Abstract: Mobile power trading information service promotes the business scale of power generation enterprises, sale companies and purchasing users of electricity, it forms multilateral trade member modes in the power market and also realizes multi-category complementary trading between supply and demand. The transaction behaviors of mobile power market members are analyzed, and the KNN algorithm based on group active learning strategy is presented. It is effective to build the training set by the group active learning strategy. Firstly, the unlabeled samples are randomly selected by group to construct a candidate set. Secondly, the individual deviation values for distance cumulative means are computed in unlabeled grouping samples. And then the candidate samples sets are filtered by satisfying the support degree values and added to the training sample set. Finally, the KNN classification algorithm based on the group active learning strategy is proposed as an implementation step description. The case study of the mobile power trading user behavior data is implemented by the proposed methodology, and the person coefficients are computed by characteristics elements with customer satisfaction, region, time, power market clearing price, to classify the most similar power purchasers. Experimental results show that the time and accuracy of group active learning KNN algorithm meet the expected requirements. The proposed active learning algorithm is more effective, and is applicable to analysis and decision on the mobile power trading market.

Key words: Active learning, K-nearest neighbor, classification algorithm, power trading, mobile terminal

摘要： 移动端电力交易信息服务提升发电企业、售电公司、购电用户的业务规模，市场成员多边交易，实现多品类交易供需互补。分析移动端电力市场成员的交易行为，提出基于群体主动学习的KNN算法。群体主动学习策略有效构造训练集，首先随机分组选择未标记样本构成候选集，其次计算未标记分组样本的个体距离累加平均值的偏差，接着筛选满足偏差支持度的候选集，加入训练集中，最后给出相应的算法步骤。结合移动端电力市场交易数据进行算例分析，计算电力用户满意度、地域、时间、成交电价综合特征的皮尔逊相关系数，分类出相似购电用户。多种算法实验进行对比和性能分析，结果表明：群体主动学习KNN算法的时间和精确度达到预期要求，具有较好的分类效果，适用于移动端电力市场交易行为分析和供需决策。

关键词: 主动学习, K-最近邻, 分类算法, 电力交易, 移动端