Day-ahead probabilistic forecasting for French half-hourly electricity loads and quantiles for curve-to-curve regression
Day-ahead probabilistic forecasting for French half-hourly electricity loads and quantiles for curve-to-curve regression
复制标题
DOI：
10.1016/j.apenergy.2021.117465
复制
发表时间：
2021-11
期刊：
影响因子：
11.2
通讯作者：
Xiuqin Xu;Ying Chen;Y. Goude;Q. Yao
中科院分区：
文献类型：
--
作者：
Xiuqin Xu;Ying Chen;Y. Goude;Q. Yao
电力负荷的概率预测是在波动和竞争激烈的能源市场中进行有效调度和决策的关键。我们提出了一种新的方法来构建概率预测曲线（PPC）的电力负荷，这导致正确定义的预测带和分位数的曲线到曲线回归的背景下。建议的预测模型不仅提供准确的每小时负荷点预测，而且还生成定义良好的概率带和全天的轨迹的负载在任何概率水平，预先指定的管理人员。我们还定义了预测分位数曲线，展示了极端情况下的未来负荷，并为电力供应管理中的对冲风险提供了见解。当应用于法国半小时电力负荷的前一天预测时，PPC优于几种最先进的时间序列和机器学习预测方法，具有更准确的点预测（平均绝对百分比误差为1.10%，而替代方案为1.36%-4.88%），全天负荷轨迹的覆盖率更高（覆盖率为95.5%，而替代品为31.9%-90.7%）和预测带的平均长度较窄。在一系列的数值实验中，PPC进一步证明了强大的性能和普遍适用性，实现了准确的覆盖概率下的各种数据生成机制。
The probabilistic forecasting of electricity loads is crucial for effective scheduling and decision-making in volatile and competitive energy markets with ever-growing uncertainties. We propose a novel approach to construct the probabilistic predictors for curves (PPC) of electricity loads, which leads to properly defined predictive bands and quantiles in the context of curve-to-curve regression. The proposed predictive model provides not only accurate hourly load point forecasts, but also generates well-defined probabilistic bands and day-long trajectories of the loads at any probability level, pre-specified by managers. We also define the predictive quantile curves that exhibit future loads in extreme scenarios and provide insights for hedging risks in the supply management of electricity. When applied to the day-ahead forecasting for French half-hourly electricity loads, the PPC outperform several state-of-the-art time series and machine learning predictive methods with more accurate point forecasts (mean absolute percentage error of 1.10%, compared to 1.36%–4.88% for the alternatives), a higher coverage rate of the day-long trajectory of loads (coverage rate of 95.5%, against 31.9%–90.7% for the alternatives) and a narrower average length of the predictive bands. In a series of numerical experiments, the PPC further demonstrate robust performance and general applicability, achieving accurate coverage probabilities under a variety of data-generating mechanisms.