Comparative Analysis of Deep Neural Networks for Oil-in-Water Concentration Prediction Using PSO-Based Hyperparameter Optimization

Accurate prediction of Oil-in-Water (OiW) concentration in de-oiling hydrocyclone systems is crucial for maintaining environmental compliance and optimizing offshore water treatment operations. This study focuses on a comparative evaluation of Particle Swarm Optimization-based Long Short-Term Memory (LSTM) and Auto-Regressive Long Short-Term Memory (AR-LSTM) networks for modeling OiW dynamics by proposing a new optimization problem to fine-tune the network hyperparameters. The AR-LSTM model integrates an auto-regressive feedback mechanism to dynamically incorporate prior predictions, enhancing its adaptability to the time-dependent and irregular patterns characteristic of offshore processes. In contrast, the LSTM model provides a robust baseline for capturing temporal dependencies without feedback integration. The performance of both models is analyzed in terms of prediction accuracy and computational efficiency. The findings provide valuable insights into the trade-offs between the two architectures and underscore the role of hyperparameter optimization in advancing predictive capabilities for complex industrial systems.