The feasibility of using an ANN method to predict the mercury emission and speciation in the flue gas of a power station under un-tested combustion/operational conditions is evaluated. Based on existing field testing ...The feasibility of using an ANN method to predict the mercury emission and speciation in the flue gas of a power station under un-tested combustion/operational conditions is evaluated. Based on existing field testing datasets for the emissions of three utility boilers, a 3-layer back-propagation network is applied to predict the mercury speciation at the stack. The whole prediction procedure includes： collection of data, structuring an artificial neural network （ANN） model, training process and error evaluation. A total of 59 parameters of coal and ash analyses and power plant operating conditions are treated as input variables, and the actual mercury emissions and their speciation data are used to supervise the training process and verify the performance of prediction modeling. The precision of model prediction （ root- mean-square error is 0. 8 μg/Nm3 for elemental mercury and 0. 9 μg/Nm3 for total mercury） is acceptable since the spikes of semi- mercury continuous emission monitor （SCEM） with wet conversion modules are taken into consideration.展开更多
基金The National Basic Research Program of China (973Program) (No.2006CB200302)the Natural Science Foundation of JiangsuProvince (No.BK2007224).
文摘The feasibility of using an ANN method to predict the mercury emission and speciation in the flue gas of a power station under un-tested combustion/operational conditions is evaluated. Based on existing field testing datasets for the emissions of three utility boilers, a 3-layer back-propagation network is applied to predict the mercury speciation at the stack. The whole prediction procedure includes： collection of data, structuring an artificial neural network （ANN） model, training process and error evaluation. A total of 59 parameters of coal and ash analyses and power plant operating conditions are treated as input variables, and the actual mercury emissions and their speciation data are used to supervise the training process and verify the performance of prediction modeling. The precision of model prediction （ root- mean-square error is 0. 8 μg/Nm3 for elemental mercury and 0. 9 μg/Nm3 for total mercury） is acceptable since the spikes of semi- mercury continuous emission monitor （SCEM） with wet conversion modules are taken into consideration.
