Prediction Of Coal Ash Fusion Point Based On Support Vector Machine

Currently,Coal fired power plants in C hina burn more mixed coal,which causes many problems of Slagging.and slagging is one of the most important factors that affect the safety of coal fired boilers.In order to solve the safety problems,it is significant to conduct the ash fusion point measurement,prediction and analysis.This paper has Analyzed and summarized the factors effecting on the coal ash fusion point which include the reaction atmosphere and the coal ash composition.Combined with the actual situation of a power plant,the experiment of the mixed coal ash fusion point test was carried out under the condition of Weak reducibility;Based on the analysis and description of the basic principle and parameter selection method of Support Vector Machine,Support vector machine was used to predict the ash fusion point(softening temperature,ST value)of the experimental coal samples and mixed coal samples and at the same time,the standard of evaluation result of the prediction is given:(1)accuracy index,mean absolute percentage error(MAPE)and mean square error MSE;(2)reliability index,error range Rag.In this paper,the problem of insufficient representation of training samples in the prediction process of support vector machine is analyzed in detail.By increasing the number of training samples,respectively,20 kinds of experimental coal samples in 2,4 and 6 kinds of single coal data are added into the training samples,then I analyzed the remaining experimental coal sample and I found that increas ing the number of training samples is helpful to improve the prediction accuracy.Similarly,the number 8 ~ 13 of blended coal samples are added into the initial training sample in order to improve representative of training samples and the results show that: for the prediction of prediction of blended coal coal mixed 1 ~ 6 number and 19 ~ 36,it have a good accuracy and the mean absolute percentage error MAPE is 1.90%;the mean square error of MSE is 1050;the error range for Rag is(-55,15)?.It is worth mentioning that,although the number 8 ~ 13 of the blended coal samples' ST values hve not been predicted,by adding them to the training samples,using this method the prediction results for similar coal samples can avoid the probability of the maximum deviation.