基于VaR和集成神经网络分位数回归的短期负荷概率预测

Short-term load probability forecasting based on VaR and integrated neural network quantile regression

中文关键词: 负荷概率预测  神经网络  注意力机制  峰值指示特征  分位数回归

英文关键词: load probabilistic prediction  neural network  attention mechanism  peak indicator feature  quantile regression

中文摘要:

      短期负荷预测在电力系统规划与运行中起着重要作用。提出一种融合注意力机制和分位数回归的混合卷积双向长短期神经网络短期负荷概率预测模型。首先,利用相关性分析选取合适的天气变量和历史负荷。其次,通过Copula模型计算出风险阈值,该值被用于构造峰值二进制指示输入特征。接着,将所选特征集输入到卷积双向长短期神经网络预测模型,引入注意力机制给予数据不同关注。然后,采用核密度估计对负荷进行概率预测。最后,使用平均绝对百分比误差和均方根误差对模型预测性能进行评估。仿真结果表明,该模型具有更高的预测精度。

英文摘要:

      Short-term load forecasting plays an important role in power system planning and operation. A hybrid short-term load probability density forecasting method based on convolutional bi-directional long short-term memory quantile regression blending attention mechanism is proposed. Firstly, the weather variables and historical loads are selected by using relevant mechanisms.Next, the Copula model is used to calculate the risk threshold,which is used to construct the peak binary indicator input characteristics. Then, the selected feature sets are input into the convolutional bi-directional long short-term memory quantile regression blending attention mechanism prediction model. Then, kernel density estimation is used to fit the load probabilistic prediction. Finally, the prediction performance is evaluated using the mean absolute percentage error and root mean square error. The simulation results show that proposed model has higher prediction accuracy.