基于QR-NFGLSTM与核密度估计的风电功率概率预测

PROBABILITY PREDICTION OF WIND POWER BASED ON QR-NFGLSTM AND KERNEL DENSITY ESTIMATION

为提高风电功率概率预测精度和缩短长短期记忆网络的训练时间,提出一种基于分位数回归结合新遗忘门长短期记忆(NFGLSTM)网络与核密度估计的风电功率概率预测方法。该方法对长短期记忆网络的结构改进,提出一种新的遗忘门结构,以缩短训练时间。基于分位数回归和NFGLSTM网络建立组合预测模型,得到风电功率点预测值和某一置信度下的预测区间,采用Cosine核函数的核密度估计求解预测值的概率密度函数。基于某风电场的实测数据的算例分析表明,和传统预测方法相比,该方法可缩短长短期记忆网络的训练时间,提高概率预测精度。

This paper proposes a wind power probability prediction method in order to improve the accuracy of wind power probability prediction and reduce the training time of long short-term memory network. This method is based on quantile regression combined with a new forget gate long short-term memory(NFGLSTM)network and kernel density estimation. The structure of long short-term memory network is improved and a new forget gate structure is proposed,which is used to shorten the training time. A combined forecasting model is established based on quantile regression and NGFLSTM network. So,the point prediction value of wind power and the prediction interval under a certain confidence are obtained. The kernel density estimation of the Cosine kernel function is used to solve probability density function of the predicted value. The case study shows that the proposed method can shorten the training time of long and short-term memory networks and improve the probability prediction accuracy.