中文摘要：

提出了一种基于稀疏贝叶斯分类与Dempster-Shafer（D-S）证据理论的短期风电功率概率分布非参数估计方法,预测时间尺度为48 h。该方法首先通过支持向量机（support vector machine,SVM）对风电功率进行点预测;进而将SVM预测误差的范围离散为多个区间,通过建立稀疏贝叶斯分类器对SVM预测误差落入各预定区间的概率进行估计。然后应用D-S证据理论对所有区间对应的概率估计结果进行整合,得到SVM预测误差的整体概率分布。最后叠加误差分布与SVM预测的风电功率值,得到风电功率的概率分布结果。该方法基于稀疏贝叶斯架构构建,具有高稀疏性,确保了模型的泛化能力与计算速度。该方法还系统地计及了风电场输出功率必须满足在[0,GN]（GN为风电场装机容量）内取值的边界约束,使预测结果更加符合实际。以某74 MW的风电场为例对上述方法进行了验证,结果表明了该方法的有效性。

英文摘要：

This paper proposes a nonparametric approach for probabilistic wind generation forecast based on sparse Bayesian classification（SBC） and Dempster-Shafer（D-S） theory. Forecast time horizon is 48 hours. Firstly, the approach makes a spot forecast of wind generation based on Support Vector Machine（SVM）. Then, SVM forecast error range is discretized into multiple intervals, and conditional probability of each pre-designed interval is estimated by building a sparse Bayesian classifier. Thirdly, D-S theory is applied to combine probabilities of all intervals to form a unified probability distribution function（PDF） of SVM forecast error. Finally, forecast result is obtained by superposition of SVM forecast result over mean value of forecasted error. The approach built on sparse Bayesian framework has high sparseness, ensuring its generalization ability and computation speed. Boundary constraint that wind generation should be within [0,GN] with installed capacity GN of wind farms, is taken into account, making forecast results well in line with actual results. Tests on a 74 MW wind farm illustrate effectiveness of the approach.