新能源进展
新能源進展
신능원진전
Advances in New and Renewable Enengy
2013年
3期
224-229
,共6页
黄磊%舒杰%崔琼%姜桂秀
黃磊%舒傑%崔瓊%薑桂秀
황뢰%서걸%최경%강계수
风功率预测%风功率新息序列%ARMAX%GARCH%区间预测
風功率預測%風功率新息序列%ARMAX%GARCH%區間預測
풍공솔예측%풍공솔신식서렬%ARMAX%GARCH%구간예측
wind power forecasting%wind power innovation series%ARMAX%GARCH%interval forecasting
目前风功率预测多为风功率期望的点预测,且以采样间隑较大的功率序列作为建模序列,这样会降低预测模型对风功率时序特征模拟的准确度和可信度。文中基于小采样间隑风功率序列,提出 ARMAX-GARCH 风功率预测模型。通过构造风功率新息序列,结合小时平均风功率序列,建立ARMAX点预测模型,采用BIC最小信息准则和相关性分析实现模型定阶和外生变量选择;采用GARCH模型模拟残差的波动特性实现区间预测。以海岛微电网实测风功率数据为例,迚行提前1 h风功率预测。结果表明,与持续法、ARMA和RBF神经网络相比,该预测模型能显著提高风功率期望的点预测精度幵具有较好的区间预测效果。
目前風功率預測多為風功率期望的點預測,且以採樣間隑較大的功率序列作為建模序列,這樣會降低預測模型對風功率時序特徵模擬的準確度和可信度。文中基于小採樣間隑風功率序列,提齣 ARMAX-GARCH 風功率預測模型。通過構造風功率新息序列,結閤小時平均風功率序列,建立ARMAX點預測模型,採用BIC最小信息準則和相關性分析實現模型定階和外生變量選擇;採用GARCH模型模擬殘差的波動特性實現區間預測。以海島微電網實測風功率數據為例,迚行提前1 h風功率預測。結果錶明,與持續法、ARMA和RBF神經網絡相比,該預測模型能顯著提高風功率期望的點預測精度幵具有較好的區間預測效果。
목전풍공솔예측다위풍공솔기망적점예측,차이채양간기교대적공솔서렬작위건모서렬,저양회강저예측모형대풍공솔시서특정모의적준학도화가신도。문중기우소채양간기풍공솔서렬,제출 ARMAX-GARCH 풍공솔예측모형。통과구조풍공솔신식서렬,결합소시평균풍공솔서렬,건립ARMAX점예측모형,채용BIC최소신식준칙화상관성분석실현모형정계화외생변량선택;채용GARCH모형모의잔차적파동특성실현구간예측。이해도미전망실측풍공솔수거위례,중행제전1 h풍공솔예측。결과표명,여지속법、ARMA화RBF신경망락상비,해예측모형능현저제고풍공솔기망적점예측정도견구유교호적구간예측효과。
Wind power forecasting models are often built for point forecasting using wind power series with large sampling intervals, which reduces the accuracy and reliability of the forecasting models. Based on wind power series with small sampling interval, this paper proposes an ARMAX-GARCH wind power forecasting model. The ARMAX model is built for point forecasting by combining the constructed innovation series of wind power and the hourly-average wind power series. The model identification and exogenous covariates selection are based on Bayesian information criterion (BIC) and correlation analysis. The GARCH model is used for interval forecasting to simulate the fluctuation characteristic of the residual series. To demonstrate the effectiveness, the model for 1 h ahead forecasting is applied and tested on a microgrid located on an island in the south of China. Comparing with persistent method, ARMA and RBF neural network, simulation results demonstrate that the proposed forecasting model improves the accuracy of point forecasting significantly and has a better interval forecasting result.