A filter design for T-S fuzzy systems based on moving horizon estimator with measurement noise

Hui Gao, Yixuan Wang, Jing HuPeerJ Comput. Sci 2023 · DOI

TL;DR

This paper introduces a novel filter design for Takagi-Sugeno (T-S) fuzzy systems that leverages a Moving Horizon Estimator (MHE) to handle unknown system dynamics and measurement noise. This approach enhances system robustness against non-Gaussian disturbances and guarantees input-to-state stability (ISS), making it a valuable tool for complex nonlinear systems.

What Problem Does This Paper Address?

Many real-world systems, particularly in control engineering, are nonlinear and exhibit unknown dynamics. Traditional filtering methods, like the Kalman filter, often struggle with these complexities, especially when dealing with non-Gaussian noise or when the system model is not perfectly known. Takagi-Sugeno (T-S) fuzzy systems offer a way to model these nonlinearities using a set of linear "if-then" rules, but designing effective filters for them, particularly in the presence of disturbances, remains a challenge. Existing methods often assume Gaussian noise or require precise system models, limiting their applicability. This paper aims to address this gap by proposing a robust filtering approach for T-S fuzzy systems that can handle unknown dynamics and non-Gaussian measurement noise, ensuring system stability.

Methods

The core of this paper is the design of a filter based on the Moving Horizon Estimator (MHE) for T-S fuzzy systems. Here's a breakdown of the key methods:

  • T-S Fuzzy System Modeling: The paper starts by representing the nonlinear discrete-time system using the Takagi-Sugeno fuzzy model. This involves defining fuzzy rules, premise variables, and system matrices that capture the system's behavior across different operating conditions.
  • Moving Horizon Estimator (MHE): Unlike traditional filters that process data point-by-point, MHE uses a sliding window of past measurements and inputs to estimate the current state. This "horizon" allows the filter to consider a history of data, leading to more robust estimations, especially in the presence of noise.
  • Optimization-Based Estimation: The MHE formulates the estimation problem as an optimization problem. It aims to find the state trajectory within the horizon that best fits the observed measurements by minimizing a cost function. This cost function typically penalizes deviations from the system model and the measurements.
  • Input-to-State Stability (ISS) Guarantee: A significant contribution is the theoretical analysis to prove that the proposed MHE-based filter guarantees input-to-state stability (ISS). This means that if the external disturbances are bounded, the system's states will also remain bounded, ensuring predictable behavior.
  • Handling Non-Gaussian Noise: The MHE's inherent ability to process data over a window makes it less sensitive to outliers and non-Gaussian noise compared to methods that rely on statistical assumptions like Gaussianity.

Datasets

Not discussed in the paper. The paper focuses on simulation-based validation using generated noise signals.

Key Insights and Ideas

  • MHE as a Robust Alternative: The paper highlights that MHE offers a powerful alternative to traditional filters like the Kalman filter, particularly for nonlinear systems with unknown dynamics and non-Gaussian noise. Its window-based approach inherently improves robustness.
  • ISS for Enhanced Predictability: Proving Input-to-State Stability (ISS) is crucial. It assures that even with external disturbances, the system's internal states will not diverge uncontrollably, which is vital for safety-critical applications.
  • T-S Fuzzy Systems and MHE Synergy: The combination of T-S fuzzy modeling and MHE is shown to be effective. The fuzzy model provides a structured way to represent nonlinearities, while MHE provides a robust estimation framework for these models.
  • Effectiveness Against Non-Gaussian Disturbances: The simulations demonstrate that the MHE filter can effectively smooth out and estimate states even when the measurement noise is not normally distributed, a common limitation of Kalman filters.

Key Results

The paper demonstrates the effectiveness of the proposed MHE filter through simulation examples:

  • Gaussian Noise: The MHE filter significantly smooths the system's output compared to the raw measurements when subjected to Gaussian noise. The estimated value curve fluctuates within a much smaller range than the true value curve.
  • Non-Gaussian Noise (Sine-form): The MHE filter effectively counteracts the influence of a bounded, non-Gaussian sine-form noise. The estimation performance is most pronounced during the initial period, showing a clear improvement in smoothness.
  • Non-Gaussian Noise (Uniformly Distributed): The simulations with uniformly distributed noise also show that the MHE filter makes the system output smoother, confirming its ability to handle various types of external disturbances.

Limitations and Future Work

  • Computational Complexity: While MHE offers robustness, it can be computationally more intensive than simpler filters due to its optimization-based nature. The paper doesn't explicitly detail the computational burden but implies it through the iterative nature of the algorithm.
  • Tuning of MHE Parameters: The performance of MHE often depends on tuning parameters like the window length (L) and the weighting matrices in the cost function. The paper uses a fixed window length (L=5) for simulations, but optimal tuning for different systems might require further investigation.
  • Future Work: The authors suggest that the proposed method is feasible for discrete-time filtering problems in T-S fuzzy systems with external disturbances, even non-Gaussian ones. Future work could explore real-world applications and further optimization of the MHE algorithm for computational efficiency.

Summary updated

BibTeX
@article{gao2023filter,
  author = {Hui Gao and Yixuan Wang and Jing Hu},
  title = {{A filter design for T-S fuzzy systems based on moving horizon estimator with measurement noise}},
  journal = {PeerJ Comput. Sci.},
  year = {2023},
  doi = {10.7717/peerj-cs.1208},
}
Figure