A Variable Step-Size Hybrid Adaptive Nonlinear Filter for Solar Radiation Prediction

• Published in: International Conference on Systems and Informatics pp. 1 - 5
• Main Authors: Xiao, Y., Ma, L., Khorasani, K.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.12.2024
• Subjects: Accuracy; Adaptive filters; Artificial neural networks; Finite impulse response filters; functional link artificial neural network (FLANN); Informatics; Kernel; Linearity; Nonlinear filters; nonlinearity; Solar radiation; solar radiation prediction; tracking capability; variable step size (VSS); Volterra filter; Weather forecasting
• ISSN: 2689-7148
• DOI: 10.1109/ICSAI65059.2024.10893851

In this paper, a new hybrid adaptive nonlinear filter (HANF) scheme with variable step sizes (VSS) is proposed for solar radiation prediction. Our methodology consists of a Volterra filter and a functional link artificial neural network (FLANN) filter. The Volterra filter with the first- and second-order kernels is included, that is capable of expressing both linearity and nonlinearity. The FLANN filter can handle the nonlinearity that may be expressed by higher-order exponential terms that the Volterra filter with a limited number of kernels is unable to deal with. The VSSs are introduced in the two filters to allow the HANF to enjoy desirable tracking capability such that the time-varying nonlinearity underlying the nonlinear phenomena can be detected and tracked. The proposed VSS-HANF is applied to a real hourly solar radiation time sequence to confirm its improved prediction performance as compared to its counterpart with fixed step sizes.