法根据概率将优化粒子群优化（PSO）算法和列文伯格-马夸尔特（LM）算法相融合。这种新颖的算法可用于伪沃伊特型布里渊散射光谱，以提高拟合度和频移提取精度。该算法以PSO算法为主框架。首先，PSO算法用于全局搜索，每次经过一定次数的优化后生成一个随机概率rand(0, 1)。如果rand(0, 1)小于或等于预定概率P，则将PSO算法获得的最优解用作LM算法的初始值。然后，LM算法用于局部深度搜索，并且LM算法的解用于替换先前PSO算法的最优解。再次使用PSO算法进行全局搜索。如果rand(0, 1)大于P，则仍使用PSO算法进行搜索，等待下一次优化生成随机概率rand(0, 1)进行判断。两种算法交替使用以获得理想的全局最优解。仿真分析和实验结果表明，新算法克服了单一算法的缺点，提高了布里渊散射光谱中频移提取的拟合度和精度，充分证明了新方法的实用性和可行性。
