SPE 28394 A Methodological Approach For Reservoir Heterogeneity Characterization Using Artificial Neural Networks

For the past few years Artificial Neural Networks (ANN) have made a strong comeback to the scientific community. They are used in a variety of tasks where adaptive computing can enhance process performance. There has been a handful of papers suggesting the use of artificial neural networks in the petroleum industry . These 1-3 papers can be classified into two major categories. First category includes papers that recommend the use of ANN in classification of lithologies from well logs. Second category includes papers that employ ANN to pick the proper reservoir model for well testing purposes. This paper introduces a new implementation of the neuro-computing technology in petroleum engineering. It is shown in this study that artificial neural networks possess numerous capabilities, and can be much more useful to petroleum engineers than previously thought. An implementation of artificial neural networks in characterization of reservoir heterogeneity is presented in this paper. A methodology is introduced through which different rock properties in highly heterogeneous reservoirs can be predicted with good accuracy using information deduced from geophysical well logs. Examples of such networks are presented using field data for verification. The underlying reasons (theories) that make achievement of such complex tasks possible are discussed. The notion that artificial intelligence and neural networks in particular have immense potentials in solving complex engineering and scientific problems are addressed. The innovation now lies on the creativity of the researchers to recognize and define petroleum engineering problems that can be addressed by artificial intelligence technology. Introduction Heterogeneity in a hydrocarbon reservoir is referred to nonuniform, non-linear spacial distribution of rock properties. Characterization of porosity, permeability, oil, gas and water saturation of hydrocarbon bearing rocks is the focus of this technical paper. Calculating formation porosity and water saturation from geophysical well logs has been practiced since well or modeled non-linear relationship between porosity and water saturation with density and resistivity log responses. The term modeled non-linearity has been used here to emphasize that in these calculations it is assumed that a known function (linear or non-linear) is sufficient for modeling the relationship between these rock parameters and the aforementioned well log responses. A thorough investigation of such relationships and a comparison of their results with those produced by an artificial neural network will be addressed in a separate paper. Here, simply some preliminary results and a short discussion on the performance of artificial neural networks in predicting rock porosity as well as water, gas and oil saturations in heterogeneous reservoirs, where all three phases exist, is presented. Characterizing rock permeability and its spacial distribution in a heterogeneous reservoir is a problem with no direct and known solution. To date, there are only two generally reliable ways of acquiring knowledge on rock permeability. These are laboratory measurements and well test interpretation. Laboratory measurement of the cores attained from the field or core archives, provides precise (assuming adequate equipment) permeability values that are used in reservoir simulation studies as well as any other design and development studies on the field. The second method for permeability determination is pressure transient analysis which provides a volumetrically averaged permeability for the volume of the reservoir that has been investigated during the test. It should be noted that during the well testing procedure the length of the test is an important issue. Test should be designed so that it is long enough to achieve reliable and usable data. On the other hand, the longer the test time, the larger the volume represented by the calculated permeability. In this paper the authors introduce a new method for permeability determination. This technique is quite inexpensive. It does not require production interruption and provides permeability values that are comparable to those obtained 2 SPE 28394 MOHAGHEGH, S., AREFI, R., AMERI, S., HEFNER, M.H. by laboratory measurements of cores. In a feasibility study on this processing of a body of information, all of which are available at method of permeability prediction/estimation, Mohaghegh, et. al. the same time. The parallel distributed information processing 1 showed that such efforts are indeed fruitful. In that study, Mohacharacteristics of neural networks accommodate this necessity. ghegh, et. al. demonstrated that with a limited number of data, a The science of pattern recognition is concerned with three major carefully designed and developed artificial neural network can issues; 1) The appropriate description of objects, physical or provide acceptable results. conceptual, in terms of representation space, 2) The specification Methodology tion space into interpretation space . Another important characterPetroleum engineers have shown a high degree of open-mindedness in utilizing new technologies from different disciplines to solve old and new petroleum engineering problems. Use of CT-Scan, MRI, Microwave, and even expert systems are good examples. Artificial Intelligence in general and neural networks specifically are no exceptions. The key in using artificial neural nets in petroleum engineering, or in any other discipline for that matter, is to observe, recognize, and define problems in a way that will be addressable by neural nets. It is obvious that neural network is not a panacea for petroleum industry, but it very well may help solve problems that conventional computing has not been successful in solving. Artificial Intelligence is generally divided into two basic categories, rule based (expert) systems and adaptive (neural) systems. Neural network, a biologically inspired computing scheme, is an analog, adaptive, distributive, and massively parallel system that has been used in many disciplines and has proven to have potential in solving problems that require pattern recognition. The main interest in neural network has its roots in the recognition that the brain processes information in a different manner than conventional digital computers. Computers are extremely fast and precise at executing sequences of instructions that have been formulated for them (algorithm). A human information processing system is composed of neurons switching at speeds about a million times slower than logical computer gates . Yet, humans are more 5 efficient than computers at computationally complex tasks such as speech understanding and other pattern recognition problems. Artificial neural systems, or neural networks, are physical cellular systems which can acquire, store, and utilize experiential knowledge. The knowledge is in the form of stable states or mapping, embedded in networks that can be recalled in response to the Big Injun formation, in the Granny Creek field in West Virginia presentation of cues. was chosen for this study (Figure 1.) Located approximately 25 This statement means that, unlike a digital, sequential computer is structurally situated on the northwest flank of a syncline which with a central processor that can address an array of memory strikes N 15-20 degrees east to S 15-10 degrees west. Upper locations, neural networks store knowledge in the overall state of Pocono Big Injun sandstone is the oil producing formation in the the network after it has reached some equilibrium condition (stable Granny Creek field. Big Injun sandstone has been sub-divided into state.) In other words, knowledge in a neural network is not stored several sections. Using grain-size distribution and bulk density in a particular location. One can not look at memory address 1354 variations it has been sub-divided into A, B and C members. In an to retrieve the value of permeability. Knowledge is stored both in engineering and geological study of this field, Big Injun formation the way processing elements are connected, and in the importance has also been subjected to other sub-divisions. Using depositional of each input to the processing element (embedded mapping.) environment, the formation was sub-divided into 5 separate Knowledge is more a function of the network's architecture or sections and using lithofacie it was sub-divided into two sections structure than the contents of particular locations . (Figure 2.) In a preliminary study using neural networks, the 6 Pattern recognition has proven to be one of the neural nets' strong distribution as well as bulk density variation (the A, B, C subpoints. The essence of pattern recognition is the concurrent divisions) should be incorporated in this work. of an interpretation space, and 3) The mapping from representa-