The combustion process in a coal power plant is a complex process where optimisation is an objective to reach the best results from technical, economical and environmental points of view. This paper presents SEVILLA which is an intelligent tool able to help the operation of the combustion process of a coal power plant. Its objective is the diagnosis of deviations in respect to the normal dynamical behaviour of the variables that are most representative of the combustion process. In order to do this, several models based on neural networks characterise particular relationships among several variables which are important for the combustion process. The models are fitted using real data corresponding to different situations of the power plant when it is considered to be in normal operation. Once the models are available, they can be used in real-time taking information from sensors that monitor the combustion process. The models use real input and predict expected values for variables that are compared with their real values. If some important difference is observed, a diagnosis based on knowledge is started in order to infer a conclusion that can explain the problem observed. SEVILLA works in real-time taking images of the flame of the boiler and other relevant operation parameters at the Meirama coal power plant (550 Mw) belonging to Union Fenosa Generación, a Spanish electrical company. This paper will describe the different modules of SEVILLA and the experience collected till now.
Keywords: Combustion, coal power plant, expert system, neural network, on-line diagnosis, models of normal behaviour
Conferencia INFUB (6th European Conference on Industrial Furnaces and Boilers) 2-5 de Abril de 2002. Estoril (Portugal).
Published: April 2002.