With the development of the global power market reform, the monopoly of the power sector and government control pattern has gradually broken. Due to the unique properties of electricity, electricity prices show high volatility and uncertainty, bringing significant challenges to the accurate prediction of electricity prices. The sudden occurrence of a few spike prices in the electricity spot market has significantly affected electricity price forecasting accuracy. We propose a novel two-stage electricity price forecasting scheme (TSEP). A multi-source data-based spike occurrence prediction scheme is presented in the first stage, which adopts a deep neural network (DNN) to predict whether the price to be forecasted is a spike or not. Specifically, to alleviate the impact of low spike price samples, the oversampling method is used to synthesize some spikes at the data level. A loss function with a misclassification penalty to increase the cost of missing price spikes is designed at the algorithm level. Based on the outputs of the first stage, in the second stage, TSEP exploits the variance stabilizing transformations respectively suitable for pre-processing spike and normal prices and combines an artificial neural network (ANN) based spike calibration model to improve the accuracy of electricity price forecasting further. The experimental results on the European Power Exchange for France (EPEX-FR) demonstrate that TSEP increases spike occurrence prediction accuracy compared with the conventional models and significantly improves the accuracy of spike electricity price forecasting without affecting the accuracy of forecasting normal electricity price.

随着全球电力市场改革的推进，电力行业的垄断和政府控制格局逐渐被打破。由于电力的独特属性，电价表现出高度的波动性和不确定性，给电价的准确预测带来了重大挑战。电力现货市场突然出现的几起尖峰电价，对电价预测的准确性造成了显着影响。我们提出了一种新颖的两阶段电价预测方案（TSEP）。第一阶段提出了一种基于多源数据的尖峰发生预测方案，该方案采用深度神经网络（DNN）来预测要预测的价格是否为尖峰。具体来说，为了减轻低尖峰价格样本的影响，过采样方法用于在数据层面合成一些尖峰信号。在算法级别设计了一个带有错误分类惩罚的损失函数，以增加丢失价格峰值的成本。基于第一阶段的输出，在第二阶段，TSEP利用分别适用于预处理尖峰和正常价格的方差稳定变换，并结合基于人工神经网络（ANN）的尖峰校准模型来提高电价的准确性进一步预测。