基于聚类和时间序列分析的变压器状态评价方法

传统的电力变压器DGA故障诊断方法,仅能二值化地判断设备处于健康或故障状态,无法表征变压器的潜在故障情况,也无法确定变压器向故障状态转化的趋势。对此,提出了一种基于聚类和时间序列分析的变压器状态评价方法。首先,基于点密度判据进行数据预处理,消除噪声影响。其次,基于大数据聚类思想,计算采样数据和历史故障数据簇的相对邻近度,根据计算结果将设备状态划分为健康、潜伏故障或故障。在此基础上判断故障设备的故障类型,基于故障类型关联权重计算健康设备的健康得分,通过时间序列相似性分析方法获取潜伏故障设备的预测故障发展时间。算例分析验证了该方法的可行性与有效性。

Evaluation method of transformer state based on clustering and time series analysis

Conventional DGA fault diagnosis methods of power transformer can only judge whether the equipment is in normal or fault condition, but they can neither characterize the potential failure of the normal transformer nor identify the trend of transformer converted into fault state. In order to solve this problem, this paper proposes a transformer state evaluation method based on clustering and time series analysis method. Firstly, the data preprocessing method is carried out to prevent the influence of noise. Then, based on big data clustering method, the proximity of sampled data and historical fault data clusters is calculated. Based on the result, the equipment condition is sorted into healthy, incipient faulty or faulty. For equipment in faulty state, the fault type is identified. For healthy equipment, the health score is calculated based on related weights of fault types. For incipient faulty equipment, the time series similarity analysis method is used to further predict the time span for the equipment changing from current state to fault state. The example analysis verifies the feasibility and effectiveness of the proposed method. This work is supported by Key Program of National Natural Science Foundation of China (No. 51537010) and Science and Technology Project of State Grid Corporation of China (No. 52182016001J).