Distributed State Estimation with Deep Neural Networks for Uncertain Nonlinear Systems under Event-Triggered Communication

Abstract

Distributed state estimation is examined for a sensor network tasked with reconstructing a system's state through the use of a distributed and event-triggered observer. Each agent in the sensor network employs a deep neural network (DNN) to approximate the uncertain nonlinear dynamics of the system, which is trained using a multiple timescale approach. Specifically, the outer weights of each DNN are updated online using a Lyapunov-based gradient descent update law, while the inner weights and biases are trained offline using a supervised learning method and collected input-output data. The observer utilizes event-triggered communication to promote the efficient use of network resources. A nonsmooth Lyapunov analysis shows the distributed event-triggered observer has a uniformly ultimately bounded state reconstruction error. A simulation study is provided to validate the result and demonstrate the performance improvements afforded by the DNNs.

Multi-Area Distributed State Estimation in Smart Grids Using Data-Driven Kalman Filters

“…Special cases of multi-area distributed state estimation, in which areas are defined based on each bus and its onehop neighborhood, which we term 1-hop neighborhood state estimation, have also been considered in the literature. Examples of 1-hop neighborhood distributed state estimation techniques in power systems include techniques based on neural networks [32], belief propagation [33], and KF [34,35]. For instance, the work in [34] uses diffusion-based KF for 1-hop neighborhood distributed state estimation.…”

A model-based technique for fault identification of sensors in autonomous systems using adaptive neural networks and extended Kalman filtering

“…19,20 In this study, a new method for detecting and managing faults in actuators and sensors in cylinder-shaped AUVs is developed. A full 6-DoF model of the remote environmental monitoring unit (Remus) AUV 21,22 was used to demonstrate the application of the fault detection strategy outlined in this paper. Using the proposed strategy, we implemented an adaptive neural network (ANN) with input from a nonlinear observer where the ANN weights are being updated using an extended Kalman filter (EKF).…”