Estimating Reservoir Porosity: Probabilistic Neural Network

Evaluating Reservoir Porosity: Probabilistic Neural Network Estimation of Reservoir Porosity Using Probabilistic Neural Network Watchwords: Porosity Seismic Attributes Probabilistic Neural Network (PNN) Features: Porosity is assessed from seismicattributes utilizing Probabilistic Neural Networks. Impedance is determined by utilizing Probabilistic Neural Networks reversal. Multi-relapse investigation is utilized to choose input seismic traits. Unique Porosity is the most central property of hydrocarbon store. In any case, the porosity information that originate from well log are just accessible at well focuses. Subsequently, it is important to utilize different strategies to assess store porosity. Introduction is a basic and broadly utilized strategy for porosity estimation. In any case, the exactness of addition strategy isn't agreeable particularly in where the quantities of wells are little. Seismic information contain rich lithology data. There are natural relationships between's repository propertyand seismic information. Along these lines, it ispossible to evaluate store porosity by utilizing seismic information andattributes. Probabilistic Neural Network is a neoteric neuralnetwork modelbased on factual theory.It is a useful asset to separate mathematic connection between two informational collections. For this case, it has been utilized to separate the mathematic connection among porosity and seismic properties. In this i nvestigation, initially, a seismic impedance volume is determined by seismic reversal. Also, a few fitting seismic qualities are separated by utilizing multi-relapse investigation. At that point, a Probabilistic Neural Network model is prepared to acquire mathematic connection among porosity and seismic characteristics. At long last, this prepared Probabilistic Neural Network model is applied to compute a porosity information volume. This procedure could be utilized to discover beneficial regions at the beginning time of investigation. Furthermore, it is likewise useful for the foundation of supply model at the phase of store improvement. 1. Presentation As of late, clear advances have been made in the investigation and use of shrewd frameworks. Wise framework is an amazing asset to extricate quantitative definition between two informational indexes and has started to be applied to the oil business (Asoodeh and Bagheripour, 2014; Tahmasebi and Hezarkhani, 2012; Karimpouli et al., 2010; Chithra Chakra et al., 2013). There are inborn relationships between's repository properties and seismic traits (Iturrarã ¡n-Viveros and Parra, 2014; Yao and Journel, 2000). Accordingly, it ispossible to gauge store porosities by utilizing seismic information and qualities. Past examinations have demonstrated that it is doable to assess repository porosity by utilizing measurable techniques and wise frameworks (Na’imi et al., 2014; Iturrarã ¡n-Viveros, 2012; Leite and Vidal, 2011). Probabilistic NeuralNetwork (PNN) is a neoteric neural system model dependent on factual hypothesis. It is basically a sort of equal calculation dependent on the base Bayesian hazard measure (Miguez, 2010). It is not normal for customary multilayer forward system that requires a mistake back spread calculation, however a totally forward figuring process. The preparation time is shorter and the exactness is higher than conventional multilayer forward system. It is particularly reasonable for nonlinear multi characteristics investigation. For this case, PNN has great execution on concealed information. In this investigation, the propounded system is applied to assess the porosity of sandstone store prosperously. 2. Probabilistic Neural Network PNN is a variation of Radial Basis Function arranges and inexact Bayesian measurable strategies, the mix of new information vectors with the current information stockpiling to completely order the info information; a procedure that like human conduct (Parzen, 1962). Probabilistic Neural Network is an elective sort Neural Network (Specht, 1990). It depends on Parzen’s Probabilistic Density Function estimator. PNN is a four-layer feed-forward system, comprising of an information layer, an example layer, a summation layer and a yield layer (Muniz et al., 2010). Probabilistic NeuralNetwork is actuallya scientific introduction strategy, however it has a structure of neural system. It has preferable insertion work over multilayer feed forwardneural arrange. PNN’s necessity of preparing information test is as same as Multilayer Feed Forward Neural Network. It incorporates a progression of preparing test sets, and each example compares to the seismic example in the investigation window of each well. Assume that there is an informational index of n tests, each example comprises of m seismic traits and one supply parameter. Probabilistic Neural Network accept that each yield log worth could be communicated as a direct mix of info logging information esteem (Hampson et al., 2001). The new example after the property blend is communicated as: (1) The new anticipated logging esteems can be communicated as: (2) where㠯⠼å ¡ (3) The