Study On Automatic Generation Control Based On Fuzzy-Q Algorithm

As the most basic function of power generation scheduling and control,automatic generation control (AGC) plays an important role in the control system andstable operation of interconnected power systems, for it is directly related to the safetyand economic operation of the power system. Based on the power market mechanism,modern power grids have been developed into multiple-area interconnected systems.Automatic generation control is the mainly means to regulate the power andfrequency of the modern interconnected power system, therefore, its controlperformance will have an effect on power quality. The error of power and frequencyof the tie-line in interconnected power system will be changed as the load fluctuates.How to control the active power output of generators to track the random variation ofthe load and to improve the power quality have been one of appealing subjects now.Intelligent control algorithms have some highlights that traditional ones do nothave, such as good convergence, adaptability, robustness and so on. On the basic ofexisting research results, this paper presents two orientations to introduce Q-learninginto fuzzy control to optimize the relevant parameters of the universes and rules, andthen applied these two new algorithms to automatic generation control. The mainwork the writer has done includes:（1）This paper introduces the basic principles, advantages and disadvantages offuzzy control and reinforcement learning, and the control object AutomaticGeneration Control are also outlined.（2）In the third chapter, this paper reflects some well-known concepts andcontrol theory of traditional fuzzy controller, proposes the structure of AGC systembased on fuzzy control.（3）In consideration of the shortcomings of traditional fuzzy controller, which isthat its adjustment is usually rather rough, this paper shows a method which optimizethe universes of the fuzzy controller by Q-learning to overcome this problem. In thisproposed method, contraction-expansion factors and geometric factors are used toregulate the universes to improve its adjustment, thereby reducing the distortion in the process of controlling.（4）In the fifth chapter, this paper describes how to optimize the parameters offuzzy rules by Q-learning. This new method redistributes the reinforcement value ofeach action in each rule, to enhance the control performances.