Author(s) :

Abdulrahman Rufai, Hassan A. Bashir.

Volume/Issue :

Abstract :

This paper develops the identification of thermal power plant using artificial neural network (ANN) technique taking Ajaokuta steel power plant as a case study. The thermal power plant, built in 1987, has a full capacity of 60MW at a frequency of 50Hz. Data was collected from the Ajaokuta steel power plant for the input/output parameters of the main components for the identification process. This include governor valve position, superheated steam flow rate to the turbine, turbine blades torque as well as the frequency of the generated electrical power. Firstly, identification models for the governor, turbine and generator were separately developed. To simplify the process, each model is treated as a single input single output (SISO) system. Then, the thermal power plant was modelled as a whole taking frequency as the input and generated electrical power as the output. The obtained data is used to train a feedforward neural network with one hidden layer in Matlab environment. An exhaustive search routine was developed to control the choice of the architecture (training function and number of layers of neurons) for the feedforward ANN models during the identification process. This optimizes the developed ANN model for every component of the power plant. A correlation analysis on the inputs/output dependencies reveals 95% confidence. To validate the developed ANN models, dynamic linear models were estimated using ARMAX method. Similarly, the order of each ARMAX model was chosen to minimize the mean square error. Compared to the ARMAX models, the accuracy of the proposed ANN models outweigh that of their numerical counterparts by 6.32, 0, 99.82, and 3.09 percent for the governor, turbine, generator and the complete plant respectively. Thus, the identified ANN models provide better representation of the thermal power plant.

No. of Downloads :

19