Research On Data-driven Modeling Methods In Power Plant Thermal Process

Magnanimity of data are accumulated during the operating time of power plants,which reflect the running condition of units.Data-driven modeling methods are effective ways to convert data into information to help people estimate the running level and improve the safety and economy of units.With the development of informatization of power plants,which makes it easier to storage and obtain the operation data,data-driven modeling methods will own widely applied spaces in the future.The main research contents are following:First of all,to solve the hybridity in thermal process data,the pretreatment methods and dynamic behaviour of thennal process data were studied in this paper,focusing on the wavelet de-noising method and R-statistic method,the selection of wavelet transform parameters and steady-state threshold were studied.This paper combined those two methods together,results showed that after de-noising,more steady-state data were extracted in the same steady-state threshold.Secondly,the dynamism of thermal process will influce the perfonnance evaluation of system and equipments,this paper use heat rate and SCR system to introduce the dynamic modification method to acquire static relation from dynamic data.Simulation platforms were built to verify the method.For another,the dynamic parts in SCR reactor were researched and came up with the correction methods of NH3/NOX and denitration efficiency.Using operating data to identify time constant,simulation platforms were built to generate simulated data to verify the methods.Results show that methos can eliminate inconsistency of real-time data effectively,and the adjusted data can reflect static characteristics of reactor.Furthermore,the adaptive modeling methods of thermal process were researched in this paper,and the Cluster-RBF-based regression algorithm for concept drifting data streams was proposed to automatically adjust the model when the character of thermal object varies.This algorithm can ensure high precision and right input output relationship in whole working condition.This paper used test function data and CSTR process data to verify this algorithm and showed how it worked in getting degradation trend of SCR denitrification catalyst.Finally,based on the China Datang Corp SCR denitrification catalyst management database platform,the life monitoring modules of denitration catalyst was developed and the software design idea,system structure,operating process was given.