Modelling Of Boiler’s Convective Heat Transfer Surfaces Fouling On Large Thermal Generation Units

Fouling and slagging on convictive heat transfer surfaces have a significant effect on the economic and safe performance of coal-fired boiler of power plant. In most cases, the power plants use a approach of scheduled cleaning to clean ash deposits on surfaces, which usually depends on operating experience or recommended scheme from boiler design specification. According to this approach, operators clean fouling of each heat transfer surfaces sequentially without taking into account of their different fouling levels, which has a poor economy. This work is aimed to build models of fouling and slagging problem on convictive heat transfer surfaces by studying the theories and methods about the on-line monitoring。Firstly,the principle of on-line monitoring of fouling and slagging on convictive heat transfer surfaces based on mechanism model and its limitations are introduced in detail. On this basis, combined with the Analysis of some factors, such as working conditions and coal quality, a model about solving the issue of monitoring fouling and slagging on convictive heat transfer surfaces is built by using Artificial Neural Network(ANN). The model has been optimized through a detailed description of sample selection and processing, input and output Parameters Picking, ANN architecture design and case sensitive test. The weights of designed neural network are optimized by using Genetic Algorithm(GA) and Particle Swarm Optimization(PSO), in order to overcome the problem of the BP neural network easily falling into local minima and its poor generalization ability. At last, after a comparation of optimization effect between the two methods, a monitoring model by using PSO-BP is built.A 600MW boiler of Shanxi Tashan Power Station is taken as the research object in this work. To obtain accurate sample data for model training, a series of soot-blowing experiments were conducted. on this basis, the relevant simulation experiments were completed on MATLAB platform in order to verify the accuracy and practical applicability of this model. The result shows that the model has good generalization ability, and the predicted results of this model in different working conditions have higher accuracy and reliability, which can meet the requirements of actual monitoring.