Predicting Petroleum Reservoir Properties from Downhole Sensor Data using an Ensemble Model of Neural Networks

Abstract

The acquisition of huge sensor data has led to the advent of the smart field phenomenon in the petroleum industry. A lot of data is acquired during drilling and production processes through logging tools equipped with sub-surface/down-hole sensors. Reservoir modeling has advanced from the use of empirical equations through statistical regression tools to the present embrace of Artificial Intelligence (AI) and its hybrid techniques. Due to the high dimensionality and heterogeneity of the sensor data, the capability of conventional AI techniques has become limited as they could not handle more than one hypothesis at a time. Ensemble learning method has the capability to combine several hypotheses to evolve a single ensemble solution to a problem. Despite its popular use, especially in petroleum engineering, Artificial Neural Networks (ANN) has posed a number of challenges. One of such is the difficulty in determining the most suitable learning algorithm for optimal model performance. To save the cost, effort and time involved in the use of trial-and-error and evolutionary methods, this paper presents an ensemble model of ANN that combines the diverse performances of seven "weak" learning algorithms to evolve an ensemble solution in the prediction of porosity and permeability of petroleum reservoirs. When compared to the individual ANN, ANN-bagging and RandomForest, the proposed model performed best. This further confirms the great opportunities for ensemble modeling in petroleum reservoir characterization and other petroleum engineering problems.