基于BP神经网络的煤粉锅炉飞灰含碳量研究

飞灰含碳量是反映电站煤粉锅炉燃烧效率的一个重要指标。基于误差反向传播（BP）神经网络方法，建立了11-23-1型BP神经网络模型。根据某电站四角切圆煤粉锅炉特点选取了煤粉细度、燃烧器摆角、烟气含氧量、5个煤种参数、燃烧器喷口运行组合等11个影响燃烧的参数作为神经网络的输入因子，对建立的模型进行训练，得到模型参数。以此进行预测，与实际值的误差不超过6％。在此基础上，又提出了单参数影响飞灰含碳量的简化分析方法，使神经网络包含的多维非线性规律在一定条件下简洁、直观地反映出来。计算和分析结果表明，本模型方法能有效提取各参数对飞友含碳量的影响规律，可用于锅炉飞灰含碳量的分析、预测和优化调节。

The Investigation of Carbon Content in Fly Ash for a BP Neural Network-based Pulverized Coal-fired Boiler

Carbon content in fly ash is a major index, which reflects the combustion efficiency of a pulverized coal-fired utility boiler. On the basis of a BP (inverse propagation of error) neural network method set up was a 11-23-1 type BP neural network model. In accordance with the specific features of a four-corner tangentially fired pulverized-coal utility boiler 11 parameters which can influence combustion have been selected to serve as input factors of the neural network. The parameters include: pulverized coal fineness, burner tilting angle, oxygen content in flue gas, parameters of 5 ranks of coal, operation combination of burner spray nozzles, etc. A training course was conducted for the established model and model parameters were obtained. The error predicted by using the model is less than 6% when compared with actual values. On this basis the authors have also proposed a simplified method for analyzing the carbon content in fly ash, which may be affected by a single parameter. This makes it possible to attain under certain conditions a concise and intuitive reflection of multi-dimensional non-linear law contained in the network. The results of the calculation and analysis indicate that the model-based method can effectively identify the mechanism of various parameters in influencing the carbon content in fly ash and may be employed to conduct the analysis, prediction and optimized regulation of carbon content in boiler fly ash.