Research On Data Preprocessing And Model Diagnose Technology

Well test is usually employed to get reservoir parameters by analyze geologicalcondition and production history. Reservoir parameters can be estimated by measuringpressure and production data. Right now all of the well test interpretation plates are based onfixed production. If we use actual pressure curve iftting theory chart is bound to cause someerror, therefore we need to preprocess the data to get a more viable result, which includegetting rid of odd points and convert pressure data under transient flow rate into data withsteady condition. Deconvolution can be used to identify stratigraphic model and reservoirparameters without using any stratigraphic model. Meanwhile, its calculating result is muchreliable than conventional explanation method.This article employs deconvolution for data preprocessing, which can get a more reliableformation pressure and reservoir properties than traditional pressure drop or pressurerecovery test. Not only can this method improve the economic benifit of a well test butenhance the level of development and management of the wells. Conventional deconvolutionemploys the Fouirer deconvolution and Laplace deconvolution in well test analysis. Butbecause deconvolution demand a very high accuracy of test data, it behave a strongsensitivity to measurement error, even if the error is so faint to notice, it will still cause agreat change in the result, the least square method can be used to eliminate the instability, sothis dissertation put a lot of energy in study the nonlinear least squares method deconvolution.The paper finally studied the test model diagnosis，which could draw a model diagnosticplot using the deconvolution result. Analyzing the character of this plot，wells can becategorized to its fitting model and the corresponding reliable parameter.