为了解决国内对火电厂引风机故障预警方法相对缺乏的问题，本文提出了一种基于SSAPSOLightGBM的故障预警算法。通过建立LightGBM(Light gradient boosting machine)正常轴承温度预测模型，并创新性地引入融合麻雀搜索算法的改进粒子群优化算法（SSAPSO）优化模型超参数，最终获得引风机轴承温度预警阈值，实现引风机早期故障预警。实验证明，基于SSAPSOLightGBM的故障预警方法在预测精度、泛化能力等方面相比传统预算法效果更好；该方法能够提前2 h对风机进行故障预警，对火电厂运维具有一定的指导意义。

In response to the relative lack of fault waming methods for induced draft fans in thermal power plants in China,a novel fault early warning method based on SSAPSOLightGBM is proposed in this paper.To set warning threshold and realize novel fault early warning for induced draft fan. normal bearing temperature prediction model is established by LightGBM and SSAPSO is introduced innovatively to optimize the hyperparameters of SSAPSOLightGBM.The experimental results show that compared with SVM and other traditional machine learning prediction algorithm the novel fault early warning method based on SSAPSOLightGBM is more accurate and has a higher generalization ability.This method can predict induced draft fan fault 2 hours in advance and has a certain guiding significance for thermal power plant operation.

TM621.7

上海市“科技创新行动计划”地方院校能力建设专项项目（19020500700）