基于主成分分析与神经网络复合模型的汽轮机排汽焓计算

以某300 MW汽轮机为例，建立了基于主成分分析与神经网络复合模型的汽轮机排汽焓计算模型。首先分析了主成分分析和人工神经网络计算原理，然后采集了影响汽轮机排汽焓的各个主要参数的历史数据，并对采集到的数据进行了数据预处理，对剔除坏点后的历史数据做主成分分析，得到了累计贡献值大于99.95%的4个主要成分，并以这4个主要成分作为BP神经网络的输入变量，汽轮机排汽焓作为输出变量，建立基于主成分分析与神经网络复合模型的汽轮机排汽焓计算模型，通过对模型的训练和验证，得到了汽轮机排汽焓计算模型，便于在线监测中进行实时调用。研究结果表明：主成分分析能够确定合理的BP神经网络输入变量个数，提高训练精度和训练速度；主成分分析与神经网络复合模型对排汽焓的计算精度符合工程要求；排汽焓在各个负荷工况下波动不大。

The Calculation of Turbine Exhaust Enthalpy Based on the Hybrid Model of the Principal Component Analysis and the BP Neural Network

Taking a 300 MW turbine as an example, in this paperthe calculation of turbine exhaust enthalpy based on the hybrid model of the principal component analysis and the BP neural network is established. The principal component analysis and the BP neural network are introduced at first. Then,the historical data are collected as the main parameters that affect the steam turbine exhaust enthalpy.The data pre-processing is applied to exclude the bad points. The four major components, with the cumulative contribution value greater than 99.95%, are identified. The calculation model of turbine exhaust enthalpy is established with the four identified components as the BP neural network input parameters and the steam turbine exhaust enthalpy as the output parameter. After being trained and tested, the calculation model of turbine exhaust enthalpy is obtained to facilitate the real-time online monitoring. The results show that the principal component analysis can help to determine the reasonable BP neural network input parameters and improve the accuracy and the speed of the training. The precision of the hybrid model meets the requirements of the project. The fluctuation of the exhaust steam enthalpy is not big in all working conditions.