基于改进支持向量机的非理想因素声呐基阵鲁棒波束形成方法研究

Analysis and research on robustness beamforming algorithms of sonar array with nonideal factors based on improved support vector machine (SVM) can provide important theoretical foundation for direction of arrival (DOA) estimation. In the existing methods, the performances of robustness beamforming algorithms are known to degrade severely in the complex underwater acoustic environment. The robustness of beamforming algorithms based on SVM is faced with the difficulties of weak robustness, heavy computation and model .parameters selection. In this project, the research is focused on robustness beamforming of sonar array with nonideal factors. The idea of optimization theory is introduced into the parameter’s selecting of SVM model and then the new improved robustness SVM model based on the optimization theory will be developed. Based on the simulation results and experimental results, the new algorithm performance can be analyzed. The research can not only enrich and improve the beamforming theory of sonar array, but also provide novel theoretical references for improving algorithm’s robustness under nonideal factors. This project could be considered as the deepening and expansion of the previous projects funded by Chinese Nature Science Foundation. The prospective achievements of the project can be expected to have wide applications in underwater acoustic countermeasure, torpedo sonar self-guide system and marine geological prospecting ability as well as other fields.

基于SVM理论的非理想因素声呐基阵鲁棒波束形成理论研究与算法性能研究可为目标方位的精确估计提供重要理论基础。现有的鲁棒波束形成算法在实际复杂水声环境中算法性能将急剧下降，而基于SVM的鲁棒波束形成算法则面临着鲁棒性弱、运算量大、模型参数选取困难等问题。本项目围绕非理想因素下的声呐基阵鲁棒波束形成算法展开研究，将最优化理论引入到SVM模型参数的选取优化中，研究基于最优化理论的改进SVM模型和求解算法，然后将改进的SVM模型与波束形成技术相结合，建立基于改进SVM的非理想因素鲁棒波束形成新方法，联合仿真研究和实验研究结果，系统分析算法的性能，进一步丰富声呐基阵波束形成理论体系，为提高非理想因素下的波束形成算法鲁棒性提供新的理论依据。本项目是前期国家自然科学基金项目研究的深化和拓展，预期研究成果在水声对抗、鱼雷声自导和海洋地质勘探等领域具有广泛应用前景。

本项目瞄准未来水声目标探测、鱼雷声自导和海洋地质勘探的重大应用需求，以非理想因素影响下的声呐基阵鲁棒自适应波束形成技术为切入点，围绕鲁棒SVM、非理想因素声呐基阵波束形成开展理论研究和实验研究。具体研究内容如下：（1）进一步完善SVM-LCMV和SVM-LSMI鲁棒波束形成模型。将传统LCMV和LSMI波束形成算法与SVM相结合，并在求解过程中引入PSO理论，有效减小运算量，通过实验验证两种模型的正确性及鲁棒性。（2）分析不同仿生算法在非理想因素下对基于SVM的波束形成模型鲁棒性的影响。分别将CSO、ICSO、PSO、FA、GWO、LGA算法引入到 SVM 模型参数的选取优化中，研究在无失配情况下、DOA失配和阵元位置扰动三种常见的非理想因素作用下，不同仿生算法对基于SVM波束形成模型造成的影响。（3）基于多核SVM的鲁棒波束形成算法。将前面研究的基于SVM的鲁棒波束形成模型中支持向量机的单一核函数更改为多个核函数的组合，得到基于多核SVM的鲁棒波束形成算法。重点分析将多种核函数分别进行两两组合的多核SVM波束形成算法。通过实验验证改进SVM的鲁棒波束形成算法不仅可以抑制干扰信号，同时具有控制旁瓣的能力。本项目的研究为实际复杂水声环境中声呐基阵的精确探测定位能力提供有效的技术途径。研究成果对于鲁棒波束形成在水声对抗、鱼雷声自导和海洋地质勘探等领域具有重要的意义，在军事和民用领域均有十分广阔的应用前景。