基于Epanechnikov核与最优窗宽组合的中期电力负荷概率密度预测方法

利用神经网络分位数回归获得预测当天在不同分位点上的电力负荷预测值,将Epanechnikov核函数与不同的最优窗宽选择方法相组合,得到中期电力负荷概率密度估计函数以及在所有分位点上连续的概率密度曲线图。此外,通过选取概率密度曲线峰值处的点预测值,比较不同窗宽组合方法。相对于传统高斯核密度估计方法的组合方式,Epanechnikov核函数的组合方式较优。最后将获得的最优方法与现有的预测方法进行对比,结果表明通过选取最优窗宽可以提高预测精度,更好地反映中期电力负荷的波动性。

Mid-term power load probability density forecast based on Epanechnikov kernel and optimal window bandwidth

The neural network quantile regression is applied to forecast the power loads at different quantiles of a day, and combined with different selection methods of optimal window bandwidth, the Epanechnikov kernel function is applied to obtain the probability density estimation function of mid-term power load and the continuous probability density curves of all quantiles. The predicted peak values of the probability density curves are picked out and compared among different selection methods. The combination forecast method of Epanechnikov kernel function is better than that of traditional Gaussian kernel density estimation method. The obtained optimal combination method is compared with some existing forecasting methods and results show that, the selection of optimal window bandwidth improves the prediction accuracy and better reflects the fluctuation of mid-term power load.