Abstract:In this paper, we investigate the distributed estimation problem of continuous-time stochastic dynamic systems over sensor networks when both the system order and parameters are unknown. We propose a local information criterion (LIC) based on the penalty term. By minimizing LIC at the diffusion time instant and utilizing the continuous-time diffusion least squares algorithm, we obtain a distributed estimation algorithm to simultaneously estimate the unknown order and the parameters of the system. By dealing with the effect of the system noises and the coupling relationship between estimation of system orders and parameters, we establish the almost sure convergence results of the proposed distributed estimation algorithm. Furthermore, we give a simulation example to verify the effectiveness of the distributed algorithm in estimating the system order and parameters.
What problem does this paper attempt to address?
The paper is primarily dedicated to addressing the problem of distributed estimation in continuous-time stochastic dynamic systems within sensor networks, particularly when the system order and parameters are unknown. Specifically, the authors propose a Local Information Criterion (LIC) based on an L0 penalty term, and by minimizing this criterion, they simultaneously estimate the unknown order and parameters of the system. To achieve this goal, the paper employs a continuous-time diffusion least squares algorithm and establishes the almost sure convergence of the proposed distributed estimation algorithm by handling the influence of system noise and the coupling relationship between system order and parameter estimation.
The key contributions of the paper include:
1. **Problem Background**: With the development of computer and communication technologies, sensor networks have been widely applied in engineering systems. Compared to centralized algorithms, distributed algorithms offer higher flexibility, robustness, and ease of deployment, effectively utilizing data within sensor networks to estimate unknown parameters.
2. **Research Method**: A two-step distributed algorithm is proposed. First, the system order is estimated by minimizing the local information criterion. Then, based on the order estimation, the unknown parameters are estimated using the diffusion least squares algorithm.
3. **Theoretical Analysis**: Under cooperative incentive conditions, the almost sure convergence of the distributed order estimation and parameter estimation is proven.
4. **Application Validation**: The effectiveness of the proposed distributed algorithm is validated through simulation examples.
Overall, this paper proposes an effective solution for the problem of distributed order and parameter estimation in continuous-time stochastic dynamic systems within sensor networks, and its effectiveness is validated through theoretical analysis and simulations.