Recently, streamline-based flow simulation models have offered significant potential in integrating dynamic data into highresolution reservoir models. A unique feature of the streamlinebased production data integration has been the concept of a traveltime match that is analogous to seismic tomography, allowing the use of efficient and proven techniques from geophysics. In this paper, we propose a generalized travel-time inversion method for production data integration that is particularly well-suited for large-scale field applications with gravity and changing conditions. Instead of matching the production data directly, we minimize a travel-time shift derived by maximizing a cross-correlation between the observed and computed production response at each well. There are several advantages of our proposed method. First, it is general and extremely computationally efficient. The traveltime sensitivities can be computed analytically with a single forward streamline simulation that can be much faster than a conventional reservoir simulator. Second, it is robust and the minimization is relatively insensitive to the choice of the initial model. Finally, it is field-proven because we utilize established techniques from geophysical inverse theory.We demonstrate the power and utility of our proposed method using synthetic and field examples. The synthetic examples include a large-scale 3D example with a quarter-million grid cells involving infill drilling and pattern conversions. The field example is from the Goldsmith San Andres Unit (GSAU) in West Texas and includes multiple patterns with 11 injectors and 31 producers. Starting with a reservoir model based on well-log and seismic data, we integrate water-cut history for 20 years in less than 2 hours on a PC.

[1] This paper presents an efficient trajectory-based approach to integrate transient pressure data into high-resolution reservoir and aquifer models. The method involves alternating traveltime and peak amplitude matching of pressure response using inverse modeling and is particularly well suited for high-resolution subsurface characterization using hydraulic tomography or pressure interference tests. Compared to traveltime inversion only, our proposed approach results in a significantly improved match of the pressure response at the wells and also better estimates of subsurface properties. This is accomplished with very little increase in computational cost. Utilizing the concept of a ''diffusive'' time of flight derived from an asymptotic solution of the diffusivity equation, we develop analytical approaches to estimate the sensitivities for traveltime and peak amplitude of pressure response to subsurface properties. The sensitivities are then used in an iterative least squares minimization to match the pressure data. We illustrate our approach using synthetic and field examples. In the field application at a fractured limestone formation the predominant fracture patterns emerging from the inversion are shown to be consistent with independent geophysical experiments and borehole data.

Recently streamline-based flow simulation models have offered significant potential in integrating dynamic data into high-resolution reservoir models. A unique feature of the streamline-based production data integration has been the concept of a ‘travel time’ match that is analogous to seismic tomography and has allowed the use of efficient and proven techniques from geophysics. In this paper we propose a ‘generalized travel time’ inversion method for production data integration that is particularly well-suited for large-scale field applications with changing conditions. Instead of matching the production data directly, we minimize a ‘travel time shift’ derived by maximizing a cross-correlation between the observed and computed production response at each well. There are several advantages with our proposed method. First, it is general and extremely computationally efficient. The travel time sensitivities can be computed analytically using a single streamline simulation that can be orders of magnitude faster than a conventional reservoir simulator. Second, it is very robust and the minimization is relatively insensitive to the choice of the initial model. Finally, it is field-proven because we utilize efficient techniques from geophysical inverse theory. We demonstrate the power and utility of our proposed method using synthetic and field examples. The synthetic examples include a large-scale 3D example with quarter-million grid cells involving infill drilling and pattern conversions. The field example is from the Goldsmith San Andres Unit (GSAU) in West Texas and includes multiple patterns with 11 injectors and 31 producers. Starting with a reservoir model based on well log and seismic data, we integrate water-cut history for 20 years in less than 2 hours in a PC. Introduction It is well known that geological models derived from static data only, such as geological, well log, core and seismic data, often fail to reproduce the production history. Reconciling geologic models to the dynamic response of the reservoir is critical to building reliable reservoir models. Past few years, have seen significant developments in the area of such dynamic data integration through the use of inverse modeling.1–13 Streamline models have shown great promise in this regard.9–13 The key advantages of streamline-based production data integration are its computational efficiency as a "forward" model and analytic computations of sensitivities of the production response with respect to reservoir parameters using a single flow simulation.9–11 Sensitivities describe the change in production response because of a small perturbation in reservoir properties such as porosity and permeability and are a vital part of the dynamic data integration process. Our previous works on streamline-based production data integration followed directly from seismic waveform inversion and utilized a two-step procedure9,10 (i) a travel time match that involves matching of the ‘first arrival’ or breakthrough times and (ii) an amplitude match involving matching of the actual production response. The two-step approach has been shown to substantially speed-up the computation and also prevents the solutions from being trapped by secondary peaks in the production response. However, a majority of the production data misfit reduction occurs during the travel time inversion and most of the large-scale features of heterogeneity are resolved at this stage.9,10

