Real-Time Adaptive Control Scheme for Superior Plasma Confinement
Department of Energy
Agency Tracking Number:
Solicitation Topic Code:
Small Business Information
Intelligent Optical Systems, Inc.
2520 West 237th Street, Torrance, CA, 90505
Socially and Economically Disadvantaged:
Abstract61063 The U.S. Department of Energy (DOE) seeks intelligent automated tuning software to improve control and functionality of plasma fusion reactors. Currently installed tokamak control systems typically fail to make use of the full potential of available equipment. These systems often require significant operator intervention to tune control action between discharges. This project will develop a neural network-based approximator to produce a robust adaptive control scheme. An adaptive control algorithm will be designed to perform real-time plasma shape and boundary control. In Phase I, measurement data on various plasma equilibrium modes was acquired and analyzed. A Matlaba-based toolbox was developed, consisting of linear and neural network approximators that were capable of learning and predicting with high accuracy the behavior of plasma parameters. The development of a control algorithm capable of using the model of the plasma obtained by the neural network approximator was begun. In Phase II, the development of the neural-network-based control algorithm will be completed. The algorithm will be integrated into currently used control software for real-time control of a tokamak. The successful performance of the algorithm will be verified by comparing the predicted outputs to a well-established computer model of tokamak operation and to actual measurements from the data acquisition equipment. Commercial Applications and Other Benefits as described by the awardee: The tools developed in this project will demonstrate the feasibility of using neural network architectures for control of a plasma fusion reactor and should play a large role in the success of fusion as a viable energy source. The development of a generic adaptive control package that uses neural networks to estimate complex input-output relationships in multivariable systems should contribute to improvements in the performance of various other nonlinear systems, such as aircraft engines, power generators, chemical processes, and turbines
* information listed above is at the time of submission.