深圳大学学报理工版

针对将连续域蚁群优化算法应用于最大似然（maximumlikelihood，ML）估计中存在计算量过大的问题，提出一种基于改进蚁群优化（modifiedantcolonyoptimization，MACO）算法的最大似然波达方向（maximumlikelihooddirectionofarrival，ML-DOA）估计方法．采用精英反向学习策略获得较优初始解群体，结合全局跨邻域搜索和高斯核函数局部搜索对蚁群的寻优方式进行优化，扩大了算法的搜索空间并加快了收敛速度，最终得到ML估计方法的非线性全局最优解．仿真结果表明，与基于粒子群优化（particleswarmoptimization，PSO）算法、蚁群优化（antcolonyoptimization，ACO）算法的ML估计方法相比，ML-MACO算法的收敛速度是ML-ACO算法的4倍，计算量是ML-ACO算法的1/3，分辨成功率高于ML-PSO算法和ML-ACO算法，估计误差小于ML-PSO算法和ML-ACO算法．ML-MACO算法以更低的计算量保持了ML算法的优良估计性能，收敛性能更优且估计精度更高．

In order to alleviate the huge amount of calculation when using the continuous domain ant colony optimization (ACO) algorithm to handle the maximum likelihood (ML) estimation problem, we propose a maximum likelihood direction of arrival (ML-DOA) estimation method based on modified ant colony optimization (MACO) algorithm. Firstly, MACO algorithm adopts the elite reverse learning strategy to obtain a better initial solution group. Secondly, the optimization method of ant colony is conducted by combining global cross-neighborhood search and local search of Gaussian kernel function to expand the algorithm search space and accelerate the convergence speed. Finally, the nonlinear global optimal solution of ML estimation method is obtained. Simulation results show that, compared with ML estimation methods based on particle swarm optimization (PSO) algorithm and ACO algorithm, the convergence speed of the ML-MACO algorithm is 4 times faster than that of the ML-ACO algorithm, the computational load is 1/3 that of the ML-ACO algorithm, the resolution success rate is higher than the ML-PSO algorithm and the ML-ACO algorithm, and the estimation error is less than the ML-PSO algorithm and the ML-ACO algorithm. The ML-MACO algorithm maintains the excellent estimation performance of the ML algorithm with lower computational effort, better convergence performance, and higher estimation accuracy.

引言
1 ML估计方法
2 MACO算法原理
3 基于MACO的ML-DOA方法
4 算法仿真分析
5 水池实验及结果分析
结语

图1 SNR=10 dB时，ML-ACO算法与ML-MACO算法对双目标估计的优化迭代结果Fig. 1 (Color online) The optimization iteration results of ML-ACO algorithm and ML-MACO algorithm for double objective estimation with SNR=10 dB.

图2 四种算法在不同信噪比条件下的分辨概率Fig. 2 Resolution probability of four algorithms under different SNR conditions.

图3 四种算法在不同信噪比条件下的对（a）目标1和（b）目标2进行估计的均方误差比较Fig. 3 MSE comparison of four algorithms for estimation of (a) target 1 and (b) target 2 under different SNR conditions.

表1 ML-ACO算法和ML-MACO算法在不同信噪比条件下的平均迭代次数Table 1 Average number of iterations of ML-ACO algorithm and ML-MACO algorithm under different SNR conditions

表2 声源定标实验结果Table 2 Experimental results of sound source calibration （°）

表3 不同归一化夹角情况下采用ML-ACO算法和ML-MACO算法对双目标方位的估计结果Table 3 Estimation results of ML-ACO algorithm and ML-MACO algorithm for bearings of two targets under different normalized included angles.

