基于改进模糊聚类与IPSO-SVM的燃煤电站NO_x排放多模型预测

Multi-model NO_x Emission Prediction Based on IGASA-FCM and IPSO-SVM for Coal-fired Power Plants

通过挖掘大量脱硝系统现场运行数据,提出一种新的多模型选择性催化还原(SCR)脱硝系统建模方法:首先对SCR脱硝系统进行理论分析和实际运行研究,应用改进遗传模拟退火的模糊聚类算法对训练集进行聚类划分,得到最优聚类效果;然后建立相应的支持向量机子模型,并采用改进的粒子群算法对模型参数进行优化,所建立的子模型通过隶属度值加权融合得到最终的整体预测模型。以某电站锅炉脱硝系统为例,对所提出的方法进行验证,并与其他建模方法进行比较。结果表明:所建立的模型具有较高的泛化能力和预测精度。

A novel multi-model modeling method was proposed for the SCR denitrification system by mining massive historical data of the system. Through theoretical and practical analysis of SCR denitrification systems, the training data set was divided to obtain the optimal clustering results using fuzzy C-means clustering(FCM) based on improved genetic algorithm simulated annealing(IGASA). Then corresponding sub-models were established by support vector machine(SVM), and the sub-model parameters were subsequently optimized based on improved particle swarm optimization(IPSO). Finally, an integral prediction model was built up by weighted fusion of the sub-model membership values. Taking the denitrification system of a power plant boiler as an example, the model was verified and then compared with that built by other modeling methods. Results show that the model proposed has strong generalization ability and high prediction accuracy.