贝叶斯多元线性样条在电力负荷中期预测中的应用

Application of Mid-Term Load Forecasting Based on BMLS

贝叶斯多元线性样条BMLS是基于贝叶斯框架下的分段线性回归技术,实现了回归面边界光滑性。本文分析了BMLS方法的原理,结合EUNITE网络2001年举办的电力负荷预测比赛提供的数据进行了相关数据分析,建立了相应的电力负荷中期预测模型。我们使用BMLS方法对两种训练样本集进行了训练,并计算出预测期的预测值,取得了理想的预测结果,并结合其它方法对试验结果进行了分析。文章最后总结了BMLS方法用于预测的特点,并与其它方法,如相关向量机等进行了比较。使用BMLS的模拟近似平均技术进行预测可以实现较好的精度。

Bayesian Multivariate Linear Splines(BMLS) is a piecewise linear regression using Bayesian methods to achieve boundary smoothness between pieces. After discussing privary technologies used for BMLS, it is proposed to be used for mid-term load forecasting in this paper. After analyzing data from EUNITE-network-sponsoring competition, we did experiments choosing different train samples, getting good results. We compare experimental results with other methods and present some explanations. In the end of the paper we do summaries on BMLS advantages and disadvantages, comparing with other methods, such as the Relevance Vector Machines, etc. Predications with analogy approximation averaging used in BMLS, can achieve better performances.