燃煤锅炉飞灰含碳量的BP神经网络模型

BP neural network modelof fly ash carbon content of coal-fired boiler

燃煤锅炉是复杂的多变量系统，其飞灰的含碳量形成机理复杂，不能用简单的数学公式估算。现场实炉测试这些数据具有工作量大，测试工况有限等缺点；燃煤锅炉运行参数及燃料特性等因素影响着飞灰的含碳量，其相互耦合，导致分析数据过程困难。神经网络建模将燃煤锅炉视为黑箱，应用该方法可以良好的描述其输入输出之间的黑箱特性，因此，人工神经网络应用广泛。利用燃煤锅炉试验数据，采用3层BP（back propagation）神经网络构建了锅炉飞灰的含碳量排放特性模型。通过锅炉的实测数据验证，该BP神经网络对飞灰含碳量相对预测误差在0．19％～0．50％，预测效果良好。测试结果表明，建立的神经网络预测模型可以准确逼近验证样本数据，也能够较好的逼近非验证样本数据，具有良好的泛化能力。

Coal-fired boiler is a complicated multi-variable system, and the formation mechanism of the carbon content of fly ash is complicated which cannot be estimated by simple mathematical formula. The disadvantages of furnace tests include big data workload, test conditions, etc. Besides, the carbon content of fly ash is affected by factors of the coal-fired boiler characteristics, which causes the difficulty in data analysis. The modeling method is that it treats the coal-fired boiler as a b black-box can be well described. By using coal-fired boiler model was established tO conduct the analysi validation of the measured data shows that the operati advanta on parameters and the fuel ge of the neural network lack box, thus the input and output of the test data, a three layer BP neural network s on the boiler fly ash carbon relative prediction error of the and 0.50 %. Further, the results show that the data of both the validation sample can be accurately predicted with assistance of this modelling method. emission features. The model is between 0.19% sample and the ordinary