复杂环境下水声网络目标协同跟踪的DPMM方法

On the national maritime power strategy and the urgent demand of securing marine defense, this proposal aims cooperatively tracking targets over the underwater acoustic network to solve the model mismatching problem in the complicated ocean environment, where the non-countervailable errors are raised from multi-paths interference, quantization error and transmission error. Considering the multi-modal statistical distributions of sensors’ observations due to multi-paths interference between sensor nodes and targets, we shall construct a Dirichlet process mixture model (DPMM) which is enable with adaptation and learning ability, we shall propose off-line and on-line algorithms to estimate the parameters of this model, therefore, we shall propose a novel collaborative target tracking method based on DPMM under complicated underwater environment. Furthermore, we will build the measurement models of local sensors and fusion center in the complicated ocean environment, respectively. We shall propose a performance evaluation criterion based on Posterior Cramer-Rao Lower Bound (PCRLB) for underwater target tracking. It will indicate the mechanism that how these non-countervailable errors influence on the tracking performance. Theoretical simulations and water pool trails will be conducted for cooperatively tracking targets over the underwater acoustic network to show the effectiveness of the proposed method and possibly to improve it. We expect the research results have important theoretical significance and great value for the development of target cooperative monitoring system under the demand of marine space security in China.

“海洋强国”战略布局下，密切结合海洋空间安全的迫切需求，针对复杂环境不可抗误差引起的模型性失配问题，研究适用于水声网络应用环境的目标协同跟踪方法。针对由传感器节点和目标间观测信道中多径干扰引起的具有多模态统计分布特性量测，构建具有数据自适应学习能力的Dirichlet过程混合模型(DPMM)，并提出模型中参数的离线和在线估计方法，形成复杂环境下水声网络目标协同跟踪的DPMM新方法。建立水声网络中传感器节点和融合中心的量测模型，构建基于后验克拉美罗下界(PCRLB)的目标协同跟踪性能评估体系，揭示水声网络中复杂环境下不可抗误差对目标协同跟踪性能的影响机理。开展试验研究，进一步完善所提出的理论方法。研究成果对我国海洋空间安全需求下的目标协同监测体系发展有着重要的理论意义和应用价值。

海洋空间安全的核心任务是对某一重点海域进行有效防御和作战监视，从而应对水下入侵目标的威胁，水声传感器网络（水声网络）是实现重点海域水下空间大范围、立体化远程预警监测的重要手段。受海洋传输介质不均匀性以及海底、海面反射等因素的影响，水声信号传输存在明显的多径干扰，导致传感器节点信号到达时间（TOA）量测具有多模态统计特性，且模态数量取决于多径传播环境。针对传统目标协同跟踪方法难以适配多径环境导致的跟踪精度差的问题，开展了复杂环境下水声网络目标协同跟踪的DPMM方法研究，充分利用了Dirichlet过程混合模型(DPMM)可描述无限维混合模型特点以及层次Dirichlet过程混合模型(HDPMM)可描述无限维相关混合模型特点，构建了多径环境下水声网络目标协同跟踪的DPMM和HDPMM模型，从模型角度解决了传统目标协同跟踪方法模型性失配问题。在此基础上，提出了基于吉布斯采样的目标跟踪批处理算法(Gibbs-UKF-RTS)，利用Gibbs采样中降维的思想有效解决了高维后验概率采样困难的问题，利用UKF-RTS解决了目标非线性量测导致后验概率计算复杂的问题，实现了目标状态批处理估计；提出了基于粒子滤波的水声多径环境下水声网络目标跟踪算法(PF-Gibbs)，实现了目标状态的序贯采样估计；提出了基于变分推断(VI) 的水声多径环境下目标跟踪方法，在K-L散度优化框架下，通过变分推断逼近后验概率密度，实现了目标状态的序贯逼近估计。与传统方法相比，所提方法跟踪性能提升了25%以上。同时，构建了基于后验克拉美罗下界的水声网络复杂环境下目标协同跟踪性能评估体系，揭示了水声网络中复杂环境下不可抗误差对目标协同跟踪性能的影响机理。本项目研究内容对我国建设海洋强国需求下的高精度水下目标跟踪有着重要的理论意义和应用价值。