基于Wavenet的燃煤锅炉NOx排放浓度预测模型

NOx Emission Concentration Prediction Model for Coal-Fired Boilers Based on Wavenet

摘要: 基于数据驱动的方法对国内西北地区某火电厂660 MW燃煤机组的SCR脱硝入口NOx浓度建立预测模型。首先对火电厂DCS系统采集的数据进行数据预处理，包括开机判定、LOF算法异常值筛除，然后采用随机森林进行特征选择，最后分别采用Wavenet (波网)、LSTM (长短期记忆网络)、SVR (支持向量回归)三种方法分别对SCR脱硝入口NOx排放浓度进行预测。结果表明：无论是预测时间还是各项评价指标，Wavenet都表现更为优秀，其预测结果的平均百分比误差(Mean Absolute Percentage Error, MAPE)仅为1.31%，r2决定系数达到了0.97，充分显示Wavenet有比较好的预测能力。

Abstract: Based on the data-driven method, a prediction model was established for the NOx concentration at the SCR denitrification inlet of a 660MW coal-fired unit of a thermal power plant in northwest China. Firstly, the data collected by the DCS system of the thermal power plant is preprocessed, including the start-up judgment and LOF algorithm to screen out the data for outliers. Then, the random for-est algorithm is used for feature selection. Finally, the NOx concentration at the SCR denitrification inlet was predicted by Wavenet model, LSTM (Long Short-Term Memory Network) model and SVR (Support Vector Regression) model. The results show that Wavenet performs better in terms of prediction time and evaluation indicators. The mean absolute percentage error (MAPE) is only 1.31%, and the r2 coefficient of determination reaches 0.97. This fully shows that the proposed Wavenet model has relatively good predictive ability.