Enhancing Flow Energy Harvesting Efficiency Through Hybrid Deep Learning Models

In the context of increasingly scarce environmental resources, ocean energy harvesting has emerged as a critical research area. Among various ocean energy sources, flow energy has gained prominence due to its extensive reach and numerous advantages. However, due to the complex and irregular nature of the flow energy domain, existing Maximum Power Point Tracking (MPPT) algorithms are often not suitable for effectively addressing the challenges in this field. Therefore, this paper proposes a novel hybrid deep learning model to address these challenges and improve the efficiency of flow energy harvesting. Specifically, we employ Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, an attention mechanism, and a custom-designed ensemble loss function to enhance prediction accuracy. This combination leverages the strengths of each component, allowing the model to effectively capture both spatial and temporal patterns in flow energy data, while also addressing the inherent variability and complexity of the energy conversion process. Comparative analyses with common techniques such as Random Forest (RF) and LSTM reveal that our model achieves lower error rates, with accuracy improvements of approximately 9% over RF and 7% over LSTM, based on data from our flow energy harvesting experiments. Additionally, we demonstrate the necessity of each part of our model through ablation experiments. Our experiments offer key insights for enhancing flow energy harvesting efficiency, serving as a reference for future advancements in the field.