基于多元状态估计技术的给水泵汽轮机故障预警研究

Research on Fault Warning of Steam Turbine of Feed Water Pump Based on Multivariate State Estimation Technique

为实现给水泵汽轮机故障早期预警,提出了一种基于多元状态估计技术(MSET)的参数估计和基于残差预警的故障早期预警方法。以某350 MW超超临界机组的给水泵汽轮机为研究对象,选取表征其运行特征的关键参数,并且采集分散控制系统(DCS)运行历史数据。通过截断法和箱线图法对异常数据进行清洗,采用等间距抽样法构建运行参数的正常工况样本库,基于MSET得到参数估计值,并采用滑动窗口法确定残差预警阈值,建立了给水泵汽轮机故障早期预警模型。最后,用实际运行数据进行验证。结果表明:在正常工况下,模型估计值的平均相对误差小于3%,能有效识别给水泵汽轮机运行中的异常状态;将模型应用到电厂的大数据平台,实现了给水泵汽轮机各工况的智能监测和故障早期预警,具有较大的工程实用价值。

In order to realize the fault early warning of steam turbine of feed water pump, based on parameter estimation of multivariate state estimation technique(MSET) and residual warning, a fault early warning method was proposed. The steam turbine of feed water pump of a 350 MW ultra-supercritical unit was selected as the research object, the key parameters characterizing its operation characteristics were selected, and the operation history data from distributed control system(DCS) were collected. The abnormal data were cleaned by truncation method and boxplot method, and the normal operating condition sample database of operating parameters was constructed by equidistant sampling method. The parameter estimates were obtained based on MSET, and the residual warning threshold was determined by sliding window method, and the fault early warning model of steam turbine of feed water pump was established. Finally, the actual operating data was used for verification. Results show that, under normal working conditions, the average relative error of the estimated value of the model is less than 3%, which can effectively identify the abnormal state in the operation of the steam turbine of feed water pump. The model is applied to the big data platform of the power plant to realize the intelligent monitoring and fault early warning of the steam turbine of feed water pump, which has great engineering practical value.