For thermal technologies for heavy oil and oil sand reservoir extraction, such as cyclic steam stimulation and steam-assisted gravity drainage (SAGD), suboptimal steam conformance leads to recovery factors between 25-50%. Although preliminary research using Proportional-Integral-Derivative (PID) control in SAGD operations has proved beneficial towards steam conformance, PID control is responsive only to deviation from set-points and lacks constraint-handling capabilities. This results in suboptimal actuation signals that are sometimes unattainable. This paper summarizes research on a Model-Predictive-Controller (MPC) with proactive adjustments of steam injection rate. The steam injection rate was determined based on recursive parameter updates of a suitable time varying dynamic model describing the implicit relationship between the subcool temperature difference and the input heat rate, to achieve optimal steam conformance. Furthermore, the steam injection rate was constrained such that the pressure with which the steam impinged on the formation, called well bottom hole pressure (BHP), was below the formation fracture pressure of 4500 kPa at all times. The real time control study was made possible by establishing a bidirectional communication link between the Computer Modelling Group (CMG) STARS T M , and MATLAB/Simulink software. The threedimensional heterogeneous reservoir model, developed in STARS T M acted as a virtual plant and the MPC, developed in MATLAB/Simulink, acted as an onsite controller. Results show 35.7% improvement in oil recovery and a more efficient cumulative steam-to-oil ratio (cSOR) profile in comparison to the base case of steam injection at a constant BHP of 4000 kPa.
Cognitive architectures such as ACT-R and EPIC are being applied to human factors research problems with increasing frequency. However, it is unclear whether such systems can model continuous motor tasks that were once staples in the field but have since been largely displaced by more cognitively-oriented problems. Recent research on a challenging continuous motor control task has revealed interesting patterns in skill acquisition that appear compatible with the learning mechanisms present in ACT-R. However, what was not clear was whether ACT-R could model expert performance in a high-frequency motor control task. Unmodified, ACT-R could not. However, by making some small changes in ACT-R's motor system and capitalizing on ACT-R's ability to imagine visual objects, ACT-R was able to achieve expert-level performance in this task. Whether ACT-R will be able to mirror the skill acquisition data is still an open question.
There are many domains that still require use of complex manual control, despite the general shift in the field toward research on supervisory control. One of the problems in complex manual control is training; we currently lack models that can help guide training. The research reported here is part of an effort to fill that gap. In this study, we used the Neverball video game as a motor control task and used performance metrics from the game to measure learning. In addition, we collected motion data to determine what basic movements correlated with game performance. Subjects showed evidence of learning in almost all of the performance metrics, which will enable comparisons with the motion data. The ultimate goal is to use the motion data to identify basic movements that underlie successful performance to provide as feedback during training, and hopefully accelerate learning.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.