基于独立成分分析预测电站锅炉NOx排放量

为了解决燃煤电站锅炉低NO_x燃烧特性建模输入值高维数据众多,以及大样本处理造成模型运行速度慢、精度低的问题,将独立成分分析（ICA）应用到建模数据预处理领域,提出一种基于快速独立成分分析（FastICA）的BP（back propagation）神经网络建模方法,并用该方法对某220 MW热电机组NO_x排放浓度进行预测。研究结果表明：经FastICA降维预处理后所建的神经网络模型（ICA-BP）性能优于直接构建的神经网络（BP）模型;ICA-BP模型计算结果与实测结果相对误差仅约2.5%,说明ICA方法有助于实现降低维数的同时保留更多原始数据特性的目的,是系统建模数据前处理的有效工具。

Forecasting the NOx Emissions from Utility Boilers Based on Independent Component Analysis

There are many high dimensional input values in the modeling process for low NO_x combustion property of coal-fired utility boilers, Moreover, processing such a large amount of sample data may slow down the calculations and reduce forecasting accuracy. To solve these problems, the independent component analysis(ICA) is applied to data pretreatment and then the back propagation(BP) neural network model based on fastICA algorithm is established in this paper. The ICA-BP model is used to predict the NO_x emission from a 220-MW thermal power unit. The results indicated that the neural network model based on fastICA algorithm outperforms the one without data preprocessing. The relative error of the ICA-BP model was only about 2.5% between the calculation and the measured results, in which the ICA method is verified to be an effective tool for data pretreatment in system modeling in the aspect of lowering the dimensions while more original data characteristic information still retained simultaneously is.