基于区间样本和回声状态网络的风电功率不确定性预测

Forecast of wind power generation with uncertainty based on interval sample and echo state network

风电功率预测对并网运行的稳定性控制、市场经济调度等具有重要意义.但受风力波动性等影响,风电功率具有极大的不确定性,如何在功率预测中有效反映该不确定性对提高预测可靠性至关重要.针对当前大多采用点预测方法存在的不足,提出一种量化不确定性的区间预测模型.基于区间相似准则和相似日理论,首先给出反映风电功率不确定性的区间样本选择策略;针对选择的时序区间样本,给出基于回声状态网络的区间预测方法;最后利用区间覆盖率、区间平均宽度等指标评价预测结果.实验结果表明了所提方法的有效性.

The wind power forecasting was essential to the stability control of the grid connected operation,the economical dispatch,and so on. However,due to the variety of nature of wind,wind power had great uncertainties. Effectively expressing the uncertainties in wind power forecasting is crucial for improving the reliability of the forecast. Most existing methods focued on point forecasting,which can hardly quantify the uncertainties. To overcome the weekness,this paper proposed a novel interval-based forecasting model to quantify the uncertainties. A new interval sample selection method was firstly presented to reflect the uncertainties of wind power based on similar days and interval similar metric. Secondly,the echo state network were designed to predict the interval-based wind power in a short time due to its merits in time series predictions. The outstanding stability of the forecasting model was guaranteed by employing the recursive least squares algorithm to adjust the output weights of the echo state network. The prediction interval coverage probability（PICP） and mean prediction interval width（MPIW） were applied to evaluate the performance of our interval forecast on wind power. The experiments empirically demonstrated the feasibility and effectiveness of the proposed algorithm.