摘要
针对火电厂SO_(2)污染物排放问题,提出了一种基于互信息的粒子群寻优(PSO)最小二乘支持向量机(LSSVM)模型预测方法,通过筛选出与SO_(2)实测入口浓度相关性较高的辅助变量,将其作为模型的输入,实现对主导变量SO_(2)浓度的预测。利用互信息筛选出的辅助变量相比于机理分析、皮尔逊相关性筛选出的辅助变量具有更高的相关性。利用互信息筛选出的辅助变量作为LSSVM模型的输入以及粒子群法确定LSSVM的参数,不仅缩短了计算时间,还提高了预测精度。将该方法应用到某火电厂的SO_(2)浓度软测量中,利用现场数据进行仿真,结果表明预测精度较高,相对误差较低,预测趋势更贴近实际值,减小了实际值与预测值的误差(均方根误差为2.485,平均相对误差为0.2603%),为现场的SO_(2)浓度提前控制提供了软件技术支持。
Aiming at the problem of SO_(2) pollutant emission in thermal power plant,a PSO-LSSVM model prediction method was proposed based on the mutual information.The auxiliary variables of high correlation with the measured inlet concentration of SO_(2) were selected as the input of the model to realize the prediction of the dominant variable SO_(2) concentration.The auxiliary variables screened by mutual information had the higher correlation with the auxiliary variables selected by mechanism analysis and Pearson correlation.The auxiliary variables selected by mutual information were used as the input of the LSSVM model and the particle swarm optimization(PSO)was used to determine the parameters of the LSSVM not only reduced the calculation time,but also improved the prediction accuracy.The method was applied to the soft measurement of SO_(2) concentration in a thermal power plant,and the simulation was carried out by using field data,which showed that the prediction accuracy was higher,the relative error was lower,the prediction trend was closer to the actual value,and the error between the actual value and the predicted value was reduced(the square root error is 2.485 and the average relative error is 0.2603%).It provided the software technical support for the on-site SO_(2) concentration advance control.
作者
金秀章
李京
JIN Xiu-zhang;LI Jing(School of Control and Coputer Engineering,North China Electric Power University,Baoding,Hebei 071003,China)
出处
《计量学报》
CSCD
北大核心
2021年第5期675-680,共6页
Acta Metrologica Sinica
