文献类型：期刊文献

中文题名：基于组合数据清洗与NL-ConvLSTM模型的多步风电功率预测

英文题名：Multi-step Wind Power Prediction Based on Combined Data Cleaning and NL-ConvLSTM Model

作者：吴平雄 肖迎群 张苏 林兴宇

第一作者：吴平雄

机构：[1]贵州大学电气工程学院,贵州贵阳550025;[2]贵州理工学院大数据学院,贵州贵阳550003

第一机构：贵州大学电气工程学院,贵州贵阳550025

年份：2023

卷号：41

期号：1

起止页码：13-19

中文期刊名：机械与电子

外文期刊名：Machinery & Electronics

收录：CSTPCD

语种：中文

中文关键词：多步预测;风电功率预测;ConvLSTM;数据清洗;非局部操作

外文关键词：multi-step prediction;wind power prediction;ConvLSTM;data cleaning;non-local operation

摘要：针对风电数据在采集与传输过程中会产生大量缺失值和异常值,采用DBSCAN算法和最优组内差分法(OIV)组合筛删异常值,随机森林(RF)算法填补缺失值,提升数据准确性;并建立基于以ConvLSTM为单元的编码-预测(EF)网络的风电多气象输入多步预测模型,为了更好利用气象特征信息,在ConvLSTM模型的输入侧添加具有自注意力机制的非局部(NL)模块增强数据特征表现,从而搭建组合数据清洗方法的NL-ConvLSTM多步风电功率预测模型。实验结果表明,该方法能够进一步提高风电功率多步预测精度和稳定性。
Considering that a large number of missing values and outliers will be generated during the collection and transmission of wind power data,DBSCAN algorithm and optimal interclass variance(OIV)method are used to filter out outliers,and random forest algorithm is used to fill in the missing data to improve the accuracy of the data.The encoding-forecasting network based on ConvLSTM cell is a multi-step forecast model for wind power with multi-meteorological input.In order to make better use of meteorological feature information,a non-local module with self-attention mechanism is added to the input side of the encoding-forecasting network to enhance the representation of data features,so as to build the NL-ConvLSTM multi-step wind power prediction model,a combining data cleaning method.The experimental results show that this method can further improve the accuracy and stability of multi-step forecasting of wind power.