Stochastic modeling of earth’s hydrocarbon resources using artificial neural network

Abstract

Modeling of earth’s hydrocarbon resource and reserve is a complex phenomenon. Resource/reserve can be estimated using both deterministic and stochastic methods. However, according to the guidelines prescribed by Project resource management system (PRMS), stochastic method is a better approach. In this paper, a stochastic approach has been described which uses the power of two-layered Artificial neural network (ANN) technique. One layer is the input layer and the other one is a hidden layer. In resource estimation, six such input layers were considered, namely, area of hydrocarbon pool, pay thickness of hydrocarbon reservoir, saturation of hydrocarbon reservoir, Formation Volume Factor and Recovery Factor. The output is the model describing the stochastic range of hydrocarbon resource/reserve. Training of the network is performed with 100 random data value of each input reservoir parameter. 90% of the data have been used for training and remaining 10% for validating the network. The deterministic calculation acts as a target for stochastic inversion of data. The model performs best when quality data are fed during training. Mean square error was calculated which is the average of squared difference between normalized outputs and targets. Any value of error over 0.6667 signifies high error. Probability was assigned to the output layer ranges. Two pay zones were considered to demonstrate the efficacy of the system. Prospective recoverable resource with minimum (1P), most likely (2P) and maximum (3P) were determined and presented in this paper.