One of the key objectives doing well testing is to derive effective reservoir properties, such as permeability, to provide input for reservoir simulation. Traditional approach in well test analysis is first, to separate transient pressure draw down and build ups' due to constant flowing rate, then analyze them by forcing a pre-selected model to derive effective reservoir permeability through an inversion process. However, the results obtained this way are not really dynamic or at most is "pseudo-dynamic", because these are based on the fact that a continuous signal was broken into discontinuous signals. Some information (more than a changed permeability value) of the reservoir system response may already lose, while uncertainties were increased or multiplied in the process of data evaluation and analysis.
This paper presents a new approach, which will allow for continuous reservoir system response analysis and interpretation. Therefore the information derived from such analysis can reflect the real time, dynamic changes of the reservoir. This has been proved more powerful in analyzing transient pressure, particularly for that from permanent down hole gauges (PDG).
Numerical well testing procedures, assisted by neural network method, were used to analyze the data from simulated and field cases in a continuous, systematic fashion. First, the flowing rate history was recovered from the measured transient pressure. Transient pressure histories are then re-produced through simulations of both well test forward model and neural network black box model. A match between measured pressure response (PDG data) and re-produced transient pressure histories will then be made, to achieve the ultimate goal of well testing - real time reservoir monitoring, model calibration and reservoir management. This final matching process was named "guided (by neural network model) history matching".
Introduction
Along with the advent of reliable permanent down-hole gauge available to the oil industry, the importance of continuous reservoir monitoring in mature field and its value delivered by up-to-date information has been recognized as the essential element in modern reservoir management (Athichanagom et al., 1999; Rossi et al., 2000; Chiriti et al., 2001; Ballinas and Owen, 2002; Haddad et al., 2004; Olsen and Nordtvedt, 2005; Weaver et al., 2005; Chorneyko 2006; Frota and Destro, 2006; Olsen and Nordtvedt, 2006). However, the key to ensure the information delivered correctly is a robust analysis method, which can, not only handling the large, noisy data set, but also distil the "message" from the dynamic data for right decision making. Quite a few studies have been released in recent years using wavelet algorithm (Kikani and He, 1998; Soliman et al., 2001; Ouyang and Kikani, 2002; Guan et al., 2004; Olsen and Nordtvedt, 2005; Ribeiro et al., 2006; Zheng and Li, 2007), however, these are more about pre-analysis data processing, the analysis method used after data processing is more or less based on constant terminal rate pressure Draw-Down (DD) solution being regarded as traditional approach.
