Abstract
The neural network technique that is a field of artificial intelligence (AI) has proved to be a good model classifier in all areas of engineering and especially, it has gained a considerable acceptance in well test interpretation model (WTIM) identification of petroleum engineering. Conventionally, identification of the WTIM has been approached by graphical analysis method that requires an experienced expert. Recently, neural network technique equipped with back propagation (BP) learning algorithm was presented and it differs from the AI technique such as symbolic approach that must be accompanied with the data preparation procedures such as smoothing, segmenting, and symbolic transformation. The BP approach is able to identify the model with incomplete or distorted data without data preparation procedures. However, this approach uses partial set of data to reduce computing time and memory, and may miss the points representing the characteristics of the shape. The resulting identified model would not be correct and this requires sequential neural nets to get the correct model.
In this paper, we developed BP neural network with Hough transform (HT) technique to over-come data selection problem and to use single neural network rather sequential nets. The Hough transform method was proved to be a powerful tool for the shape detection in image processing and computer vision technologies.
With the aid of a HT method applied in this work, we can extract one simple pattern from the full set of data of pressure derivative type curve (PDTC) containing noisy and extraneous points. By using extracted simple pattern from HT method, we can then minimize number of data points when BP neural network is performed for the identification process.
Along these lines, a number of exercises were conducted with the actual well test data in two steps. First, the newly developed AI model, namely, ANNIS (Artificial intelligence Neural Network Identification System) was utilized to identify WTIM. Secondly, we obtained reservoir characteristics with the well test model equipped with modified Levenberg-Marquart method. The results show that ANNIS was proved to be quite reliable model for the data having noisy, miss-ing, and extraneous points. They also demon-strate that reservoir parameters were successfu1ly estimated.
