针对目前变压器结构复杂、绕组早期故障样本数据量小的问题，为提高变压器绕组故障类型诊断的准确率，提出基于ReliefF-mRMR漏磁场特征优选和POA-LSSVM的变压器绕组早期故障诊断模型。校验变压器物理实体与仿真模型的一致性，将变压器绕组早期故障的漏磁场信息作为故障特征状态量，通过ReliefF和mRMR(maximum relevance and minimum redundency)特征选择算法对漏磁场故障特征优选、提取关键特征，经优选后故障特征量输入最小二乘支持向量机(least squares support vector machine, LSSVM)中进行故障诊断，并用鹈鹕优化算法(pelican optimization algorithm, POA)优化LSSVM参数，结果表明经过ReliefF-mRMR算法故障特征优选后的POA-LSSVM故障诊断模型能够有效区分不同的变压器绕组早期故障类型，相比于GA(genetic algorithm)-LSSVM算法、PSO(particle swarm optimization)-LSSVM算法、POA-LSSVM算法，在故障诊断效率和分类正确率方面均有显著提高，最后通过变压器的动模实验进行了故障模拟实验，验证该故障诊断模型的有效性。
英文摘要:
Aiming at the problem of complex transformer structure and small amount of early winding fault sample data, in order to improve the accuracy of transformer winding fault type diagnosis, this paper proposes an early fault diagnosis model of transformer winding based on ReliefF-mRMR leakage magnetic field feature optimization and POA-LSSVM. Firstly, the consistency of the physical entity of the transformer and the simulation model is verified. The leakage magnetic field information of the early fault of the transformer winding is taken as the fault characteristic state quantity, and the fault characteristics of the leakage magnetic field are optimized and the key features are extracted by the ReliefF and mRMR feature selection algorithms. After optimization, the fault feature is input to the least squares support vector machine (LSSVM) for fault diagnosis, and the LSSVM parameters are optimized by the pelican optimization algorithm (POA). The results show that the POA-LSSVM fault diagnosis model after fault optimization by ReliefF-mRMR algorithm can effectively distinguish different early fault types of transformer windings, and compared with GA-LSSVM algorithm, PSO-LSSVM algorithm and POA-LSSVM algorithm, the fault diagnosis efficiency and classification accuracy are significantly improved. Finally, the dynamic simulation experiment of transformer is carried out to verify the effectiveness of the fault diagnosis model.