在智能电网电力营销发展的新向标下，电网企业必须要精确定位优质客户，转变原有的思维模式，科学配置服务资源，要以普通企业的视角看待市场营销。研究和分析电力用户的行为数据，精确定位用户的需求、消费习惯、行为趋势和心理变化，对国家电网等电力企业放开售电市场，改善国内用户服务质量，提高海内外市场核心竞争力具有重要意义。
    文首先对电网现有的营销业务应用系统、95598客服系统中积累的电力用户行为历史数据进行清洗转化等数据预处理工作，得到去除异常值和空值的规范数据集。在电力缴费数据用户价值分析模型缺乏的情况下，基于传统RFM消费者价值分析模型，结合电力用户缴费行为的特点，构建了CRFMO电力缴费用户综合价值评价模型，在对CRFMO特征采用最小-最大标准化处理后，基于K-means算法实现了此CRFMO特征价值模型的电力缴费用户价值分群。其次，为了优化K-means聚类算法存在的初始聚类中心随机的问题，本文结合DPC密度峰值算法，提出一种优化的KD-means聚类算法，通过计算基于加权欧氏距离的相似度矩阵，得到所有样本点的局部密度和高密度距离，进而获得簇中心选择指数，从而确定初始簇中心点。通过对比聚类评估指标，优化后的KD-means模型具有更好的聚类效果。针对电力缴费用户价值分群，分析其CRFMO各特征聚类中心值，勾画特征雷达图，根据不同用户类别的表现特征差异，为电力用户价值划分等级，进行用户价值细分和定位，总结各用户群特征及响应策略。
    最后，基于KD-means聚类算法对电力缴费行为的价值分群模型，设计并开发电力用户缴费行为数据分析原型系统，实现电力用户分群结果的可视化。通过系统数据分析和呈现，直观展示电力缴费用户的特点，为电力企业开展市场购售电侧改革、制定高效售电营销策略提供理论依据，有效帮助电力企业实现精准营销策略。

    Under the new direction of the development of smart grid power marketing, grid companies must accurately locate high-quality customers, change the original thinking mode, scientifically allocate service resources, and view marketing from the perspective of ordinary enterprises. Research and analyze the behavior data of power users, accurately locate the needs, consumption habits, behavior trends and psychological changes of users, it is of great significance for the national grid and other power companies to open up the electricity sales market, improve the service quality of domestic users, and improve the core competitiveness of domestic and overseas markets.
    Firstly, perform pre-processing work such as cleaning and transformation of historical power user behavior data accumulated in the existing marketing business application system of the power grid and 95598 customer service system to obtain a standardized data set for removing outliers and null values, and analyze the user value model of power payment data in the absence of circumstances, based on the traditional RFM consumer value analysis model, combined with the characteristics of power user payment behavior, a CRFMO power payment user comprehensive value evaluation model was constructed, and the minimum-maximum standardization of CRFMO characteristics was adopted. Based on this CRFMO characteristic value model, K-means user value grouping of the algorithm. Secondly, in order to optimize the initial clustering center randomness problem of the K-means clustering algorithm, combines the DPC density peak algorithm and proposes an optimized KD-means clustering algorithm. By calculating the similarity matrix based on the weighted Euclidean distance, we obtain the local density and high-density distance of all sample points are used to obtain the cluster center selection index, so as to determine the initial cluster center point. By comparing the clustering evaluation indexes, the optimized KD-means model has better clustering effect. For the user group obtained by clustering, analyze the clustering center value of each CRFMO feature, draw a feature radar chart, classify the value of power users according to the difference in performance characteristics of different user categories, subdivide and position the user value, and summarize each user group features and response strategies.
    Finally, based on the KD-means clustering algorithm for the value grouping model of power payment behavior, a prototype system for data analysis of power user payment behavior is designed and developed to realize the visualization of power user grouping results. Through systematic data analysis and presentation, the characteristics of power payment users are displayed intuitively, which provides a theoretical basis for power companies to carry out market purchase and sale side reforms and formulate efficient power sales marketing strategies, effectively helping power companies achieve accurate marketing strategies.