基于BP神经网络的水电厂排水泵故障预警研究

Early Warning of Drainage Pump Faults in Hydropower Plant based on BP Neural Network

基于BP神经网络的水电厂排水泵故障预警方法,通过分析排水系统的历史运行数据,确立激活函数、隐含层节点数等重要参数,从而建立BP神经网络模型;再根据集水井水位、来水量预测、排水泵的理论运行时间,与实测数据比对,发现异常及时给运行人员推送告警,安排检修。相比于传统的限值预警,该系统在某大型水电厂计算机监控系统内试运行中的预测精度良好,为排水泵及早发现故障提供了一种新手段。由于本系统故障判定采用的是固定阈值方法,随着排水泵老化,其排水能力会越来越差,如何动态的确定故障阈值将是一个值得研究的问题。

An early warning system for drainage pump faults in hydropower plant is introduced. A BP neural network model is built by analyzing the historical operation data of the drainage system and defining important parameters such as the activation function and the number of hidden layer nodes. Then, the predicted operation time of the drainage pump according to the water level of the collecting well and the inflow water is compared with the actual measurement data. In case of any abnormality, warning message is sent to operation personnel timely for maintenance arrangement. Compared with the traditional limit warning system, the new system is of good prediction accuracy in its trial operation in the computer monitoring system in a large-scale hydropower plant, which provides a new approach for the early fault detection of drainage pumps. However, with the ageing of drainage pumps, the drainage capacity is getting worse. While the warning system adopts fixed threshold method for fault determination. The dynamic determination of the fault threshold is still a challenging topic.