«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

34

期数:

2024年10期

页码:

186-191

栏目:

新型计算应用系统

出版日期:

2024-10-10

文章信息/Info

Title:

Short-term Photovoltaic Power Prediction Based on MPBiLSTM

文章编号:

1673-629X(2024)10-0186-06

作者:

陈君; 郭立颖*; 赵小会; 李维乾; 季虹

西安工程大学 计算机科学学院,陕西 西安 710048

Author(s):

CHEN Jun; GUO Li-ying*; ZHAO Xiao-hui; LI Wei-qian; JI Hong

School of Computer Science,Xi’an Polytechnic University,Xi’an 710048,China

关键词:

短期; 光伏功率预测; 残差反转一维卷积; 双向长短期记忆网络; Luong注意力机制; 深度学习

Keywords:

short-term; photovoltaic power prediction; Residual Reverse One-dimensional Convolution; Bidirectional Long Short Term Memory; Luong Attention; deep learning

分类号:

TP39

DOI:

10.20165/j.cnki.ISSN1673-629X.2024.0204

摘要:

由于化石能源对环境有一定程度的危害,太阳能作为可再生的绿色能源,受到广泛关注。 光伏发电是太阳能的利用途径之一,其相关技术正飞速发展。 然而,光伏发电由于受到天气及其他因素的影响,具有不稳定性的特点。 因此,为了保证发电策略的科学性,光伏发电功率预测极为重要。 为了提高短期光伏发电预测的准确性,提出了一种基于特征融合和多路径的深度学习模型。 首先,该模型使用变分模态分解(Variational Mode Decomposition,VMD)对历史发电功率序列进行分解,并结合斯皮尔曼相关系数(Spearman Correlation Coefficient,SCC)处理无关序列和异常值,形成每个包含本征模函数序列的矩阵。 接着,将矩阵数据输入预测模型,该模型利用残差反转一维卷积(Residual Reverse One-dimensional Con-volution,RROC),通过为每个结构提供不同数量的卷积核以及多路径结构来实现特征融合。 此外,该方法还引入了堆叠的双向长短期记忆网络(Bidirectional Long Short Term Memory,BiLSTM)和 Luong 注意力机制,使网络更加精密。 最终,将每个本征模函数的输出相加得到每个点或区间的预测值。 与其他方法相比,基于多路径双向长短期记忆网络(Multiple-Path BiLSTM,MPBiLSTM)的模型具有更好的预测结果。

Abstract:

As fossil energy has a certain degree of harm to the environment,solar energy,as a renewable green energy source,has received widespread attention. Photovoltaic power generation is one of the ways to utilize solar energy,and its related technology is developing rapidly. However, photovoltaic power generation is characterized by instability due to the influence of weather and other factors.Therefore,in order to ensure the scientific validity of the power generation strategy,photovoltaic power prediction is extremely important.In order to improve the accuracy of short-term power generation prediction,a deep learning model based on feature fusion and multipath techniques had been proposed. Initially,the model used Variational Mode Decomposition ( VMD) to decompose the historical power generation sequence and combined Spearman Correlation Coefficient (SCC) to process irrelevant sequences and outliers to form a matrix that contains each of the Intrinsic Mode Function. Subsequently,the matrix was fed into a prediction model that utilized the Residual Reverse One-dimensional Convolution (RROC) network,which achieved feature fusion through varying numbers of convolution kernels for each structure as well as a multi-path structure. In addition,the proposed method incorporated stacked Bidirectional Long Short Term Memory (BiLSTM) and Luong Attention to complicate the network. Eventually,the output of each Intrinsic Mode Function is summed to obtain the predicted value for each point or interval. It has been proved that the model based on Multiple - Path BiLSTM (MPBiLSTM) has better prediction results compared with other methods.

常用功能

更新日期/Last Update: 2024-10-10