风速及风电功率超短期动态选择线性组合预测  

ULTRA-SHORT-TERM WIND SPEED AND POWER FORECAST BASED ON DYNAMIC SELECTIVE LINEAR COMBINED FORECAST

作　　者：刘彦华[1] 董泽[1] 

机构地区：[1]华北电力大学,河北省发电过程仿真与优化控制工程技术研究中心,保定071003

出　　处：《太阳能学报》2016年第4期1009-1016,共8页Acta Energiae Solaris Sinica

基　　金：中央高校基本科研业务费专项

摘　　要：提出动态选择线性组合预测方法,采用不同BP、RBF神经网络和序列最小优化SMO(sequential minimaloptimization)算法为预测模型,先用K近邻法收集预测数据的泛化误差构成性能矩阵,在此基础上动态选择泛化误差较小的预测模型,经等权平均形成最终预测输出。以风速和功率的时间序列为原始数据,实现对单台风电机组2 min内功率及风速的超短期滚动预测。研究表明:该方法的预测精度高于任意单一模型及传统线性组合预测,可有效减少预测点出现较大误差的概率,将2 min内的功率及风速的平均相对误差控制在10%内,验证了其正确性和有效性。The dynamic selective equal weight average combined forecast method was presented, at first, the generalization errors of forecast data collected by K nearest neighbor method was used to construct the performance matrix, different BP neural networks, RBF neural networks and Sequential Minimal Optimization （SMO） algorithm forecast models were used for the forecast models, based on these, the forecast models with smaller generalization errors were dynamically selected and the final forecast output was formed by weighted average. Taking the time sequence of wind speed and wind turbine power as the original data, the ultra-short-term advance forecast of wind speed and power in two minutes of single turbine was realized. The research results show that the prediction accuracy of DSLCF is higher than that of any single model and that of traditional linear combined forecasting model. It can effectively reduce the probability having larger error at forecast point, control the mean relative error of 2 minutes power and wind speed forecast as low as 10%, and verify the validity and effectiveness of the DSLCF.

关 键 词：组合预测 动态选择 超短期预测 神经网络 序列最小优化 

分 类 号：TK89[动力工程及工程热物理—流体机械及工程]