论文部分内容阅读
对利用人工神经网络方法来预测电站锅炉在未知的燃烧或运行工况下烟气中汞组分进行了可行性评估.基于已掌握的三个电站锅炉现场测试的汞排放数据库,建立了一个三层误差反向传播神经网络模型用以对烟囱处汞排放的组分进行预测.全部预测过程包括:数据的采集整理、构建人工神经网络模型、训练过程和误差评估4部分.总共选取了59个煤样、灰样以及电站运行工况参数作为输入变量,利用部分实际汞排放测试数据来指导训练过程,其余的实测数据用来校验网络预测模型的准确性.结果表明,模型获得的预测精度对单质汞元素的均方根误差为0·8μg/Nm3,对全汞的均方根误差为0·9μg/Nm3.这样的误差在当考虑到现场采用半连续释放测量(SCEM)方法,由湿法测试模块所产生的峰值误差时是完全可以接受的.
The feasibility of using artificial neural networks to predict the mercury content of flue gas in power plant boilers under unknown combustion or operating conditions was evaluated.According to the available mercury boiler data from three field boiler tests in power plant boilers, The layer error back propagation neural network model is used to predict the components of mercury emissions from the chimney.The whole forecasting process includes collecting and arranging data, constructing artificial neural network model, training process and error assessment.A total of 59 Coal samples, ash samples and power station operating parameters as input variables, some of the actual mercury emissions test data to guide the training process, the rest of the measured data used to verify the accuracy of the network prediction model.The results show that the prediction accuracy obtained by the model The root mean square error of elemental mercury was 0.8 μg / Nm3 and the root mean square error of total mercury was 0.9 μg / Nm3. Such errors were caused by the use of the semi-continuous emission measurement (SCEM) Wet test module generated by the peak error is completely acceptable.