新一代信息系统呈现出网络规模巨型化、节点部署密集化、分析决策智能化的新特征。采用分布式参数估计实现对海量数据的高效处理，已成为网络中智能信号处理的关键。本课题针对大规模复杂网络应用，研究基于置信传播的分布式参数估计方法。首先，针对线性模型中置信传播算法的收敛性问题，采用阻尼方法改善其收敛特性，揭示阻尼系数与收敛速度之间的本质关系，从而优化求解最佳阻尼系数。其次，针对置信传播算法在非线性模型中不存在闭合表达式的问题，通过对模型的级数展开和多维高斯分布的分块边缘化处理，推导参数化置信传播算法，并设计双环迭代更新机制，提升算法收敛性和估计精度。最后，针对全局目标参数的一致性估计问题，通过引入软约束条件构建因子图模型，将全局参数边缘分布共识问题转化为多参数的一致性估计问题，实现全局目标参数与局部参数的联合估计。本课题的研究工作丰富了分布式参数估计理论，对大规模复杂网络中的信号处理应用具有重要价值。

The new generation information system has the characters of large-scale network size, high-density node deployment, and intelligent decision-making. Developing efficient distributed parameter estimation methods is very critical for intelligent signal processing in large-scale networks with massive data. In this project, distributed parameter estimation methods based on belief propagation are investigated for large-scale networks. First, in linear systems, damping is used to improve the convergence properties of belief propagation. The relationship between the convergence properties and damping factors is disclosed, and damping factors are optimized to achieve the fastest convergence rate. Then, to develop parametric belief propagation in nonlinear systems, the second-order Taylor expansion of nonlinear model and block marginalization of high-dimensional Gaussian distribution are used. To further improve the convergence and accuracy of the parametric belief propagation, a double-loop iterative algorithm is designed. Finally, to realize the consensus estimation of a single global parameter, belief propagation on a factor graph with soft constraints is proposed. Moreover, in the joint estimations of global parameters and local parameters, the consensus estimation of the global parameter's marginal distribution is derived and transformed into that of multiple parameters. In summary, this project will contribute to provide new fundamental theories on distributed parameter estimation and practical implementations for signal processing in large-scale networks.