A Distributed State Estimation Approach to Condition Monitoring of Nonlinear Electric Power Systems

Abstract

This paper analyzes distributed state estimation methods for condition monitoring of electric power transmission and distribution systems. When a fault occurs in such large‐scale systems, it is usually difficult to detect it and to determine its exact position. Moreover, due to the cost of installation and maintenance of measurement devices and due to the excessive size of the electric power grid, the complete monitoring of the associated infrastructure is impractical. Therefore, to monitor the condition of the power grid, some form of estimation is required. As suitable approaches for distributed state estimation this paper proposes the extended information filter (EIF) and the unscented information filter (UIF). The Extended Information Filter is actually an implementation of distributed extended Kalman filtering while the unscented information filter is an implementation of distributed unscented Kalman filtering. With the use of the aforementioned filtering algorithms on processing units located at different parts of the power grid, one can produce local estimates of the system's state vector which in turn can be fused into an aggregate state estimation. The produced global state estimate enables continuous monitoring of the condition of the electric power system and early fault diagnosis if used by a suitable fault detection and isolation algorithm.

Distributed State Estimation of Smart Grids with Packet Losses

“…In order to find the optimal gain i k , taking the partial derivative of i k|k in (20) with respect to i k and applying (21) and (22) yields:…”

“…wherêi k|k is the ith estimation error covariance matrix by (20), 4 is the 4 by 4 identity matrix and is the auxiliary variable for minimizing the global estimator error covariance. As the error covariance is minimized, so that the estimated states match the true states.…”

Distributed State Estimation Using RSC Coded Smart Grid Communications

“…In [26], a distributed hierarchical structure is provided in which local state estimation is computed independently by the local KF at each sensor node. In [27], the distributed extended information filter and unscented information filter are considered for condition monitoring of power transmission and distribution systems. Here, the local estimated states and covariance matrices are fed to an aggregator filter.…”

“…The centralized SE means a huge amount of state information is collected and processed at the central state estimation unit. This not only causes communication and computational burdens but also creates a possibility for central point failure [27], [46]. For this reason, the distributed estimation approaches are an striking alternate as they may need less communication bandwidth and allow parallel processing [44].…”

Full- and reduced-order distributed Bayesian estimation analytical performance bounds

“…As stated earlier, previous derivations of the PCRLB are limited to the centralized and hierarchical estimation architectures [15], and only recently has a suboptimal PCRLB expression [16] been derived for the decentralized architectures. In this paper, optimal dPCRLB algorithms are derived for two different estimation methodologies: 1) full-order distributed approaches [23][24][25][26][27][28][29][30][31][32][33], where the entire state vector is estimated locally at each observation node without resorting to a fusion center; and 2) reduced-order distributed approaches [34][35][36][37][38][39] for large-scale dynamical systems [40][41][42][43][44][45][46], where the overall system is partitioned [43,45] into several lower-dimensional subsystems with only a subset of state variables estimated at each observation node. The reduced-order subsystems are coupled, and connectivity between neighboring subsystems [34] is maintained.…”