ABSTRACT

The present work focused on modeling the nitrogen oxides (NO_X) generation characteristics based on the interpretable machine learning algorithm for an in-service coal-fired power plant. Computational Fluid Dynamics is available to obtain the NO_X generation data, which coupled with the historical operation data collected from Distributed Control System were used to improve the model’s prediction ability. The results showed that the depth and integrity of the dataset could be improved by adding simulation data. Compared with the Artificial Neural Network (ANN) and Support Vector Regression (SVR), the Gradient Boost Regression Tree (GBRT) model had higher accuracy than that of ANN and SVR model, and the GBRT model with more vital nonlinear transformation expression and time sequence is more suitable for the dataset, where the mean absolute error and coefficient of determination of the GBRT model were 3.85 and 0.98, respectively. Moreover, the Shapley additive interpretation analysis approach was presented for the GBRT model of NO_X generation prediction, which is helpful to the field operators to realize the efficient and low pollution operation of boiler equipment.

摘要

目前的工作重点是基于可解释的机器学习算法对在役燃煤电厂的氮氧化物 (NO _{X ) 生成特征进行建模。}计算流体动力学可用于获得 NO _X发电数据，结合分布式控制系统收集的历史运行数据，用于提高模型的预测能力。结果表明，通过添加模拟数据可以提高数据集的深度和完整性。与人工神经网络（ANN）和支持向量回归（SVR）相比，梯度提升回归树（GBRT）模型比人工神经网络和SVR模型具有更高的精度，并且GBRT模型具有更重要的非线性变换表达式和时间序列更适合数据集，其中 GBRT 模型的平均绝对误差和确定系数分别为 3.85 和 0.98。_{此外，还提出了用于 NO X}的 GBRT 模型的 Shapley 加性解释分析方法。发电量预测，有助于现场操作人员实现锅炉设备高效低污染运行。