基于机器学习的火电机组一次风量软测量技术

Soft Sensor for Thermal Power Units Primary Air Flow Based on Machine Learning

在电厂燃煤机组中,一次风用于煤粉输送和锅炉燃烧,直接关系到炉膛内的实际燃烧工况,适当的一次风量对于磨煤机乃至整台机组的正常运行具有重要意义。然而受现场多种因素的影响,现有测量方法得到的一次风量误差很大。针对这一问题,基于最小二乘支持向量回归机算法建立了风量软测量模型,对辅助变量的选取及数据预处理方法进行了分析和讨论,并采用PSO算法对LSSVM软测量模型参数进行优化。以某电厂DCS历史数据中选取的数据作为训练样本和测试样本,对风量软测量模型进行了实验验证,结果表明该方法得到的预测值能够很好的跟踪实际风量的变化,且计算简便、预测速度快,具有较好的应用前景。

In the power plant coal-fired units, the primary air is used for boiler combustion and coal powder conveying. It directly related to the actual combustion chamber conditions. The appropriate primary air flow is of great significance for coal mill and even the whole units with the normal operation. However, due to influence of various factors, the error of primary air flow measured using existing methods is great. Aiming at the question, the air flow soft sensor model is established based on the least square support vector machine （LSSVM）. The method of selecting auxiliary variable and data preprocessing is discussed, and the parameters of LSSVM soft sensor model are optimized by PSO algorithm. Selecting a power plant DCS historical data as training samples and test samples, the air flow soft sensor model is tested. The results indicate that estimate of the model can forecast the change of actual air flow, and the calculate is simple, forecasting speed is rapid. The soft sensor method has a good application prospects.