Development of neural network controllers for power industry applications. Volume 1, Final report
- Stanford Univ., CA (United States). Information Systems Lab.
This report details an investigation in the development of the theory of the design of neural network controllers, and their application to power system problems. The current algorithm for the supervised training of neural networks, backpropagation, is reviewed. Using signal flow graph theory to develop the backpropagation-through-time algorithm, further insight of the training of neural network controllers in dynamic control systems is gained. The strengths of neural networks are brought out in this work, demonstrating that neural networks can solve highly nonlinear problems for which no closed-form, analytical solutions exist. Demonstrating the ability of a neural network to estimate complex processes, the backpropagation algorithm is used to train a neural network to predict void fraction. The results show that the accuracy of t@e neural network estimator is favorable. Demonstrating the ability of a neural network to control dynamic systems, the backpropagation-through-time algorithm is used to train neural networks to solve excitation control problems. Work is performed in developing a neural network controller for regulating load-frequency and terminal voltage for a simplified single synchronous generator model and for two generators linked by a tie-line. Results are favorable and show promise for neural network controllers in power system problems. Volume I contains the results of these applications of neural networks. Volume II contains two Ph.D. theses, which were written, in part, as a result of the research, and are provided as a supplementary reference. These two theses present algorithms and discuss contributions of the research to the science of neural networks, in the areas of control, nonlinear filtering, system identification and prediction.
- Research Organization:
- Electric Power Research Inst. (EPRI), Palo Alto, CA (United States); Stanford Univ., CA (United States). Information Systems Lab.
- Sponsoring Organization:
- Electric Power Research Inst., Palo Alto, CA (United States)
- OSTI ID:
- 205958
- Report Number(s):
- EPRI-TR-105533-Vol.1
- Resource Relation:
- Other Information: PBD: Nov 1995
- Country of Publication:
- United States
- Language:
- English
Similar Records
Optimal groundwater remediation using artificial neural networks and the genetic algorithm
Empirical modeling of nuclear power plants using neural networks