考虑配电设备多源监测数据缺失的深度森林状态评价方法

Condition Evaluation of Power Distribution Equipment based on Deep Forest Method Considering Missing Values in Multi-source Monitoring Data

配电设备在线监测技术渐进式的发展过程中形成了大量含缺失项的历史数据,为改善基于机器学习的配电设备状态评价方法在处理缺失数据方面的效果,本文提出基于概率权重深度森林的配电设备状态评价方法。首先对数据缺失产生的原因进行说明,接下来使用概率权重处理含缺失项样本并构建决策树,最后在此基础上构造随机森林和深度森林以实现基于数据驱动的配电设备状态评价。算例分析证明了本文方法在处理高比例有缺失数据方面的优越性。

With the gradual development of the online monitoring technology of power distribution equipment,a large number of historical samples with missing items were naturally produced.To improve the effect of the condition evaluation method of the power distribution equipment based on machine learning,this paper proposed an approach based on deep forest algorithm with probability weights.Firstly,the reason for the data loss was explained.Then,a method to build the decision tree upon samples with missing items using probability weights was introduced.Finally,the random forest and deep forest are constructed to realize the data-driven condition evaluation of power distribution equipment.Example analysis shows the superiority of those methods proposed in this paper in dealing with high proportion of missing data.