obscure amount D(x, xi) is the â€Å"distance† between input point and each preparation test point. This separation is estimated by seismic characteristics in multidimensional space and it is communicated by the obscure amount ÏÆ'j. Eq. (1)and Eq. (2) speak to the utilization of Probabilistic Neural Network. The preparation procedure incorporates deciding the ideal smoothing parameter set. The objective of the assurance on these parameters is to make the approval blunder minimization. Characterizing the kth target point approval result as follows: (4) At the point when the example focuses are not in the preparation information, it is the kth target test forecast esteem. In this way, if the example esteems are known, we can compute the forecast blunder of test focuses. Rehash this procedure for each preparation test set, we can characterize the all out expectation blunder of preparing information as: à £Ã¢â€š ¬Ã¢â€š ¬Ã£ £Ã¢â€š ¬Ã¢â€š ¬ à £Ã¢â€š ¬Ã¢â€š ¬(5) The expectation blunder relies upon the decision of parameter ÏÆ'j. This obscure amount understands the minimization through nonlinear conjugate inclination calculation. Approval blunder, the normal mistake of all barred wells, is the proportion of a potential expectation mistake during the time spent seismic traits change. The prepared Probabilistic Neural Network has the attributes of approval least mistake. The PNN doesn't require an iterative learning process, which can oversee extents of preparing information quicker than other Artificial Neural Network models (Muniz et al., 2010). The element is a consequence of the Bayesian technique’s conduct (Mantzaris et al., 2011). 3. Procedure The informational collections utilized in this examination have a place with 8 wells (comprising of W1 to W8) and post-stack 3D seismic information in Songliao Basin, Northeast China. The objective layer is the principal individual from the Cretaceous Nenjiang Formation that is one of the fundamental repositories here. In this examination, the fundamental substance incorporate seismic impedance reversal, properties extraction, preparing and utilization of PNN model. The stream diagram is appeared in Fig. 1. Fig. 1. The stream outline of this investigation 3.1 Seismic impedance reversal This area is to ascertain a certified 3D seismic impedance information volume for porosity estimation. The qualities are assembled from both seismic and reversal shape. The period of information 3D seismic information is near zero at the objective layer. The information have great quality in the whole time run without observable various impedance. T6 and T5 are the top and base of stores, individually. T6-1 is a middle of the road skyline somewhere in the range of T6 and T5 (Fig. 2 (b)). This information volume covers a region of around 120 km2. The structure type of store around there is an incline. There are two blames in the up plunge course of slant (Fig. 2 (a)). (a) (b) Fig. 2. (a) T6 skyline show. (b) A self-assertive line from seismic information, line of this area is appeared in (a). Seismic datacontain bottomless data of lithology andreservoirs property. Through seismic reversal, interface sort of seismic datacan beconverted intolithology kind of loggingdata, which could be directlycompared withwell logging (Pendrel, 2006). Seismic inversionbased on logging information exploits huge region sidelong conveyance ofseismic information joined with utilizing the geologicaltheory. It is a viable strategy to examine the circulation anddetailsof repositories. PNN reversal is a neoteric seismic wave impedance reversal technique. There is mapping connection between manufactured impedance from well log information and seismic follows close to well. In PNN reversal strategy, this mapping connection will be found and a scientific model will be developed via preparing. The solid strides of PNN reversal are as follow (Metzner, 2013): (1). Develop an underlying store land model. The control purposes of model are characterized by a progression of various profundity, speed and thickness information. (2). Neural Network model foundation and preparing. At this progression, a PNN model is developed and prepared. The preparation and approval mistake of prepared PNN ought to be limited. The prepared PNN model incorporates the numerical connection between engineered impedance by well log information and seismic follows close to well. (3). Figuring of impedance by applying the PNN model to seismic information volume. PNN reversal strategy exploits all the recurrence parts of well log information, and has great enemy of obstruction capacity. PNN reversal won't lessen goals in reversal procedure, and there is no blunder aggregation. Conclusive outcomes of reversal are shown in Figs. 3, 4, 5 and Table 1. Fig. 3. Cross plot of real impedance and anticipated impedance Fig. 4. Cross Validation Result of Inversion. Correlation=0.832, Average Error=546.55[(m/s)*(g/cc)] Fig. 5. Self-assertive line from inversed impedance information volume. Base guide is appeared in the figure lowerleft. Table 1 Numerical investigation of reversal at well areas 3.2 Seismic traits determination by utilizing multi-relapse analy