Hybrid Neural Network Approaches to Produced Water Quality Forecasting and Management in Nigerian Oil Fields

Abstract

The management of produced water is a significant challenge for environmental agencies and operating companies in Nigerian oil environments, where escalating water-cut ratios persist. This research proposes and compares different artificial neural network (ANN) models with the integration strategies of Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Wavelet Transform (WT) for forecasting the quality variables TDS, Oil & Grease, Chloride, and Heavy Metals. Data for six prominent oil fields in the Niger Delta for 2015-2023 have been employed for analyses with all three models: ANN-PSO, ANN-GA, and ANN-WT. These models have confirmed that ANN-PSO models predict with higher precision compared to all other models, with regression values above 0.94 for forecasting TDS. The values for Root Mean Square Error (RMSE) have been found to vary between 23% and 31% lower than ANNs. The model ANNE-WT has outperformed with a regression value of 0.92 for forecasting chloride levels with a mean absolute percentage error (MAPE) value of 4.7% for Seasonal components.
These models offer oil field operators an adequate predictive system that can proactively optimize water treatment systems and facilitate regulatory compliance. These results help address the issue of water cut increase in a more sustainable manner within the mature oil field settings.