ANN-Based Predictive Information System for Barium Sulfate Solubility in Oilfield Brines Under Variable Temperature, Pressure, and Ionic Strength

Abstract

Accurate solubility prediction of barium sulfate (BaSO₄) is essential to prevent scale formation in oilfield operations, yet conventional thermodynamic models are computationally intensive and unreliable when dealing with nonlinear interactions between pressure, temperature, and brine composition. This paper presents an Artificial Neural Network (ANN)–based predictive information system designed to estimate BaSO₄ solubility under wide operating conditions (25–250°C, 1–500 bar, and NaCl concentrations up to 4 M). The system integrates data preprocessing, normalization, ANN model development, and automated validation within a structured workflow, making it suitable for deployment as a real-time decision-support module in industrial monitoring systems.

The optimal ANN architecture—six layers, three neurons per layer, trained using the Levenberg–Marquardt backpropagation algorithm—achieved superior predictive performance with MSE = 6.82×10⁻⁵ and R² = 0.994 during training. Validation results confirmed generalization capability with MAE = 1.3, RMSE = 2.09, and MAPE = 4.3%, effectively outperforming conventional prediction models reported in the literature. Comparative benchmarking against empirical models demonstrated that the ANN system reduces prediction error by up to 40–60%, particularly at high-temperature/pressure operational windows where classical solubility equations degrade in accuracy.

The proposed solution demonstrates that machine learning can reliably model complex solubility behavior without explicit thermodynamic formulations, enabling scalable integration into oilfield digital dashboards, predictive maintenance systems, and automated scale-prevention workflows. This validates the applicability of ANN-based soft sensors as a cost-effective and accurate alternative approach within industrial information systems and data-driven decision support.