Distributed Data-driven Unknown-input Observers

Yuzhou Wei, Giorgia Disarò, Wenjie Liu, Jian Sun, Maria Elena Valcher, Gang WangAutomatica 2024 · arXiv · OpenReview

TL;DR

This paper introduces a novel data-driven approach to designing distributed observers for estimating the state of systems with unknown inputs, a common problem in wireless sensor networks. By leveraging offline data, the proposed method allows for state estimation without requiring an explicit model of the system, making it more practical for real-world applications.

What Problem Does This Paper Address?

In many real-world systems, especially those involving wireless sensor networks, the exact mathematical model is often unknown or difficult to obtain. This is particularly problematic when "unknown inputs" are present. These unknown inputs can arise from various sources like sensor aging, calibration errors, or external disturbances. They significantly degrade the accuracy of state estimation, which is crucial for monitoring and control. Traditional methods for dealing with unknown inputs, known as Unknown-Input Observers (UIOs), typically rely on having a precise model of the system. However, in complex, interconnected systems like sensor networks, obtaining such models is a major challenge. This paper tackles the problem of designing distributed UIOs for continuous-time linear time-invariant (LTI) systems when the system model is unknown, using only collected data.

Methods

The authors propose a Distributed Data-driven Unknown-Input Observer (D-DUIO). The core idea is to move away from model-based design towards a data-driven approach.

Here's a breakdown of their methodology:

  • Model-Based Foundation: They start by recalling existing model-based conditions for designing distributed UIOs. They also propose a slightly stronger, but more data-adaptable, condition.
  • Data-Driven Design: The key innovation is using offline collected data (input, output, and state trajectories) to design the observer. They establish conditions under which this data is representative of the system's behavior.
  • Existence and Construction: The paper provides a framework to:
  • Verify, using only the collected data, if the conditions for the existence of a D-DUIO are met.
  • Explicitly derive the matrices for the D-DUIO from this data, without needing to identify the system matrices (A, B, E, C).
  • Distributed Consensus: The D-DUIO scheme utilizes a consensus strategy, allowing neighboring sensor nodes to exchange information and collectively estimate the global system state, even if no single node has complete information.

Datasets

The paper does not introduce new datasets. Instead, it assumes that offline input/output/state data can be collected from the target system. This data is then used for the design and validation of the proposed D-DUIO. The numerical example in Section 5 uses a simulated two-mass-spring system to generate these datasets.

Key Insights and Ideas

  • Data as a Replacement for Models: The most significant insight is demonstrating that for designing distributed UIOs, explicit system models can be bypassed by leveraging sufficiently rich offline data. This opens up possibilities for applying UIOs in systems where modeling is impractical.
  • Verifying Observer Existence from Data: The ability to check the conditions for a D-DUIO's existence solely from collected data is a crucial practical advancement. It means engineers can assess the feasibility of the approach without needing to know the underlying system dynamics.
  • Decoupling Unknown Inputs and System Identification: The method explicitly avoids system identification. This is important because unknown inputs (like disturbances) can make accurate system identification very difficult or impossible. The D-DUIO directly addresses the estimation problem without needing to resolve the unknown inputs first.
  • Consensus for Distributed Estimation: The use of a consensus mechanism allows for distributed state estimation, where each node contributes to the overall estimation process by communicating with its neighbors. This is vital for large-scale sensor networks.

Key Results

The paper presents a novel D-DUIO scheme for continuous-time LTI systems with unknown inputs.

  • Theoretical Guarantees: They provide conditions for the existence of such a D-DUIO and a method to construct its matrices using offline data.
  • Asymptotic Convergence: Under mild assumptions, the proposed D-DUIO guarantees that the state estimates from all nodes will converge to the true system state asymptotically.
  • Numerical Validation: A simulation example with a five-node wireless sensor network demonstrates the effectiveness of the D-DUIO. The results show that the D-DUIO achieves consensus among nodes and accurately estimates the system state, outperforming a data-driven approach that attempts system identification (ID-DUIO) and performing competitively with a model-based DUIO. Specifically, the D-DUIO showed improvements in MSE and MAE compared to the ID-DUIO.

Limitations and Future Work

  • System Type: The current work focuses on continuous-time linear time-invariant (LTI) systems.
  • Data Requirements: The method relies on having access to offline input/output/state data. The quality and richness of this data are critical for the success of the approach, as formalized by Assumption 5 (rank of pre-collected data).
  • Future Directions: The authors suggest extending the framework to more complex scenarios, including:
  • Nonlinear systems.
  • Switching network topologies.
  • Exploring integral representations of data to avoid explicit derivative computation.

Summary updated

Abstract

Unknown inputs related to, e.g., sensor aging, modeling errors, or device bias, represent a major concern in wireless sensor networks, as they degrade the state estimation performance. To improve the performance, unknown-input observers (UIOs) have been proposed. Most of the results available to design UIOs are based on explicit system models, which can be difficult or impossible to obtain in real-world applications. Data-driven techniques, on the other hand, have become a viable alternative for the design and analysis of unknown systems using only data. In this context, a novel data-driven distributed unknown-input observer (D-DUIO) for unknown continuous-time linear time-invariant (LTI) systems is developed, which requires solely some data collected offline, without any prior knowledge of the system matrices. In the paper, first, a model-based approach to the design of a DUIO is presented. A sufficient condition for the existence of such a DUIO is recalled, and a new one is proposed, that is prone to a data-driven adaption. Moving to a data-driven approach, it is shown that under suitable assumptions on the input/output/state data collected from the continuous-time system, it is possible to both claim the existence of a D-DUIO and to derive its matrices in terms of the matrices of pre-collected data. Finally, the efficacy of the D-DUIO is illustrated by means of numerical examples.
BibTeX
@article{wei2024distributed,
  author = {Yuzhou Wei and Giorgia Disarò and Wenjie Liu and Jian Sun and Maria Elena Valcher and Gang Wang},
  title = {{Distributed Data-driven Unknown-input Observers}},
  journal = {Automatica},
  year = {2024},
  url = {https://openreview.net/forum?id=iJZuzxCXnf},
}
Figure