首页 | 本学科首页   官方微博 | 高级检索  
     

基于变量选择与高斯过程回归的短期负荷预测
引用本文:梁智,孙国强,卫志农,臧海祥.基于变量选择与高斯过程回归的短期负荷预测[J].电力建设,2017,38(2).
作者姓名:梁智  孙国强  卫志农  臧海祥
作者单位:河海大学能源与电气学院,南京市,210098
摘    要:提高短期电力负荷预测精度是保障电网安全稳定运行的技术措施之一,通过选取影响负荷的最优输入变量集合,建立高斯过程回归(Gaussian process regression,GPR)短期负荷预测模型。负荷预测建模输入变量的选取对预测精度有很大影响,首先采用随机森林(random forest,RF)算法给出输入变量重要性评分(variable importance measure,VIM),并对各输入变量影响程度进行排序,基于序列前向搜索策略确定最优输入变量集合,避免人工经验选取的不足。其次针对共轭梯度(conjugate gradient,CG)法求解高斯过程回归模型超参数时易陷入局部最优解,且存在优化性能依赖于初值选取、迭代次数难以确定的问题,采用改进粒子群优化(particle swarm optimization,PSO)算法搜索模型超参数,形成优化高斯过程回归预测模型。最后,算例测试表明该模型的有效性。

关 键 词:短期负荷预测  输入变量选择  随机森林(RF)算法  高斯过程回归(GPR)  改进粒子群优化(PSO)算法

Short-Term Load Forecasting Based on Variable Selection and Gaussian Process Regression
LIANG Zhi,SUN Guoqiang,WEI Zhinong,ZANG Haixiang.Short-Term Load Forecasting Based on Variable Selection and Gaussian Process Regression[J].Electric Power Construction,2017,38(2).
Authors:LIANG Zhi  SUN Guoqiang  WEI Zhinong  ZANG Haixiang
Abstract:Improving the short-term load forecasting accuracy is one of the technical measures to ensure the safe and stable operation of the power grid.This paper establishes the short-term load forecasting model based on Gaussian process regression (GPR) by selecting the optimal set of input variables influencing the load consumption.The selection of input variables in load forecasting modeling has great influence on the forecasting accuracy.Firstly, we adopt random forest (RF) algorithm to obtain each input variable importance measure (VIM) and rank the input variables according to the influence degree.Based on the forward search strategy, the optimal input variable set can be determined, which can avoid the shortage of artificial experience.The conjugate gradient (CG) method is easy to fall into local optimal solution when solving the hyperparameters of Gaussian process regression (GPR) model, and the optimal performance depends on the selection of initial value, also the iteration number is difficult to be determined.In view of these problems, we use the improved particle swarm optimization (PSO) algorithm to search the model hyperparameters, and develop the load forecasting model based on optimal GPR.Finally, the effectiveness of proposed model is illustrated through testing simulation.
Keywords:short-term load forecasting  input variables selection  random forest (RF) algorithm  Gaussian process regression (GPR)  improved particle swarm optimization (PSO) algorithm
本文献已被 CNKI 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号