摘要
在常规的电力系统负荷预测过程中,所用的预测方法不能快速、有效地训练大量数据样本,并且不能有效地运用历史数据。为了解决这些问题,设计了一种基于大数据平台的电力系统负荷预测模型。运用Xgboost算法搭建训练样本模型生成决策树,为了防止在该过程中出现过拟合,采用了Gradient Boosting思想和Shrinkage思想;搭建Hadoop平台部署Xgboost算法。通过对某省M县实际负荷数据特性的分析,构建了基于负荷的时间特性、温度特性的训练样本,分别进行了夏季、冬季情况下的负荷预测,同时与随机森林(RF)算法和梯度提升决策树(GBDT)算法进行对比。预测试验对比验证了Xgboost算法具有准确性好、训练速度快等特点,且在开启多线程的情况下,Xgboost算法有更明显的提升。
In the conventional power system load forecasting process,the prediction methods used cannot train a large number of data samples quickly and effectively,either cannot effectively utilize the historical data. In order to solve these problems,a power system load forecasting model based on big data platform has been designed. The training sample model is built by using Xgboost algorithm to generate a decision tree. In order to prevent over fitting in the process,two ideas： Gradient Boosting and Shrinkage are adopted,and Hadoop platform is built to deploy Xgboost algorithm. Through analyzing the actual load data characteristics of M county of a certain province,the training samples of time characteristics and temperature characteristics based on load are constructed and the load forecast in summer and winter are respectively carried out,as well as compared with random forest（ RF）algorithm and gradient enhancement decision tree（ GBDT） algorithm. The comparison verifies that the Xgboost algorithm has the advantages of good accuracy and fast training speed,and the Xgboost algorithm has more obvious improvement when the multi-thread is started.
作者
夏博
杨超
郑凯文
XIA Bo;YANG Chao;ZHENG Kaiwen(College of Electrical Engineering, Guizhou University, Guiyang 550025, Chin)
出处
《自动化仪表》
CAS
2018年第6期81-84,共4页
Process Automation Instrumentation
