Temporal Convolutional Network-Based Model for Precision Time-Series Forecasting in Marine Fisheries

Recent advancements in sensor technology have made it increasingly feasible to collect long-term, continuous data. However, challenges related to data completeness, timeliness, and quality remain prevalent, undermining the reliability of predictive models. This study proposes a time series prediction model tailored for marine fisheries, addressing these limitations by leveraging Convolutional Neural Networks (CNNs). Specifically, the model incorporates a Temporal Convolutional Network (TCN), which excels at identifying intrinsic patterns and hierarchical structures within large-scale marine time-series data, thereby enhancing model interpretability. The TCN framework demonstrates robust performance in capturing temporal dependencies and uncovering latent patterns, providing a comprehensive approach to managing the complex dynamics of marine data. This research applies the TCN deep learning model to examine the intricate relationships and information flow among various water quality parameters, tackling issues such as sequence distribution shifts and overfitting. Additionally, an interactive learning strategy is introduced to optimize hidden layer configurations and activation functions, further enhancing the predictive accuracy of the model. The results indicate that the proposed approach significantly improves both the timeliness and accuracy of predictions, offering valuable support for intelligent management in marine fisheries.