摘要
分布式能源系统负荷预测是系统规划与经济运行的可靠前提和依据,在当前海量高维数据的背景下,有效的在线数据处理平台与精确的负荷预测方法是当前的研究重点。基于分布式能源系统负荷数据特点,在缺失数据处理、坏数据分类以及特征选择的基础上,建立了基于Spark平台与多变量L_2-Boosting回归模型的分布式能源系统短期负荷预测方法。首先,利用Spark平台分割全部数据得到多个子数据模型,通过并行计算提高数据处理效率,采用特征提取方法得出模型需要的输入向量;其次,将得出的有效数据信息输入到多变量L_2-Boosting回归模型进行训练学习,得到训练后的多变量L_2-Boosting回归模型;最后,利用测试数据测试模型。算例结果验证了所提模型的有效性。
Load forecasting for distributed energy system is precondition and basis of system planning and economic operation. Under background of current massive high-dimension data, effective online data processing platform and accurate load forecasting methods are current research focus. In this paper, considering characteristics of load data of distributed energy system, short-term load forecasting method for distributed energy system based on Spark platform and multi-variable L2-boosting regression model was established after missing data processing, bad data classification and feature selection. Firstly, Spark platform was used to divide up all data to get multiple sub-data models by parallel computing to improve data processing efficiency and using feature extraction method to obtain input vector required by the model. Secondly, effective data information obtained was input to multivariable L2-boosting regression model to train the model and obtain trained multivariable L2-boosting regression model. Finally, test data was used to test the model. Results verified validity of the proposed model.
出处
《电网技术》
EI
CSCD
北大核心
2016年第6期1642-1649,共8页
Power System Technology
基金
国家自然科学基金项目(71471059)~~
