基于深度学习的中长期风电发电量预测方法

Medium and Long Term Wind Power Generation Forecasting Method Based on Deep Learning

针对影响风电中长期预测的气象、地理等因素众多且复杂,及无法解决长期依赖时间序列的问题,提出一种基于多维特征融合网络(multi-dimensional feature fusion network,MFFN)和长短期记忆(long and short term memory,LSTM)的预测方法—多维特征提取(feature extraction,FE)-关联函数(copula,CO)-LSTM融合模型(FE-CO-LSTM)。收集来自不同地区4个风电场的特征数据,在研究云贵高原地区风电场的背景下,最大限度扩充模型数据集;使用关联结构函数构造一种提取气象特征的方法,使模型可以在一定程度上量化气象因素和风力发电之间的相关性;基于神经网络模型提出一种特征表示与融合方法,以有效表达风电场气象因素、地理位置等特征;最后提出一种基于LSTM网络的中长期发电量预测模型,以有效解决模型对风电场时间序列数据反向传播时早期月度数据信息缺失的问题。实验结果证明,FE-CO-LSTM表现出最佳的预测性能。

Aiming at the problems of numerous and complex factors such as meteorology and geography that affect the medium and long term forecasting of wind power,and long-term dependence on time series,this paper proposes a forecasting method based on the multi-dimensional feature fusion network(MFFN)and the long and short term memory(LSTM),that is multi-dimensional feature extraction(FE)-copula(CO)-LSTM fusion model(FE-CO-LSTM).By collecting the feature data from 4 wind farms in different regions and under the background of studying wind farms in the Yunnan-Guizhou plateau,the paper expands the model data set in the maximum extent.By using the correlation structure function,it constructs a method for extracting meteorological features,so that the models can quantify the correlation between meteorological factors and wind power generation to a certain extent.It also proposes a feature representation and fusion method based on the neural network model to effectively express the meteorological factors of the wind farms,geographic location and other characteristics.Finally,it propose a medium and long term power generation forecasting model based on the long and short term memory network to effectively solve the problem of the lack of early monthly data information when the model propagates back the time series data of the wind farms.The experimental results prove that FE-CO-LSTM has the best predictive performance.