电价预测的自适应支持向量机方法研究  

Research on SA-SVM methods for price forecasting

作　　者：刘庆彪[1,2] 张步涵[1] 王凯[1] 谢光龙[1] 

机构地区：[1]华中科技大学电力安全与高效湖北省重点实验室,湖北武汉430074 [2]西宁供电公司,青海西宁810003

出　　处：《电力系统保护与控制》2008年第22期34-39,共6页Power System Protection and Control

摘　　要：在电力市场中对电价进行准确的预测无论对于发电商、电力用户还是市场运营者都具有重要的意义,该文突破了传统电价预测方法基于经验风险最小化的局限性,采用数据挖掘技术实现了数据隐含特征的提取,通过判断数据特征进行了核函数的选择,采用遗传算法实现了计算参数的自适应调整,并用相似样本和邻近样本训练支持向量机,对预测结果进行了去噪声合成。利用澳大利亚NSW电力市场的数据进行了验证,单日预测的平均百分比误差(MAPE)为5.85%,明显优于神经网络和单纯支持向量机的预测结果。扩大样本长度进行研究,一周的预测结果表明该方法不但能够有效学习样本信息、去除电价毛刺,并能有效跟踪电价的突变情况,实现了学习适度的优良泛化性预测。Price forecasting is of great importance in power market, this paper used data mining techniques to extract implicit data properties, selected kernel functions according to data properties, made use of Genetic Algorithm （GA） theory to realize Self-Adapting Support Vector Machine （SA-SVM）, and then used similar samples and adjacency samples to train SVM and synthesis the final result noise-freely. When examining this method using NSW market data in Australia, we found that one day forecasted mean absolute percentage error （MAPE） was much better than Neural Network method and pure Support Vector Machine method. When extending forecasting sample ,we found one week forecasted result indicated that SA-SVM could not only study valuable information and get rid of noise, but also trace power price peaks and get quality forecasting results of good generalization and proper study.

关 键 词：电价预测 数据挖掘 支持向量机 自适应调整 电力市场 电力行业 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] F426.61[自动化与计算机技术—控制科学与工程]