Adequate frequent information on state variables of a process is sometimes needed for effective control and monitoring of the process. However, it is not often available in practice, which can be addressed using a state estimator. This work deals with distributed state estimation in large-scale processes. The decomposition of a process into observable subsystems is formulated as an optimization problem, which is solved using an efficient whale optimization algorithm. Four nonlinear state estimation methods (extended Kalman, unscented Kalman, spherical unscented Kalman, and cubature Kalman filtering) are then implemented and compared using distributed and centralized architectures on a process consisting of two reactors and a separator, and the Tennessee Eastman process. A parallelization strategy that improves the computational efficiency of the distributed architecture is proposed. Simulation results show that the parallel implementation of the distributed filtering methods is computationally more efficient than their centralized counterparts while yielding similarly accurate state estimates.

有时需要有关过程状态变量的足够频繁的信息以有效控制和监视过程。然而，它在实践中并不经常可用，这可以使用状态估计器来解决。这项工作涉及大规模过程中的分布式状态估计。将一个过程分解为可观察的子系统被表述为一个优化问题，该问题使用有效的鲸鱼优化算法来解决。然后使用分布式和集中式架构在由两个反应器和一个分离器组成的过程以及田纳西州伊士曼过程中实施和比较四种非线性状态估计方法（扩展卡尔曼、无迹卡尔曼、球形无迹卡尔曼和体积卡尔曼滤波）。提出了一种提高分布式架构计算效率的并行化策略。仿真结果表明，分布式过滤方法的并行实现比集中式过滤方法在计算上更有效，同时产生同样准确的状态估计。