Streamline-based models have shown great potential in reconciling high resolution geologic models to production data. In this work we extend the streamline-based production data integration technique to naturally fractured reservoirs. We use a dualporosity streamline model for fracture flow simulation by treating the fracture and matrix as separate continua that are connected through a transfer function. Next, we analytically compute the sensitivities that define the relationship between the reservoir properties and the production response in fractured reservoirs. Finally, production data integration is carried out via the Generalized Travel Time inversion (GTT). We also apply the streamline-derived sensitivities in conjunction with a dual porosity finite difference simulator to combine the efficiency of the streamline approach with the versatility of the finite difference approach. This significantly broadens the applicability of the streamlinebased approach in terms of incorporating compressibility effects and complex physics.The number of reservoir parameters to be estimated is commonly orders of magnitude larger than the observation data, leading to non-uniqueness and uncertainty in reservoir parameter estimate. Such uncertainty is passed to reservoir response forecast which needs to be quantified in economic and operational risk analysis. In this work we sample parameter uncertainty using a new two-stage Markov Chain Monte Carlo (MCMC) that is very fast and overcomes much of its current limitations. The computational efficiency comes through a substantial increase in the acceptance rate during MCMC by using a fast linearized approximation to the flow simulation and the likelihood function, the critical link between the reservoir model and production data. The Gradual Deformation Method (GDM) provides a useful framework to preserve geologic structure. Current dynamic data integration methods using GDM are inefficient due to the use of numerical sensitivity calculations which limits the method to deforming two or three models at a time. In this work, we derived streamline-based analytical sensitivities for the GDM that can be obtained from a single simulation run for any number of basis models. The new Generalized Travel Time GDM (GTT-GDM) is highly efficient and achieved a performance close to regular GTT inversion while preserving the geologic structure.

Accurate prediction of future well performance is of great importance for petroleum reservoir management. This paper presents a practical neural network approach to predict existing and infill oil well performance using available filed data, such as well production history and well configuration information. It serves as a practical, cost-effective and robust tool for oilfield production and management. Well production, well spacing and the time-dependent information are used to train the neural network. The time-dependent information of wells are incorporated in a manner of time series for establishment of neural network. After the neural network is established, it is used to predict future performance of existing and infill wells. No reservoir data is currently used in the establishment of neural network, therefore it can predict well production performance in absence of reservoir data. Primary production of two data sets (each has 9 wells) in North Robertson Unit located in west Texas was tested using this approach. The results demonstrate that this approach is powerful in rapid projection of existing wells’ future performance, as well as the performance prediction of infill drilling wells. Introduction Neural network is a kind of artificial intelligence technology. It mimics characteristics of biological neurons. Like human information processing system, artificial neural system, or neural network, acquire, store, and utilize knowledge by learning. The knowledge is embedded in the networks that can be recalled in response to the presented information. There are several types of neural networks. The back-propagation neural network is one of them and has been most commonly used for engineering purposes. The structure of back-propagation neural network is sketched in Fig. 1. The neural network usually consists of at least three layers. They are input layer, hide layer and output layer. Each layer has a number of neurons or nodes. Each input neuron contains actual data introduced to the network externally. The hidden layer between input layer and output layer can be one or more than one according to different applications. Each neuron in output layer give a response for a given data set. The neurons between layers are interconnected by numbers called weights.These weights determine how a particular set of inputs be sent to neurons in hidden layer, and so on from hidden layer to neurons in output layer. These weights are determined by a process called training. Establishment of a neural network is to train the neural network by adjusting the weights between neurons so that the network can give desired output sets for given input data sets by using the trained weights. In the process of training, the weights are first randomized between the range of 0 to 1. The input data matrix is then multiplied by the weights connecting input layer and hidden layer to produce a new matrix. A transfer function is used to transfer this matrix to another matrix, which is the output of middle layer neurons. The obtained matrix is multiplied by the weights associated with the middle - output layer to generate a new matrix. Similarly, a transfer function is used to transfer the matrix from middle layer to output from the output layer neurons. The obtained output results are compared with the desired outputs to calculated the discrepancy. The weights are updated based on the discrepancy from output layer to input layer (back-propagation) until a specified convergence criterion is satisfied for all input data sets and desired output sets. In recent years, there are increasing applications in the oil and gas industry1–7. Neural network has been used in various petroleum engineering areas, such as geology, geophysics, drilling and completion, formation evaluation, production and stimulation, reservoir engineering and economic etc. Neural network ha s also been proved as a very successful approach on time series prediction problems. Good examples include prices of stock market, etc.8–10