Automatic generation control of a multi area hydrothermal system using reinforced learning neural network controller

• Published in: International journal of electrical power & energy systems Vol. 33; no. 4; pp. 1101 - 1108
• Main Authors: Saikia, Lalit Chandra, Mishra, Sukumar, Sinha, Nidul, Nanda, J.
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.05.2011; Elsevier
• Subjects: Applied sciences; Automatic generation control; Bacteria; Bacterial foraging; Control systems; Dynamical systems; Dynamics; Electrical engineering. Electrical power engineering; Electrical machines; Electrical power engineering; Errors; Exact sciences and technology; Integral controller; Learning; Multilayer perceptron neural network; Neural nets; Neural networks; Operation. Load control. Reliability; Perturbation methods; Power networks and lines; Regulation and control; Reinforcement; Integral controller; Multilayer perceptron neural network; Neural nets; Bacterial foraging; Automatic generation control; Performance evaluation; Sensitivity analysis; Error estimation; Power system control; Control system; Reinforcement learning; Turbine; Neural network; Step response; Frequency characteristic; Backpropagation algorithm; Speed regulation; On line processing; Multilayer perceptrons; Frequency response; Comparative study; Multilayer network
• ISSN: 0142-0615; 1879-3517
• DOI: 10.1016/j.ijepes.2011.01.029

► Automatic generation control (AGC) of a three unequal area hydrothermal system. ► Reheat turbines in thermal areas and electric governor in hydro area are considered. ► The performance of a MLPNN controller using reinforcement learning is evaluated for the system. ► The performance of the MLPNN controller is compared with that of BFIC. This paper deals with automatic generation control (AGC) of a three unequal area hydrothermal system. Reheat turbines in thermal areas and electric governor in hydro area are considered. Appropriate generation rate constraints are considered in the areas. Bacterial foraging (BF) technique is used to simultaneously optimize the integral gains ( K Ii ) and speed regulation parameter ( R i ) keeping frequency bias fixed at frequency response characteristics. The integral controller in this case is termed as BFIC. The performance of a multilayer perception neural network (MLPNN) controller using reinforcement learning is evaluated for the system. In this reinforcement learning, the weights are dynamically adjusted online using backpropagation algorithm with error being the area control error (ACE). The performance of the MLPNN controller is compared with that of BFIC. Also, the performance of MLPNN controller over a wide range of system loading conditions and step load perturbations is compared with BFIC. Investigations clearly reveal the superior performance of MLPNN controller over BFIC. Sensitivity analysis subject to wide changes in system loading, inertia constant ( H) and size and location of step load perturbation is carried out to investigate the robustness of the controller with the optimum K Ii and R i obtained at nominal condition.