HOPFIELD METHODS FOR ECONOMIC DISPATCH OF POWER SYSTEMS

• Main Authors: Chiou, Gwo-Jen, 邱國珍
• Other Authors: Su Ching-Tzong
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/94304658300627881356

博士 === 國立中正大學 === 電機工程學系 === 86 === Due to the effectiveness of the Hopfield neural network (HNN) on solving combinatorial optimization problems, this dissertation proposes a series of Hopfield network approaches for solving the economic dispatch (ED) problems of power systems. In this research, two continuous Hopfield model based algorithms are first proposed to solve several basic ED problems. Since generating units may have prohibited operating zones lying between their minimum and maximum power outputs, the corresponding ED problem become a nonconvex optimization problem. To solve such nonconvex ED problems, the research proposes a novel strategy to prevent the unit from operating into prohibited zones. The Hopfield model essentially belongs to one kind of the gradient method, sticking at one of the local minimum points has become an unavoidable drawback while solving the nonconvex optimization problem. A novel neural network model which combines the Hopfield model with the simulated annealing (SA) technique is proposed. Conceptually, the model solution normally moves in the direction of decreasing energy, but sometimes we intentionally allow it to accept uphill movements following a probabilistic acceptance criterion, such that it will finally search out the global optimum solution for the nonconvex ED problem. To overcome the problems of curve saturation and improper selection of the shape constant of the sigmoidal function, a novel linear model is proposd to describe the input-output relationship of the neuron. Conceivably, this linear model is much more efficient, which can greatly reduce computational efforts required. Finally, a direct-computation Hopfield model is proposed to solve the ED problems. The method employs the linear input-output model for neurons to formulate analytical solutions for ED problems. By applying such formulations to the problems, direct computation instead of lengthy iteration becomes possible. Unlike conventional Hopfield methods which select the weighting factors of the energy function by trial and error, the proposed method determines the corresponding factors by calculation and is therefore relatively easy to apply. The effectiveness of the proposed methods is demonstrated by investigating several examples. Computational results from the proposed methods and the conventional methods are also compared.