飞灰含碳量自适应校正WLSSVM软测量模型

A WLSSVM based self-adaptive correction soft-sensing model for carbon content measurement in fly ash

针对锅炉飞灰含碳量难以长期准确预测的问题,从提高模型预测精度和自适应能力的角度出发,提出一种基于模型预测性能评价的自适应校正加权最小二乘支持向量机(WLSSVM)软测量模型。构造了基于最大线性无关组的软测量模型训练样本集,使WLSSVM模型具有较好的稀疏性,并减少了训练过程的计算量;建立基于数据相似度加权因子的WLSSVM软测量模型,利用双种群差分进化算法进行模型参数的优化选取;通过模型预测性能在线评估和递推校正实现了模型在线自适应校正。在某台300MW机组锅炉上进行的仿真试验结果表明,该算法模型具有良好的预测精度和自适应能力,能够有效预测锅炉飞灰含碳量。

The carbon content in fly ash is difficult to predict correctly,which has been a problem in a long term.Thus,from such aspects as improving the prediction precision and self-adaptive ability of the model,a model prediction performance evaluation based self-adaptive correction WLSSVM soft-sensing model was proposed.In order to enhance the sparsity of the WLSSVM model and reduce the calculation amount during training process,the maximal linearly independent group based training sample set of the soft-sensing model was constructed.The data similarity weighting factor based WLSSVM soft-sensing model was established,and the the double population differential evolution algorithm was employed to optimize the model parameters.By online evaluation and recursive correction of the model prediction performance,the online self-adaptive correction of the model was realized.The simulation test was conducted on a 300 MW coal-fired boiler.The results showed that the proposed soft-sensing model had high prediction accuracy and good self-adaptive ability,and can predict the carbon content in fly ash effectively.