基于改良BP神经网络的生物质锅炉飞灰含碳量预测模型研究

针对生物质锅炉飞灰含碳量较高的问题,文章提出了基于主成分分析法(PCA)或Garson算法与普通LM-BP神经网络相结合的两种生物质锅炉飞灰含碳量预测模型。这两种模型通过对负荷、燃烧室烟气温度、烟气含氧量等17个原始输入变量进行降维得到新输入变量,再进行训练建模,提高了模型精度。利用我国某生物质电厂飞灰含碳量的实测数据对模型进行检验,检验结果表明,LM-Garson-BP神经网络的MAPE为2.09%,MSE为0.11,MAE为0.25,泛化能力最强,稳定性最好。

Prediction models of the carbon content of fly ash in a biomass boiler based on improved BP neural networks

Aiming at the high carbon content of fly ash in biomass boilers,two prediction models based on principal component analysis(PCA)or Garson algorithm combined with common LM-BP neural network were proposed.New input variables were obtained by reducing the dimensions of 17 original input variables such as load,combustion temperature and oxygen content,etc.,improving the accuracy of the models.The models were tested by using the measured data of the carbon content of fly ash in a biomass power plant in China.The results showed that the MAPE,MSE and MAE of the LM-Garson-BP neural network were 2.09%,0.11 and 0.25,respectively.Therefore,the generalization ability of the LM-Garson-BP neural network was the strongest and the stability was the best.