Comparison of machine learning models for performance evaluation of wind-induced vibration piezoelectric energy harvester with fin-shaped attachments

Three machine learning models are applied to estimate output voltage of the wind-induced vibration piezo-electric energy harvester (WIVPEH) in this paper. The dataset for machine learning model is from the wind tunnel experiment. The wind speed ranged from 1.02 m/s to 6.80 m/s. The fin-shaped attachments (FSAs) are fixed on the cylinder of the WIVPEH. The installation angle and coverage of FSAs are in the range of 0 <= theta <= 160 degrees and 20 degrees <= alpha <= 50 degrees. The range of FSA-height is H = 0.2D -0.4D. The results show that the amplitude ratio of the FSA-cylinder first increases and then decreases with the increase of FSA-height. The maximum amplitude reaches 2.3D at U = 6.8 m/s and H = 0.33D. The FSAs enhance energy extracting performance of the WIVPEH. The maximum output power of the WIVPEH with FSAs reaches 1.87 mW at H = 0.33D and U = 6.8 m/s, which is 49.21 times that of the harvester with smooth cylinder. The GBRT model with optimal parameters has the best performance for predicting output voltage of the WIVPEH with FSAs on the test dataset.