基于时间序列编码和相关向量机的负荷短期预测  

Short-Term Load Prediction Based on Relevance Vector Machine with Time Sequence Encoding

作　　者：刘英华[1] 

机构地区：[1]华中农业大学楚天学院,湖北武汉430205

出　　处：《内蒙古师范大学学报（自然科学汉文版）》2015年第3期311-315,共5页Journal of Inner Mongolia Normal University(Natural Science Edition)

基　　金：湖北省高校省级教学研究项目(2012458);华中农业大学校级科研项目(K201405)

摘　　要：针对电力负荷短期预测精度不高的问题,提出一种基于时间序列编码和相关向量机的电力负荷短期预测方法.通过收集电力负荷实际数据,研究了日最大负荷数据之间的关系、日最大负荷与节假日的关系以及当日与对应星期数的相关性,并建立了相应的电力负荷短期预测模型.该模型采用相似日选择方法,给工作日和节假日赋予不同的权重,从电力负荷时间序列中筛选出与预测日特征相似的数据,对模型进行训练.与BP神经网络和支持向量机相比,该预测方法有更高的预测精度和更好的泛化能力,而且学习速度更快.Because short-term power load forecasting accuracy is not high,a short-term electric power load forecasting model is proposed based on relevance vector machine and time series code.Firstly,the power load data is collected,and the relationship between the daily maximum load,and relationship between the maximum daily load and the holidays and the correlation between day and corresponding weeks,and the corresponding power load short term forecasting model is established.The model different weights of working days and holidays are obtained by using similar days select method,and similar data are screened from load time series for prediction day to train the model.And compared with the BP neural network and SVM,the proposed method has higher prediction accuracy and better generalization ability,and learns faster.

关 键 词：时间序列编码 相似日 电力负荷 短期预测 相关向量机 

分 类 号：TN27[电子电信—物理电子学]