基于样本权重的深度神经网络短期电价预测方法研究——以美国PJM实际电价数据检验预测精度

Short-Term Price Forecasting Method Based on Deep Neural Network with Sample Weights——Test by the Actual Data of USA PJM

电力市场环境下短期电价预测面临全新挑战,其预测结果的准确性对市场主体报价决策具有重大意义。对此,本文提出一种基于样本权重的深度神经网络(Deep Neural Network,DNN)短期电价预测方法,通过对样本进行筛选并为不同训练样本赋予相应的权重,有效提升DNN模型的电价预测精度。样本权重赋值方法的两个重要步骤为:(1)通过计算样本数据间的欧式距离衡量样本间的相关程度,并以此为依据挑选训练样本;(2)根据各训练样本数据与预测日数据之间的欧式距离为训练样本赋予不同权重,使得DNN能有选择、有重点地对训练样本进行学习。模型构建后,对2020年1月美国PJM实际电价数据进行虚拟预测,结果表明:所提方法能有效提升电价预测的准确性和可靠性,可为市场环境下市场主体提供可靠的决策依据。

The reform of power market raises a new challenge to short-term price forecasting, the accuracy of whose predicted results is also of great significance to the market participants. To improve the accuracy of price forecasting, a short-term price forecasting method based on Deep Neural Network(DNN) with sample weights is proposed in this paper. By filtering the samples and assigning corresponding weights to different training samples, this method effectively improves the prediction accuracy of the DNN model. Two important steps of this method are as follows:(1) Calculate the Euclidean distance among the sample data to measure the degree of correlation among samples, and then select the training samples according to the Euclidean distance;(2) After selecting training samples, corresponding weights can be assigned to training samples according to the Euclidean distance between the training data and the data of the forecast day. Sample weights enable the DNN to study training samples selectively. Finally, this paper uses the actual data of USA PJM in January 2020 to simulation. The results show that the proposed method can improve effectively the accuracy and reliability of price forecasting, and provide a reliable decision-making basis for market participants under the market environment.