基于隐马尔可夫模型的油浸电力变压器故障预测

Fault Prediction of Oil Immersed Power Transformer Based on Hidden Markov Model

为分析变压器在运行过程中状态的变化并准确地预测变压器的故障,本文提出一种基于隐马尔可夫模型的变压器故障预测技术,并针对隐马尔可夫模型在模型训练时参数易陷入局部最优化的情况,利用遗传算法(Genetic Algorithm,GA)的优点对其中的Baum-Welch算法和初始参数进行优化,形成精准度更高的模型,并验证了优化后效果。另外,将高斯混合模型作为聚类方法,引入“亚健康”概念来丰富变压器的状态类型,从变压器的溶解气体分析(Dissolved Gas Analysis,DGA)数据集中提取不同健康状态特征,建立溶解气体与状态类型的特征联系。建立优化后的隐马尔可夫模型,从不同的健康状态评估各个状态的转移概率矩阵。在故障预测方面,综合DGA历史运行状态,利于隐半马尔可夫模型预测,所得的结果与实际历史情况相比较,准确度较好。

In order to analyze the change of transformer state during operation and accurately predict transformer fault, a transformer fault prediction technology based on Hidden Markov model is proposed in this paper, Using the advantages of GA, Baum Welch algorithm and initial parameters are optimized to form a model with higher accuracy, and the optimized effect is verified. In addition, the Gaussian mixture model is used as a clustering method, and the concept of “Baum-Welch” is introduced to enrich the state types of transformers. The characteristics of different health states are extracted from the DGA data set of transformers, and the characteristic relationship between dissolved gas and state types is established. The optimized hidden Markov model is established to evaluate the transition probability matrix of each state from different health states. In terms of fault prediction, integrating the historical operation state of DGA is conducive to the prediction of hidden semi Markov model. The results obtained are more accurate than the actual historical situation.