Session: SYMP 7-3: Hydrokinetic Energy Harvesting

Paper Number: 139336

139336 - Enhancing Vortex Energy Harvesting Efficiency: Evaluating Predictive Algorithms 

Effectively capturing fluid mechanical energy stands as a pivotal challenge in marine energy research. The unpredictable and irregular of marine mechanical energy poses significant hurdles in developing efficient energy harvesting systems. In the quest to enhance the efficiency of harnessing ocean flow energy, this paper explores the application of Maximum Power Point Tracking (MPPT) algorithms. Data prediction aims at the output open circuit voltage of planar energy harvesters. The training data required for prediction is obtained from the water tank experiment. Specifically, our investigation delves into the prediction of voltage fluctuations in piezoelectric energy harvesting devices at specific time intervals, with the overarching goal of optimizing system parameters based on these predictions to maximize power output. Innovatively contributing to the field, this study proposes a novel approach by introducing a simple yet widely-used algorithm tailored for the unique environment of flow-induced vortex energy harvesting.

The primary focus of this research centers on a comparative analysis of two algorithms: multiple linear regression prediction and Gated Recurrent Unit (GRU) neural networks. Through a comprehensive evaluation using diverse regression metrics such as Mean Squared Error (MSE), R-squared (R²), and Huber loss, a notable trend is observed.

Contrary to the traditional viewpoint favoring complex neural networks for their presumed superiority, our experiments demonstrate the computational efficiency and accuracy of multiple linear regression prediction. The simplicity and inherent efficiency advantages of multiple linear regression become apparent in the practical environment of planar vortex energy prediction. Moreover, the time consumption of multiple linear regression is significantly lower than that of neural network algorithms, presenting a crucial advantage in real-time computations for embedded systems. Based on our results, linear regression algorithm significantly outperformed the GRU algorithm both in terms of prediction accuracy and time efficiency. With the improvement of approximately 60% in accuracy and a notable enhancement of around 90% in runtime efficiency. Our experiments demonstrate that under the specific conditions of this study, complex models may not necessarily yield optimal results.

Our experimental data stem from voltage measurements obtained from C-type and L-type piezoelectric energy harvesting devices under various conditions, including different flow speeds, beam-column sizes, and cylinder diameters. The diverse dataset enriches our comparative analysis.

This paper establishes a foundational exploration for future marine mechanical energy prediction - collection research. We offer a succinct overview of the challenges posed by irregular flow-induced vortex patterns and the imperative demand for predictive MPPT algorithm.

In conclusion, this study provides valuable insights into algorithm selection for flow-induced vortex energy prediction, advocating the adoption of simpler models such as multiple linear regression. These findings not only guide current applications but also lay the groundwork for further research into real-time parameter adjustments and the intricate dynamics of flow-induced vortex energy harvesting systems. Importantly, to the best of our knowledge, this study represents the pioneering use of deep learning methods in flow-induced vortex energy prediction.

Presenting Author: Yuanhang Xie School of Artificial Intelligence of Future Technology of Shanghai University

Presenting Author Biography: Yuanhang Xie received the bachelor's degrees from Shanghai Normal University. He is currently pursuing a M.S. degree in the School of Artificial Intelligence of Future Technology, which is a key laboratory of Ministry of Education, at Shanghai University. His research interests focus on energy harvesting and prediction algorithms. He has previously received the First Prize at the university level in the College Students' Innovation and Entrepreneurship Training Program.

Authors:

Ying Gong
Yuanhang Xie
Zhongjie Li
Wenhua Zhang