Steam Temperature Deviation Modeling Of Supercritical Coal-fired Unit Based On Operation Data

Coal-fired power generation is still the main power source in China because of the energy structure.The unit is required to operate in non-design conditions to meet the volatility of the power load and the continuous increasing volume of wind and photovoltaic power,which brought challenges to the safe and reliable operation of the unit.As a key parameter of the coal-fired unit,the steam temperature tends to be inconsistent across the superheater during running at wide range varying conditions,the resulted temperature deviation is harmful to the stable and efficient operation.The factors affecting the steam temperature deviation are numerous and complex,and it is difficult to establish the mechanism model.The modelling methods based on operating data is investigated in this thesis.The unit operation data contains a large number of operating parameters.In order to establish an accurate model of the steam temperature deviation,the cause of the steam temperature deviation is analyzed from the operation principle.For a supercritical unit,the main operating parameters affecting steam temperature deviation are analyzed,the main modelling methods for steam temperature deviation are described.For the problem of feature selection in the data-driven modeling process,the correlation between various parameters and steam temperature deviation is analyzed based on partial mutual information(PMI),and the main operating parameters affecting steam temperature deviation are selected.The influence of PMI parameters on the results of variable selection is discussed,and the data set is obtained for steam temperature deviation modeling.A weighted mixed kernel function is constructed based on polynomial and Gaussian kernel functions to improve the limitations of the nonlinear expression ability of single kernel function.The modeling results on a unit show that the model based on the mixed kernel function has higher accuracy.To further improve the modeling performance,the particle swarm optimization(PSO)algorithm is used to optimize the parameters,including the parameters of the support vector regression modeling algorithm,such as penalty coefficient,Gaussian kernel function bandwidth,mixed kernel function weighting factor.The data-driven model with optimal parameters has better generalization performance.The experimental results on a 350 MW supercritical unit show that the resulting model can better describe the quantitative impact of operating parameters on steam temperature deviations.The results can provide scientific basis and operation reference for optimizing the adjustment to reduce steam temperature deviation during load varying.