Noise-Robust Parameter Estimation in ARX Models Using Stochastic Configuration Networks

This research introduces a hybrid framework combining Stochastic Configuration Networks (SCNs) with ARX (Auto-Regressive with Exogenous Input) models to address challenges in parameter estimation, particularly in noisy and diverse data environments. Traditional ARX methods, such as least squares and maximum likelihood estimation, struggle with noise and lack adaptability to varying data constraints. SCNs, with their incremental learning and advanced regularization techniques, provide a robust solution by dynamically adjusting to data variability. The integration of SCNs with ARX models incorporates manifold and consistency regularization to improve noise robustness and adaptability to both local and global data patterns. This combination ensures accurate parameter estimation in scenarios with limited data, high noise, or large-scale datasets. Experimental results show that models like ANN and Block Level with SCN (BLSCN) outperform traditional methods, achieving low estimation errors and high computational efficiency. The framework’s application in industrial processes, predictive maintenance, and signal processing highlights its effectiveness in dynamic, noisy environments.