Any scheme for automatic analysis of well tests must have (1) a simulator that can model pressure transients accurately for a variety of well and reservoir configurations, (2) an estimator that provides unique estimates of reservoir parameters while producing a history match of the test data, and (3) a scheme for screening the data for the domain of validity of the chosen model. Most of the well test interpretation programs available today have elements (1) and programs available today have elements (1) and (2) and are capable of yielding unique estimates for parameters, provided one can devise an idealized parameters, provided one can devise an idealized model that fits the test data satisfactorily. This, however, happens very rarely. In practice, an ideal model can be found to fit only portions of the data. A routine application of a standard history matching program will lead to a forced fit of the ideal model program will lead to a forced fit of the ideal model on the data and the resulting parameter estimates would be misleading. The reason is that a "blackbox" regression algorithm is designed solely to estimate the unknown parameters in a model given that the model is valid and not to check for its validity or allow for its inadequacies. For instance, it lacks the logic to isolate that portion of the pressure build-up or drawdown data where a particular model is applicable. Without this, computerized analysis is not competitive with some of the more conventional techniques wherein the engineer has ample opportunity to verify the model using, say, type curve plots.
Realization of these has influenced our philosophy in designing WELLTEST, Chevron's program for well test interpretation. In this paper we relate our experience in its development and our approach to simulation, estimation and model discrimination.
Computer-aided or automatic methods for the analysis of well tests are receiving attention lately. The reasons are many:
They do not require extensive plotting and manual calculations required in the conventional methods such as Horner, Miller-Dyes-Hutchinson and type curve techniques.
They do not require restrictive test conditions such as constant or stabilized rates, longbuild-ups.
Tests with varying production rates, multiple drawdown and build-up tests, injection tests, PI - tests, all can be analyzed by the computer using appropriate superposition methods, thus replacing the numerous special purpose methods of specific applicability used conventionally.
The entire pressure production history in a test, rather than just build-up or drawdown, can be used in a unified manner by the automatic history-matching programs. This allows one to ensure that the reservoir model chosen and the parameters calculated are consistent with all the test data.
Highly complex reservoir geometries can, in principle, be simulated using today's reservoir simulators. This can be used in conjunction with estimation techniques in the search for improved reservoir definition.
The increasing availability of low-cost and high storage computers implies the possibility of having such programs fully portable and even distributed in remote locations. There is a definite advantage in having a powerful simulation and analysis tool at the exploratory well-site; this may aid the engineer in designing and simulating well tests, in monitoring the progress and in getting a first-hand evaluation of the data perhaps before leaving the well-site.
Some of the above have already been investigated, others, particularly the use in real-time environment, are yet to be tried. In any case the potential is there. However, we believe that the mere availability of a low-cost versatile computer program would encourage the engineer to explore a variety of possible alternatives in his search for reservoir possible alternatives in his search for reservoir definition. This is the main reason behind the development of the WELLTEST program at Chevron. Recently Shell announced their HITCH program for automatic history matching of well tests.
