基于联合分布适配的智能电表运行状态评价

针对智能电表运行状态评价中经常存在带标签的数据采集困难且不同地区的数据分布不一致的问题，在智能电表的运行状态评价领域引入迁移学习中的联合分布适配（JDA）算法。该算法寻找一个最优化的变换矩阵，使得在变换后的空间中不同地区数据的边缘分布和条件分布距离最小化。针对条件分布适配中目标域没有数据标签的问题，采用伪标签迭代的方法使得目标域伪标签不断接近真实标签。变换后空间数据中训练得到的分类模型可以运用于新的地区，实现迁移。实验结果证明了JDA算法在智能电表运行状态评价中的有效性。 

Operating status evaluation of smart electricity meters based on joint distribution adaption

In the operation status evaluation of smart electricity meters, it is often difficult to collect labeled data and the data distribution in different regions is inconsistent. To solve this problem, we introduce the joint distribution adaption (JDA) method in transfer learning in the field of operation status evaluation of smart electricity meters. This method tries to find an optimized transformation matrix to minimize the distance between edge distribution and conditional distribution in different regions in the transformed space. In order to solve the problem that there is no data label in the target domain when the conditional distribution adaptation is made, the JDA method uses pseudo-tag iteration method to make the target domain pseudo-label constantly approach the real tag. The classification model trained from the data in the post-transformation space can be applied to the new region to realize the transfer. The experimental results demonstrate the effectiveness of JDA method in the operation status evaluation of smart electricity meters.