Optimizing wind energy harvester with machine learning

• Published in: Applied mathematics and mechanics Vol. 46; no. 8; pp. 1417 - 1432
• Main Authors: Weng, Shun, Wu, Liying, Li, Zuoqiang, Zhang, Lanbin, Dai, Huliang
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.08.2025; Springer Nature B.V
• Edition: English ed.
• Subjects: Applications of Mathematics; Cantilever beams; Classical Mechanics; Energy harvesting; Evolutionary computation; Fluid- and Aerodynamics; Machine learning; Mathematical Modeling and Industrial Mathematics; Mathematics; Mathematics and Statistics; Mechanical efficiency; Neural networks; Optimization; Parameter identification; Partial Differential Equations; Piezoelectricity; Radial basis function; Wind power; Wind speed; Wind tunnels; piezoelectric effect; 68T05; differential evolution (DE); wind energy harvester (WEH); vortex-induced vibration (VIV); machine learning (ML); radial basis function neural network (RBFNN); O326
• ISSN: 0253-4827; 1573-2754
• DOI: 10.1007/s10483-025-3279-6

Optimizing wind energy harvesting performance remains a significant challenge. Machine learning (ML) offers a promising approach for addressing this challenge. This study proposes an ML-based approach using the radial basis function neural network (RBFNN) and differential evolution (DE) to predict and optimize the structural parameters (the diameter of the spherical bluff body D , the total spring stiffness k , and the length of the piezoelectric cantilever beam L ) of the wind energy harvester (WEH). The RBFNN model is trained with theoretical data and validated with wind tunnel experimental results, achieving the coefficient-of-determination scores R 2 of 97.8% and 90.3% for predicting the average output power P avg and aero-electro-mechanical efficiency η aem , respectively. The DE algorithm is used to identify the optimal parameter combinations for wind speeds U ranging from 2.5 m/s to 6.5 m/s. The maximum P avg is achieved when D = 57.5 mm, k = 28.8 N/m, L = 112.1 mm, and U = 4.6 m/s, while the maximum η aem is achieved when D = 52.7 mm, k = 29.2 N/m, L = 89.2 mm, and U = 4.7 m/s. Compared with that of the non-optimized structure, the WEH performance is improved by 28.6% in P avg and 19.1% in η aem .