An innovative neural forecast of cumulative oil production from a petroleum reservoir employing higher-order neural networks (HONNs)

• We model higher-order neural network (HONN) to forecast real field oil production.
• A pre-process step performed to reduce noise and select optimal data for HONN model.
• The higher-order synaptic operations (HOSO) were employed to train the HONN models.
• HONN model tested on two case studies by applying oil, gas, and water production data of field.
• HONN with HOSO has superior capability for forecasting with reduced computation time.

Precise and consistent production forecasting is indeed an important step for the management and planning of petroleum reservoirs. A new neural approach to forecast cumulative oil production using higher-order neural network (HONN) has been applied in this study. HONN overcomes the limitation of the conventional neural networks by representing linear and nonlinear correlations of neural input variables. Thus, HONN possesses a great potential in forecasting petroleum reservoir productions without sufficient training data. Simulation studies were carried out on a sandstone reservoir located in Cambay basin in Gujarat, India, to prove the efficacy of HONNs in forecasting cumulative oil production of the field with insufficient field data available. A pre-processing procedure was employed in order to reduce measurement noise in the production data from the oil field by using a low pass filter and optimal input variable selection using cross-correlation function (CCF). The results of these simulation studies indicate that the HONN models have good forecasting capability with high accuracy to predict cumulative oil production.