论文部分内容阅读
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 boilers are complex multivariable systems and their carbon content in fly ash is complex and can not be estimated using simple mathematical formulas. On-site real furnace test These data have the disadvantages of heavy workload and limited test conditions. The coal-fired boiler operating parameters and fuel characteristics affect the carbon content of fly ash, which leads to the difficulty in analyzing the data. Neural network modeling regards coal-fired boiler as a black box. Using this method, the black box characteristics between input and output can be described well. Therefore, artificial neural network is widely used. Based on the experimental data of coal-fired boiler, a three-layer back propagation (BP) neural network was used to construct a carbon emission model of boiler fly ash. The measured data from the boiler verified that the relative prediction error of the carbon content of fly ash by BP neural network is between 0.19% and 0.50%, and the prediction effect is good. The test results show that the proposed neural network prediction model can accurately approximate the verification sample data and also can approximate the non-verification sample data well and has good generalization ability.