LSSVM动态软测量模型在磨煤机一次风量预测方面的应用  

Soft-sensing of Primary Air Flow in a Coal Mill Based on LSSVM

作　　者：杨耀权[1] 张新胜[1] 

机构地区：[1]华北电力大学控制与计算机工程学院,河北保定071003

出　　处：《动力工程学报》2016年第3期207-212,217,共7页Journal of Chinese Society of Power Engineering

摘　　要：针对磨煤机一次风量离线软预测模型难以满足机组变负荷要求的问题,建立了一种自适应修正预测模型参数的LSSVM动态软测量模型.以总的预报误差大小作为阈值来实时更新模型参数,该阈值无需人为参与设定,且能够根据负荷变化自适应改变,并采用网格搜索结合粒子群寻优算法得到LSSVM动态软测量模型中的2个最优参数,应用电厂实际运行数据建立软测量模型并对一次风量动态预测.结果表明:所建立的LSSVM动态软测量模型正确合理、预测精度高(相对误差波动小于1.5%)、实时性好,能很好地实现磨煤机一次风量的实时预测和估计,为磨煤机一次风量的在线监测提供数据支持.To solve the problem that the off-line soft-sensing model for primary air flow in coal mill of a power unit can not satisfy the requirement of variable load operation, a dynamic soft-sensing model based on I.SSVM was proposed, which is able to adaptively modify the model parameters by taking the total pre- diction error as a threshold to update the model parameters in real time without any manual work according to the load variation. Two optimal parameters of LSSVM model were obtained by grid search and PSO al- gorithm. The model was trained with actual operation data of a power plant, and was then used to dynami- cally predict the primary air flow. Results show that the LSSVM soft-sensing model is reasonable and ac- curate （maximal error less than 1.5%）, with good real-time performance, which is able to realize real-time prediction of the primary air flow, and therefore may serve as a reference for on-line monitoring of the pri- mary air flow in coal mills.

关 键 词：一次风量 动态软测量 最小二乘支持向量机 预报误差 运行数据 

分 类 号：TK313[动力工程及工程热物理—热能工程]