Research On Measurement Of Primary Air Flow Based On Least Squares Support Vector Machine

With the rapid growth of China's economy,the demand for electricity is increasing.However,with the increasing exhaustion of energy and the implementation of the policy,every coal-fired power plant is forced to improve its operating economy under the precondition of safe production.In the coal-fired boiler system,Primary air flow system is an important part.Primary air flow accurate,real-time measurement is the key to ensure a reasonable distribution of air and optimizing combustion.Under the existing conditions of industrial environment production,some thermal parameters of the primary air flow system are restricted by the conditions of hardware instrument measurement and the environment of the site,which are difficult to measure accurately.It has a direct effect on the stability and economy of the entire industrial control and the operating system,which is not good for the productivity of the coal-fired power plant.In response to the above problems,this master's thesis uses soft measurement technology to measure the Primary air flow.First,this master's thesis combines theoretical analysis and data driven modeling way,emphatically analyzed the various factors affecting the Primary air flow.Meanwhile,this graduate thesis selects the DCS history sample data by correlation analysis,using mechanism analysis to determine the appropriate auxiliary variables.Secondly,in the process of modeling,the collected historical sample data is pretreated by the laedda criterion and normalized method to eliminate the influence of singular and isolated point.At the same time,a static measurement model is built based on the least square support vector machine algorithm(LSSVM).In view of the lack of sparse solution and the problem of long operation time,this master's thesis combined linear equation of compression perception and LSSVM,which uses the nuclear matrix as the dictionary.Then in the process of training,the orthogonal matching tracking algorithm is used to compress the support vector sets to realize the sparse of LSSVM and improve the modeling speed.Using the improved cloud adaptive particle swarm optimization algorithm for optimizing model parameters.In dynamic modeling,the model of sparse vector is used in the static modeling process,and the measurement precision is lost.This master's thesis presents a multi-scale LSSVM based on wavelet theory,which improves the measurement accuracy.In addition,based on the online correction algorithm,the model can adapt the model parameters with the change of the operating conditions.Therefore,the model can meet the real-time requirement in different working conditions.Based on the above theoretical method,this master's thesis uses the field historical data to complete the initial data processing and auxiliary variable selection,and constructs an online measurement model of primary air flow.The simulation results show that the proposed model can achieve the ideal measurement accuracy and provide effective data support for the reasonable distribution of the coal-fired units.