Heating Boiler Parameter Forecasting Based On Time-Series Model

In the modern industrial production process,in order to stable and safe operation of the industrial site,Industrial computer will real-time monitoring of various process parameters in industrial production and timely recording.In some specific industrial sites,However,One or more process parameters are particularly important to the production process and lag relative to the operational control.These parameters only can be rough controled with operational experience.If real-time prediction of these lag parameters is achieved and eliminate lag time difference,it not only contribute to improve control accuracy,but also prevent some misoperation.Therefore,Researching on time series prediction models start from the time series data recorded in the industrial field.And achieveing real-time forecasting of lag parameters depend on time series prediction model.Considering the time series characteristics of parameters recorded by the boiler combustion system,and the autoregressive moving average(ARMA)model is applied to predict the target variable firstly.This algorithm stablish a linear model of the target variable base on itself historical observations and random disturbance terms.Achieving effectively predict the target variable in the same condition.However,The effectiveness of the model is greatly reduced and mismatch occurs when conditions change.This is mainly because the ARMA model ignore the relationship between the target variable and other variables and result in weak generalization model.Aiming at the shortcomings of the univariate linear prediction model and consider the impact of multiple variables on the target variable in the system,a multivariable nonlinear prediction model is proposed.Support vector regression(SVR)is able to deal with multivariate relationship by nonlinear regression,and its parameters can be global optimization by using Particle Swarm Optimization(PSO).But the modeling process does not consider the time elements of time series data.Granger causality algorithm can extract feature variables from multivariate time series and get the causal relationship and the effective lag period.Support vector machine regression and Granger causality algorithm have significant complementary advantages in multivariate time series prediction.Therefore,the Granger-PSO-SVR model can be combined.In order to compare the model prediction results,three different predictive models were established for the oxygen content of the flue gas by using time series data generated by the boiler combustion system.The Granger-PSO-SVR model not only realizes real-time prediction of future values of target variables by using historical values of related variables but also reduced the predicted average absolute error rate by 0.83%,reduced the maximum error rate by 4.93%,and reduced the modeling time by 33%compared with PSO-SVM prediction model.Through the real-time prediction of the lag parameter by the prediction model,it can provide optimization guidance for the control operation or control algorithm of the lag parameter.So as to prevent the occurrence of misoperation and ensure the stability of the control effect.