基于改进核密度估计的光伏发电功率区间预测  

Photovoltaic power generation power interval prediction based on improved kernel density estimation

作　　者：金方承 姜建国[1] JIN Fangcheng;JIANG Jianguo(School of Electrical and Information Engineering,Northeast Petroleum University,Daqing 163000,China)

机构地区：[1]东北石油大学电气信息工程学院,黑龙江大庆163000

出　　处：《电子设计工程》2024年第24期36-41,共6页Electronic Design Engineering

基　　金：黑龙江省自然科学基金(LH2022F005)。

摘　　要：该文提出了一种基于改进核密度估计和GRO-CNN-LSTM-Attention的光伏发电功率区间预测模型。该模型通过引入淘金优化算法(GRO)与SE注意力机制提高了点预测模型的准确度,同时通过自适应带宽参数与混合概率密度函数两种策略改进核密度估计区间预测方法,使得改进后的区间预测方法在满足置信度要求的前提下平均区间宽度更窄,提高了区间预测的准确性。在三种不同的天气类型下分别以85%、90%和95%的置信度进行光伏发电功率区间预测,采用预测区间覆盖概率和预测区间平均宽度作为评价指标与其他预测方法进行对比分析,验证了所提预测模型的有效性与准确性。This article proposes a photovoltaic power generation interval prediction model based on improved kernel density estimation and GRO-CNN-LSTM-Attention.This model improves the prediction accuracy of point prediction by introducing the Gold Rush Optimizer(GRO)algorithm and SE attention mechanism.At the same time,the kernel density estimation interval prediction method is improved through two strategies:adaptive bandwidth parameters and mixed probability density function.This makes the improved interval prediction method narrower on average while meeting the requirements of confidence,thereby improving the accuracy of interval prediction.The photovoltaic power generation interval prediction was conducted with 85%,90%and 95%confidence levels under three different weather types.The Predicted Interval Coverage Probability and Predicted Interval Average Width were used as evaluation indicators to compare and analyze with other prediction methods,verifying the effectiveness and accuracy of the proposed prediction model in this paper.

关 键 词：核密度估计 淘金优化算法 注意力机制 CNN-LSTM 区间预测 光伏发电 

分 类 号：TN209[电子电信—物理电子学]