Dynamic optimization using long short-term memory and genetic algorithms for predicting marine data

Abstrak

This study aims to develop an accurate and efficient ocean data prediction model to tackle the challenges posed by climate change and complex oceanographic dynamics. The main goal is to use long short-term memory (LSTM) networks along with genetic algorithms (GA) to predict four key ocean factors at once: sea surface temperature (SST), sea surface height (SSH), sea surface salinity (SSS), and chlorophyll-a (Chl-a). An experimental quantitative approach is employed, utilizing satellite data from the Banda Sea region. This approach involves time series modeling using LSTM, which is optimized by GA for hyperparameters such as the number of neurons and batch size. The results show that the combined LSTM-GA model greatly improves prediction accuracy and successfully identifies seasonal trends and irregular changes in all variables, even when there is a lot of noise. Tests reveal that the optimal configuration varies for each variable, and the GA optimization process can expedite model convergence by as little as 10 epochs. These findings underscore the effectiveness of integrating evolutionary techniques in training deep learning (DL) models for ocean data. The implications of this research include potential applications in adaptive ocean monitoring systems, early warning initiatives, and data-driven planning in marine resource management.