Masoud Nikravesh

Shortest-prediction-horizon non-linear model-predictive control

Abstract This article concerns non-linear control of single-input-single-output processes with input constraints and deadtimes. The problem of input-output linearization in continuous time is formulated as a model-predictive control problem, for processes with full-state measurements and for processes with incomplete state measurements and deadtimes. This model-predictive control formulation allows one (i) to establish the connections between model-predictive and input-output linearizing control methods; and (ii) to solve directly the problems of constraint handling and windup in input-output linearizing control. The derived model-predictive control laws have the shortest possible prediction horizon and explicit analytical form, and thus their implementation does not require on-line optimization. Necessary conditions for stability of the closed-loop system under the constrained dynamic control laws are given. The connections between (a) the developed control laws and (b) the model state feedback control and the modified internal model control are established. The application and performance of the derived controllers are demonstrated by numerical simulations of chemical and biochemical reactor examples.

Shortest-Prediction Horizon Nonlinear Model Predictive Control 1

Abstract This article presents a continuous-time formulation of model predictive control. This formulation allows (i) to establish the connections between model predictive control and input-output linearizing control methods and (ii) to address the problems of constraint handling and windup in input-output linearizing control methods. Model predictive control laws with the shortest possible prediction horizon are derived for constrained nonlinear processes with deadtime. They have explicit analytical form, and thus their implementation does not require on-line optimization. Furthermore, in the absence of constraints, they are input-output linearizing.

Evolution of fuzzy logic: from intelligent systems and computation to human mind

Inspired by human’s remarkable capability to perform a wide variety of physical and mental tasks without any measurements and computations and dissatisfied with classical logic as a tool for modeling human reasoning in an imprecise environment, Lotfi A. Zadeh developed the theory and foundation of fuzzy logic with his 1965 paper “Fuzzy sets” (Zadeh in Inf Control 8:378–53, 1965) and extended his work with his 2005 paper “Toward a generalized theory of uncertainty (GTU)—an outline” (Zadeh in Inf Control, 2005). Fuzzy logic has at least two main sources over the past century. The first of these sources was initiated by Peirce in the form what he called a logic of vagueness in 1900s, and the second source is Lotfi’s A. Zadeh work, fuzzy sets and fuzzy Logic in the 1960s and 1970s.

Dividing oil fields into regions with similar characteristic behavior using neural network and fuzzy logic approaches

Presents the next generation of intelligent oil field surveillance and prediction software based on neural networks and fuzzy logic. We treat the entire oil field as a coupled, highly nonlinear system of water injectors and oil/water/gas producers. The oil field is divided into regions with similar characteristic behavior using neural network and fuzzy logic. Wells in each region are then modeled with specialized neural networks trained to recognize their particular behavior. The model helps to improve waterflood management, avoid reservoir damage, and increase oil recovery per unit volume of injected water. Finally, the model visualizes the global trajectory of an entire field project and allow engineers to recognize patterns of incipient reservoir damage and poor performance.

Neural Networks for Field-Wise Waterflood Management in Low Permeability, Fractured Oil Reservoirs

An optimal water injection policy maximizes oil recovery per barrel of injected water while minimizing formation damage and maintaining reservoir pressure. O ptimal water injection into low permeability, fractured oil reservoirs is problematic because of highly nonlinear and complex reservoir dynamics. Likewise, current first principle models of fluid movement in fractured, low permeability rock systems are insufficient to design, operate, and predict the performance of large scale waterfloods. Historically, the conflict between prudent reservoir management and meeting field injection-production targets has resulted in reservoir and well damage, injectant recirculation and irreversibly lost oil production. Here we present the next generation of “intelligent” field surveillance and prediction software based on neural networks and implemented on a PC. We demonstrate a new approach to field-wise performance prediction and optimization of waterfloods that recognizes an oil field as a coupled, highly nonlinear system of injectors and producers. With lease-wide historical data from a waterflood in the Lost Hills Diatomite (Kern County, CA), we construct several neural networks which recognize that individual well behavior may depend on well history and the injectionproduction conditions of surrounding wells. Some of our neural networks accurately predict wellhead pressure as a function of injection rate, and vice versa, for all injectors. Other networks history-match oil and water production on the well-by-well basis, and predict future production on a quarterly or half-year basis. Finally, our neural networks recognize and suggest water injection policies that lead to the minimum injected water and the best oil recovery.

Theoretical Methodology for Prediction of Gas-Condensate Flow Behavior

Theoretical and experimental evidence has shown that the flow of condensate in the early stages of condensate formation up to {sigma} < {sigma}{sub c} is represented by film flow. After a transition state in which the gas saturation reaches critical condensate saturation, the condensate flow is subsequently reduced to bulk flow. The condensate is formed in the smaller pores, fills these pores and will continue into the larger pores. In the presence of interstitial water saturation, the condensate is formed in the water surface in the early stages of condensate formation. In addition, the gravity field has an important effect on S{sub cc}. Therefore, ignorance of the gravity effect in the experimental and theoretical calculations leads to an overestimation of the value for S{sub cc}. It is also shown that the shape of relative permeability curves are suddenly changed for a given critical IFT value ({sigma}{sub c}).

CT scan and neural network technology for construction of detailed distribution of residual oil saturation during waterflooding

We present an integrated approach to imaging the progress of air displacement by spontaneous imbibition of oil into sandstone. We combine Computerized Tomography (CT) scanning and neural network image processing. The main aspects of our approach are I) visualization of the distribution of oil and air saturation by CT, II) interpretation of CT scans using neural networks, and III) reconstruction of 3-D images of oil saturation from the CT scans with a neural network model. The neural networks developed here construct 3-D images of fluid distribution at any time and/or location within the core. One neural network model interpolates between the CT images for a given position at different time levels and extrapolates the interval of time during which the images were collected. Likewise, the network interpolates spatially between images at a given time. After interpolation and extrapolation, other network models have been developed to reconstruct the three-dimensional distribution of oil in the core. Excellent agreement between the actual images and the neural network predictions is found.

Pattern Trees: An Effective Machine Learning Approach

Fuzzy classification is one of the most important applications of fuzzy logic. Its goal is to find a set of fuzzy rules which describe classification problems. Most of the existing fuzzy rule induction methods (e.g., the fuzzy decision trees induction method) focus on searching rules consisting of t-norms (i.e., AND) only, but not t-conorms (OR) explicitly. This may lead to the omission of generating important rules which involve t-conorms explicitly. This paper proposes a type of tree termed pattern trees which make use of different aggregations including both t-norms and t-conorms. Like decision trees, pattern trees are an effective machine learning tool for classification applications. This paper discusses the difference between decision trees and pattern trees, and also shows that the subsethood based method (SBM) and the weighted subsethood based method (WSBM) are two specific cases of pattern trees, with each having a fixed pattern tree structure.

Dynamic neural network control of nonlinear-time varying systems: stability analysis, optimal network structure, and on-line adaptation

In this paper, dynamic neural network control (DNNC) is presented as a control strategy which uses a neural network to model the process and then applies the mathematical inversion of the process model to design the controller. DNNC falls into the large class of model predictive controllers, many of which are now widely used in industry. The objectives of this paper are: to study the stability properties of DNNC; and to study the on-line adaptation properties of DNNC. In this study four basic assumptions are made. The process is modeled with continuous time functions. If the process is nonstationary, it is also necessary to assure that changes will be continuous and smooth over small intervals of time, i.e. that process parameters change in a continuous manner. The time between model adaptation is small relative to time scale of parameter changes. Process parameters will not change at once, but only a small subset of them.