基于近邻传播聚类和回声状态网络的光伏预测  

PV power forecasting based on AP-ESN

作　　者：李乐[1] 刘天琪[1] 

机构地区：[1]四川大学电气信息学院,四川成都610065

出　　处：《电力自动化设备》2016年第7期41-46,共6页Electric Power Automation Equipment

基　　金：四川省科技厅科技计划项目(2016GZ0143)~~

摘　　要：分析了天气因素对光伏出力的影响,选择预报技术较为成熟的气象信息和能反映光伏变化趋势的波动分量作为分类特征。针对传统预测算法在突变天气条件下预测精度较低的问题,提出基于近邻传播聚类和回声状态网络的组合算法,通过近邻传播算法对光伏出力进行分类,并根据预测日所属的类别建立回声网络状态方程进行光伏出力预测。仿真表明所提算法不仅能满足非突变型天气下的光伏预测要求,还能较好地跟踪突变型天气下光伏出力变化,具有更高的准确度和通用性。The influence of weather on the PV（PhotoVoltaic） output power is analyzed. The weather information by the mature forecasting technologies and the PV fluctuant components reflecting the variation tendency of PV output power are selected as the classification properties. As the accuracy of traditional forecasting algorithm is low when weather changes suddenly,an algorithm based on AP（Affinity Propagation） and ESN（Echo State Network） is proposed,which applies AP to classify PV output power and sets the ESN according to its class of the day concerned for forecasting its PV output power. Simulation shows that,with higher accuracy and universality,the proposed algorithm meets the requirements of PV output power forecasting for normal weathers and traces quite well the variation of PV output power for abnormal weathers.

关 键 词：光伏发电 预测 波动分量 近邻传播 回声状态网络 聚类算法 

分 类 号：TM615[电气工程—电力系统及自动化]