| 引用本文: | 李俊卿,陈雅婷,李斯璇.基于深度置信网络的同步发电机励磁绕组匝间短路故障预警[J].电力自动化设备,2021,41(2): |
| LI Junqing,CHEN Yating,LI Sixuan.Early warning of inter-turn short circuit fault in excitation windings of synchronous generator based on deep belief network[J].Electric Power Automation Equipment,2021,41(2): |
|
| 摘要: |
| 针对励磁绕组轻度匝间短路故障难以及时诊断的问题，提出一种基于深度置信网络(DBN)的同步发电机励磁绕组匝间短路早期故障在线预警方法。构建由2个DBN子网络组成的模型，分别以励磁电流、定子径向振动作加速度作为对应子网络的输出量；利用随机森林(RF)算法对特征的重要性进行排序，以此确定网络输出量的关联物理量，并将其作为对应子网络的输入量。通过改进粒子群优化(IPSO)算法选择网络参数，利用机组正常情况下的运行数据训练2个子网络。结合层次分析法(AHP)和反熵值法分配2个子网络的训练结果的权重，确定最终的故障预警阈值。将发电机故障情况下的数据输入训练好的网络，若总偏移距离大于故障预警阈值则判定为故障。以MJF-30-6型同步发电机为实验对象，证明所提方法可以实现同步发电机励磁绕组匝间短路早期故障预警。 |
| 关键词: 深度置信网络 同步发电机 励磁绕组匝间短路 故障预警 IPSO算法 |
| DOI：10.16081/j.epae.202011033 |
| 分类号:TM341 |
| 基金项目: |
|
| Early warning of inter-turn short circuit fault in excitation windings of synchronous generator based on deep belief network |
|
LI Junqing, CHEN Yating, LI Sixuan
|
|
School of Electrical and Electronic Engineering, North China Electric Power University, Baoding 071003, China
|
| Abstract: |
| Aiming at the problem that it is difficult to diagnose the mild inter-turn short circuit fault of excitation winding in time, an online early warning method for early fault of the inter-turn short circuit of excitation winding of synchronous generator based on DBN(Deep Belief Network) is proposed. A network model consisting of two sub-DBNs is constructed, and the excitation current and stator radial vibration acceleration are respectively taken as the output of the corresponding sub-DBN. The correlation physical quantities of network output, which are determined by the importance ranking sorted by RF(Random Forest),are taken as the input of the corresponding sub-DBN. The IPSO(Improved Particle Swarm Optimization) algorithm is used to select the network parameters. The parameters of the generator under normal operating condition are used to train the two sub-DBNs, and the weights of the training results are assigned by combining the AHP (Analytic Hierarchy Process) and the anti-entropy method to determine the final fault warning threshold. The data of the generator fault is input into the trained network. If the total offset distance is greater than the fault early warning threshold, it will be judged as a fault. Taking MJF-30-6 synchronous generator as an experimental object, it proves that the proposed method can realize the early warning of the inter-turn short circuit in excitation winding of synchronous generator. |
| Key words: deep confidence network synchronous generator inter-turn short circuit of excitation winding fault warning IPSO algorithm |
