配电台区时间序列大数据负荷预测技术研究

发布时间：2018-03-19 16:51

本文选题：台区负荷预测　切入点：聚类　出处：《华北电力大学(北京)》2017年硕士论文　论文类型：学位论文

【摘要】：随着电力用电信息采集系统的优化升级,电力用户侧除了能够采集到传统的电能使用信息外,也能获得用户电能质量、96点负荷曲线、用电行为偏好等多维度反映用户特性的数据。配电台区负荷预测作为传统负荷预测领域的细分,是进行精细化的台区用电管理、运行调度与网架结构优化的新兴手段。受制于台区负荷的随机性与多样性,传统预测方法在台区级的预测中表现欠佳,而借助大数据平台这一新兴工具,对用户侧采集到的冗杂且大量的用户特性信息加以利用,能够有效提高台区负荷预测的预测精度与预测适应性。本文分析台区负荷数据的特点及其关联因素,借助大数据平台中K-means、BIRCH、WARD等聚类算法,对上海市2977个台区进行聚类划分,结合实际台区行业构成分析了各聚类簇的特征及划分的有效性。聚类结果表明台区内用户对气象、经济等因素均有不同的敏感度,以聚类结果为依据,对不同用电特性的用户类分别建立预测模型是提高台区负荷预测精度的有效手段。提出基于机器学习中的岭回归算法与自适应思想的自适应岭回归预测模型,依据自适应程度将模型划分为3种模式,对台区负荷建模并实际训练、预测。3种模式在训练用时、预测精度、敏感度方面表现出不同特性,适用不同预测环境。在聚类与回归模型构建的基础上,进一步提出基于聚类及自适应岭回归技术的台区负荷预测方法,设计3种聚类特征选取方式与5种聚类算法作为预测自适应优化的可调模块,增强了预测的动态优化与误差调控能力。使用该方法对上海市某包含487户用户的台区进行预测,算例结果显示在不同预测环境中通过优选聚类特征、聚类算法及模型参数自适应调节,该预测方法能达到较高的预测精度与环境适应能力。
[Abstract]:With the optimization and upgrading of electric power information acquisition system, the power user side can not only collect the traditional power usage information, but also obtain the 96 point load curve of power quality. As a subdivision of traditional load forecasting field, distribution station area load forecasting is a fine management of station area power consumption. Due to the randomness and diversity of the load in Taiwan area, the traditional forecasting method is not good in the prediction of Taiwan district level, but with the help of big data platform, which is a new tool, It can effectively improve the forecasting accuracy and adaptability of station load forecasting by using the miscellaneous and large amount of user characteristic information collected by user side. This paper analyzes the characteristics of station load data and its related factors. With the help of K-means-BIRCHWARD and other clustering algorithms in big data platform, the cluster classification of 2977 stations in Shanghai is carried out, and the characteristics of each cluster and the validity of the division are analyzed in combination with the industry composition of the actual station. The clustering results show that the users in the station are sensitive to meteorology. Economic and other factors have different sensitivity, based on clustering results, It is an effective method to improve the accuracy of load forecasting for different users with different power consumption characteristics. An adaptive ridge regression prediction model based on machine learning and adaptive thinking is proposed. According to the adaptive degree, the model is divided into three models, and the load in the station area is modeled and trained in practice. The prediction of the three models shows different characteristics in training time, prediction accuracy and sensitivity. Based on the construction of clustering and regression models, a load forecasting method based on clustering and adaptive ridge regression is proposed. Three kinds of clustering feature selection methods and five clustering algorithms are designed as adjustable modules for adaptive predictive optimization, which enhances the dynamic optimization and error control ability of prediction. The proposed method is used to predict a certain station area in Shanghai, which includes 487 users. The simulation results show that the prediction method can achieve high prediction accuracy and environmental adaptability by optimizing clustering features, clustering algorithm and adaptive adjustment of model parameters in different prediction environments.
【学位授予单位】：华北电力大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM715

【参考文献】