论文部分内容阅读
汽轮机组特性随着机组老化而变化,传统上采用定期现场实测,需停机和采用专门的设备与系统,费用很高,因此基于现有自动化系统历史测量数据的特性曲线识别方法十分必要。一般汽轮机组汽耗量特性具有非凸和非连续等特点,常规的多元线性回归拟合不能适应。M5’模型树算法是一种多输入单输出系统的分段线性化的数据挖掘算法。提出采用M5’模型树的抽汽式机组汽耗量特性模型和其模型结构及参数识别算法,用于滚动利用最新的电厂测量历史数据获取最新的汽耗量特性。该方法简单、有效,逼近能力强,自动化程度高,在处理非凸形和非连续性的特性方程具有优势。通过多个热电厂的实时数据进行验证,具有很高的预测精度,效果优于多元线性回归拟合方程。
Turbine characteristics change with the aging of the unit. Traditionally, it is necessary to identify the characteristic curve based on the historical measurement data of the existing automation system by traditionally using on-site field measurements, stopping and using specialized equipment and systems. General Steam Turbine steam consumption characteristics with non-convex and non-continuous features, the conventional multiple linear regression fit can not be adapted. The M5 ’model tree algorithm is a piecewise linearized data mining algorithm for multiple input and single output systems. The steam consumption characteristic model of extraction unit with M5 ’model tree and its model structure and parameter identification algorithm are put forward. It is used to get the latest steam consumption characteristics by scrolling the latest power plant measurement history data. The method is simple and effective, has strong approximation ability and high degree of automation, and has advantages in dealing with non-convex and discontinuous characteristic equations. Validated by real-time data from multiple thermal power plants has a high prediction accuracy and is better than multiple linear regression fitting equations.