A novel interval Type-2 fuzzy framework integrating MLP inference and savitzky-golay filtering for enhanced sun tracker performance

This study proposes a novel model, the Savitzky-Golay Filtered Interval Type-2 Fuzzy Multilayer Perceptron (SGF-IT2FMLP), to enhance the tracking performance of dual-axis sun tracker systems. The primary novelty lies in integrating Savitzky-Golay Filter (SGF), Interval Type-2 Fuzzy (IT2F) logic system, and Multilayer Perceptron (MLP) into a unified framework that eliminates the computational bottleneck of traditional IT2F systems. The primary innovation of this study lies in the data-driven inference process, which employs a Multi-Layer Perceptron (MLP) to replace the iterative procedure of the Karnik-Mendel (KM) algorithm, thereby reducing the computational burden associated with type-reduction. Specifically, the proposed framework employs the MLP as a data-driven inference mechanism, effectively replacing the computationally expensive and iterative Karnik-Mendel (KM) algorithm. This approach significantly simplifies the computational process by enabling direct output generation. Furthermore, the SGF is utilized to mitigate noise and enhance sensor data quality, while the IT2F system models data uncertainty through Upper (UMF) and Lower Membership Functions (LMF). These membership degrees subsequently serve as input features for the MLP inference model to predict azimuth and altitude angles. The model was rigorously evaluated across various combinations of SGF and MLP parameters, encompassing different filter orders, activation functions, and optimizers. The results demonstrate that the SGF-IT2FMLP framework, particularly the SGF-2 LBFGS-Sigmoid configuration, effectively reduces noise, yields accurate and stable predictions, and exhibits strong generalization capabilities. Moreover, Monte Carlo-based uncertainty analysis confirms the model’s high robustness, achieving 95% confidence intervals of ±5.4723° for the azimuth angle and ±0.0134° for the altitude angle. Experimental evaluation demonstrates the superiority of the proposed SGF-IT2FMLP framework over other benchmark models. Specifically, the proposed model significantly improves tracking accuracy by reducing the RMSE to 2.414° for the azimuth angle and 0.542° for the altitude angle, while maintaining a low MAPE of 0.9%.