Production Forcasting of Oil and Gas Wells in the Niger Delta Using Artificial Neural Network (ANN)

Accurate production forecasting in the Niger Delta complex reservoirs remains challenging due to significant heterogeneity, operational disruptions, and non-stationary decline behavior. This study develops an Artificial Neural Network (ANN) framework to improve short-term production forecasts by integrating operational data with engineered temporal features. Using daily production records from 10 – 15 mature wells, a period from 2010 to 2023, the model incorporates key variables—including oil, gas, and water rates, downhole pressures, choke sizes, and derived metrics like rolling averages and time since last workover to capture dynamic reservoir behaviors. A feedforward ANN architecture with three hidden layers of 256, 128, and 64 neurons and robust regularization (dropout, batch normalization) was optimized through Bayesian hyper-parameter tuning. The ANN significantly outperformed conventional methods, achieving test-set R-squared (R²) values of 0.92 for oil, 0.91 for gas and 0.93 for water, with a 24 – 31% reduction in error metrics (RMSE, MAPE) compared to random forest benchmarks. Notably, the model predicted 78% of major water-cut increases (>10%) at least seven days in advance. Feature importance analysis revealed wellhead pressure of 28% contribution and choke size of 23% as dominant predictors, aligning with physical principles. The framework's operational value was demonstrated through field cases, including a 22% production decline prediction 14 days prior to an electric submersible pump (ESP) failure, enabling proactive mitigation. These results highlight the ANN's capability to address Niger Delta forecasting challenges by leveraging high-frequency data and nonlinear pattern recognition and a practical, data-driven alternative to traditional decline curve analysis.