基于CNN-LSTM的短期风电功率预测  

Short-Term Wind Power Prediction Based on CNN-LSTM

作　　者：赵建利[1] 白格平 李英俊[2] 鲁耀[2] ZHAO Jianli;BAI Geping;LI Yingjun;LU Yao(.Inner Mongolia Electric Power Science & Research Institute,Huhehaote 010020,China;.Ulaanchab Electric Power Bureau,Wulanchabu 012000,China)

机构地区：[1]内蒙古电力科学研究院,内蒙古呼和浩特010020 [2]乌兰察布电业局,内蒙古乌兰察布012000

出　　处：《自动化仪表》2020年第5期37-41,共5页Process Automation Instrumentation

基　　金：内蒙古电力公司2018年重点科技基金资助项目(51CB41180007)。

摘　　要：短期风电功率预测对电力系统的安全稳定运行和能源的优化配置具有重要意义。鉴于卷积神经网络(CNN)高效的数据特征提取能力,以及长短期记忆网络(LSTM)描述时间序列长期依赖关系的能力。为了提高短期风电功率预测的精度,设计了一种基于CNN和LSTM的风电功率预测模型。该模型利用卷积神经网络对风电功率、风速、风向数据进行多层卷积和池化堆叠计算,提取风电功率相关数据的特征图谱。为了描述风电功率序列的时序依从关系,将图谱特征信息作为长短期记忆网络的输入信息,计算得到风电功率的预测结果。采用西班牙某风电场的实测数据进行模型预测精度验证。结果表明,该模型较LSTM、Elman模型具有更好的预测性能。Short-term wind power forecasting is of great significance for the safe and stable operation of power systems and the optimal allocation of energy.In view of the efficient data feature extraction ability of convolutional neural network(CNN)and the ability of long-term dependence of time series described by long-term memory(LSTM)network,and improving the accuracy of short-term wind power prediction,a wind power prediction model based on CNN and LSTM network is proposed.In this model,the convolutional neural networks is used to perform multi-level convolution and pooling stacking of wind power,wind speed,and wind direction data,and extract the characteristic-map of wind power related data.In order to describe the timing dependence of the wind power sequence,the feature data is used as the input information of the long and short-term memory network,and finally the wind power prediction result is obtained.Using the measured data of a Spanish wind farm to verify the model prediction performance,the testing results show that the proposed model has better prediction performance than the LSTM and Elman models.

关 键 词：时间序列 卷积神经网络 长短期记忆网络 特征提取 深度学习 风电功率预测 

分 类 号：TH-39[机械工程]