A Deep Reinforcement Learning-Based LSTM-Enhanced A3C Model for Predicting Bacterial Gill Disease in Aquaculture Water

Aquaculture, a rapidly growing industry that plays a crucial role in global food security and contributes significantly to the economies of many countries, faces substantial challenges in managing aquatic diseases. Among such diseases is Bacterial Gill Disease (BGD), which may cause severe economic losses if not detected early and promptly treated. This paper introduces a novel deep reinforcement learning approach, the LSTM-Enhanced Asynchronous Advantage Actor-Critic (LE-A3C) model, which was proposed to predict BGD by analyzing multivariate water quality indicators. The approach integrates LSTM's sequential learning capabilities with the advantage-based optimization and the asynchronous parallel learning approach of the A3C algorithm. This could remarkably capture complex temporal dependencies as well as water quality patterns, which allows the proposed LE-A3C to identify early warning signs for outbreaking potential BGD outbreaks. The asynchronous learning in A3C would enable multiple water quality scenarios to be explored and optimized concurrently; this actually accelerates learning and increases its robustness in real-time predictive tasks. This boosts the speed and accuracy of detecting diseases and, therefore is a highly efficient tool for aquaculture systems. The proposed LE-A3C model was implemented on the “Ponds data” dataset, and experimental results clearly demonstrate its superior performance compared to existing models, achieving a predictive accuracy of 97.50%. In comparison with the existing models M-DQN achieved an accuracy of 88.52%, the M-GRU model reached 78.92%, and the S-LSTM model recorded 74.31 % accuracy.