风电场超短期功率预测及不确定性分析

UL TRA-SHORT TERM WIND POWER PREDICTION AND UNCERTAINTY ASSESSMENT

研究了在不使用数值气象预报的条件下,基于历史数据序列,采用BP神经网络对风电场未来3h功率进行预测。结合多次的计算试验,合理确定了输入神经元参数;提出两种风电场功率预测路线:一种是首先预测每台机组功率,再累加计算风电场功率;另一种是直接计算整个风电场功率。结果表明,第一种预测路线更适合我国风电场集中分布的情况,相对预测误差为9.6%。在此基础上,建立了基于独立分量分析的条件概率计算模型,对预测结果的不确定性进行了分析。

Wind power prediction is of great importance for the safety and stabilization of grids. Based on historical data, three hour＇s wind power prediction was studied using BP neural network without numerical weather prediction. The input parameters were chosen after many times of calculation. Two kinds of prediction routes were put forward： One is to predict each wind turbine＇ s power firstly, and then accumulate to get the wind farm power. The other is to predict the wind farm power directly. The result show that the first route has more accurate prediction re- sult with relative error of 9.6%. A conditional probability model based on independent component analysis was built for the uncertainty assessment of prediction results.