基于SSA-XGBoost的变压器故障诊断

Transformer fault diagnosis based on SSA-XGBoost

针对传统机器学习算法处理绝缘油中溶解气体分析(Dissolved gas analysis,DGA)的数据集准确率低的问题,提出了樽海鞘优化算法(Salp swarm algorithm,SSA)优化极端梯度提升(Extreme gradient Boosting,XGBoost)的DGA故障检测模型。对DGA数据集进行Z-score标准化预处理,消除DGA数据集各个特征间差异过大的影响;初始化SSA种群参数、迭代次数,设置樽海鞘群体的上下限,分别对应XGBoost里面的各个参数;经过樽海鞘群体的连续迭代优化,寻找XGBoost的最优参数;利用寻找到的诊断模型XGBoost的最优参数进行DGA的故障诊断。与遗传算法(Genetic algorithm,GA)优化XGBoost、粒子群算法(Particle swarm optimization,PSO)优化XGBoost和蚁群算法(Ant colony optimization,ACO)优化XGBoost的对比实验表明,针对DGA数据集,SSA相比于GA、PSO和ACO优化的XGBoost模型各个故障的准确率更高,总体准确率达到了93.4%,SSA更容易找到XGBoost的最优参数,更容易避免XGBoost的过拟合问题,能够实现对复杂样本的有效分类。结果表明,SSA-XGBoost在处理电力变压器的故障诊断有着较高的准确率,是诊断油浸式电力变压器故障的有效模型。

To address the problem of low accuracy of traditional machine learning algorithms in processing Dissolved gas analysis(DGA)datasets,the Salp swarm algorithm(SSA)optimized Extreme Gradient Boosting(XGBoost)DGA fault detection model is proposed.The DGA dataset is preprocessed with Z-score standardization to eliminate the influence of the large differences between features in the DGA dataset.The SSA population parameters and iteration times are initialized,and the upper and lower limits of the Salp swarm are set,which correspond to the various parameters in XGBoost.After continuous iterative optimization within the population,we aim to find the optimal parameters for XGBoost.The optimal parameters found diagnostic model of XGBoost is used to perform fault diagnosis on the DGA dataset.The comparative experiments involving the optimization of XGBoost using Genetic algorithm(GA),Particle swarm optimization(PSO),and Ant colony optimization(ACO)on a DGA dataset have demonstrated that,in contrast to GA,PSO,and ACO,SSA yields higher accuracy for various failure cases of the XGBoost model.The overall accuracy reaches 93.4%.The experiments suggest that SSA is more adept at finding the optimal parameters for XGBoost,mitigating overfitting issues,and achieving effective classification of complex samples.The results indicate that SSA-XGBoost has a high accuracy in handling fault diagnosis of power transformers and is an effective model for diagnosing faults in oil-immersed power transformers.