Dynamic nonlinear modelling of power plant by physical principles and neural networks

Dynamic modelling of power plants is fundamental to control system design and performance studies. This paper describes a nonlinear power plant model built by physical principles and neural network models by identification of the physical model. Every effort has been made to improve accuracy of the physical model without increasing its complexity. Practical aspects of neural network modelling for selecting testing data of the self-unbalancing system are investigated to ensure sufficient perturbations covering proper dynamic and load conditions. As an example, the generic modelling strategies are applied to a 200 MW oil-fired drum-type boiler-turbine-generator unit. The simulation results of the neural network and physical models are compared both at the trained and untrained conditions. It is shown that the accuracy of artificial neural network models depends greatly on the training data and is satisfactory within normal operating scope.