[1]SIRIANUNPIBOON S, HOWARD S D, ELTON S D. Time-decentralized DOA estimation for electronic surveil‐lance [C]//The 51st Asilomar Conference on Signals, Sys‐tems, and Computers. Piscataway, USA: IEEE, 2017:1137-1141.
[2]CHEN Shiyu, LI Yanjie, CHEN Haoyao. A monocular vi‐sion localization algorithm based on maximum likelihood estimation [C]// International Conference on Real-time Computing and Robotics (RCAR). Piscataway, USA:IEEE, 2017: 561-566.
[3]LI M, LU Y. Genetic algorithm based maximum likeli‐hood DOA estimation [C]// International Radar Con‐ference. Edinburgh, UK: IET, 2002, 5(3): 502-506.
[4]CHEN Haihua, LI Shibao, LIU Jianhang, et al. A novel modification of PSO algorithm for SML estimation of DOA [J]. Sensors, 2016, 16(12): 2188.
[5]SHARMA A, MATHUR S. Comparative analysis of ML-PSO DOA estimation with conventional techniques in varied multipath channel environment [J]. Wireless Personal Communications, 2018, 100(3): 803-817.
[6]单泽彪，石要武，刘小松，等.应用人工蜂群算法的动态波达方向跟踪[J].光学精密工程，2015，23 （3）：838-845. SHAN Zebiao, SHI Yaowu, LIU Xiaosong, et al. DOA tracking of moving targets by artificial bee colony algorithm [J]. Optics and Precision Engineering, 2015, 23 (3): 838-845. (in Chinese)
[7]SHEIKH Y A, ULLAH R, YE Z. Range and direction of arrival estimation of near-field sources in sensor arrays using differential evolution algorithm [J]. International Journal of Computer Applications, 2016, 139(4): 16-20.
[8]POUR H M, ATLASBAF Z, MIRZAEE A, et al. A hybrid approach involving artificial neural network and ant colony optimization for direction of arrival estimation [C]//Canadian Conference on Electrical and Computer Engi‐neering. Piscataway, USA: IEEE, 2008: 1059-1064.
[9]王鹏，贺雪芳，张明星，等.基于人工鱼群算法的声矢量传感器阵列的最大似然DOA估计[J].太原理工大学学报，2020，51（6）：845-851. WANG Peng, HE Xuefang, ZHANG Mingxing, et al. Maximum likelihood DOA estimation of acoustic vector sensor array based on artificial fish swarm algorithm [J]. Journal of Taiyuan University of Technology, 2020, 51(6):845-851. (in Chinese)
[10]FAN Xinnan, PANG Linbin, SHI Pengfei, et al. Applica‐tion of bee evolutionary genetic algorithm to maximum likelihood direction-of-arrival estimation [J]. Mathemati‐cal Problems in Engineering, 2019, 2019: 6035870.
[11]王辉辉，陈玉凤.一种基于连续蚁群算法的快速最大似然DOA估计[J].火控雷达技术，2014，43（2）：1-4，67. WANG Huihui, CHEN Yufeng. A fast maximum likeli‐hood DOA estimation based on continuous ant colony algorithm [J]. Fire Control Radar Technology, 2014, 43(2):1-4, 67. (in Chinese)
[12]WANG Peng, KONG Yujun, HE Xuefang, et al. An improved squirrel search algorithm for maximum likeli‐hood DOA estimation and application for MEMS vector hydrophone array [J]. IEEE Access, 2019, 7: 118343-118358.
[13]SUDHOLT D. The benefits of population diversity in evo‐lutionary algorithms: a survey of rigorous runtime analyses [M]// DOERR B, NEUMANN F. Theory of Evolutionary Computation: Recent Developments in Discrete Optimiza‐tion. Cham: Springer International Publrshing, 2020:359-404.
[14]ZHOU Yongquan, WANG Rui, LUO Qifang. Elite opposition-based flower pollination algorithm [J]. Neuro‐computing, 2016, 188: 294-310.
[15]WU Guohua. Across neighborhood search for numerical optimization [J]. Information Sciences, 2016, 329:597-618.

备注

引言

1 ML估计方法

2 MACO算法原理

3 基于MACO的ML-DOA方法

4 算法仿真分析

5 水池实验及结果分析

结语

期刊信息

备注

引言