基于机器学习的变压器预测性维护方案

Machine Learning Based Predictive Maintenance Solution for Transformers

摘要: 为了改善现有的配电变压器维护方案，更好地实现电力大数据的应用，本文提出了一种基于机器学习的变压器预测性维护方案，以变压器油中溶解气体数据特征数据，首先对变压器原始采集数据进行处理，然后使用隐半马尔可夫模型(HSMM)确定变压器的运行状态，进一步使用改进卷积神经网络对异常数据进行分类预测，最终实现辅助变电运维人员进行非周期定向维护，从而有效降低运维成本、减少机组停机时间，提高变压器的任务完成率的需要。

Abstract: In order to improve the existing maintenance scheme for distribution transformers and better realise the application of big data in electricity, this paper proposes a machine learning-based predictive maintenance scheme for transformers, using data characterised by dissolved gas data in transformer oil, firstly processing the original transformer collection data, then using a hidden semi-Markov model (HSMM) to determine the operational status of the transformer, and further using an improved Convolutional neural networks are used to classify and predict abnormal data, ultimately realising the need to assist substation operators and maintenance personnel to carry out off-cycle targeted maintenance, thereby effectively reducing operation and maintenance costs, reducing unit downtime and improving the task completion rate of transformers.