摘要
风能作为一种清洁、可再生的能源,在能源转型中扮演着至关重要的角色,准确预测风电出力对电力系统的安全高效运行非常重要,然而风速的波动性和随机性,对风电预测带来了挑战。为了提高风电预测的准确性,提出了一种基于DenseNet卷积神经网络的短期风电预测模型。该模型通过精简DenseNet201网络得到了拥有出色的密集连接结构和适当深度、宽度的DenseNet160网络,不仅能缓解训练过程中梯度消失现象,还能通过密集连接将浅层的信息反映到深层,实现深度监督。基于巴西纳塔尔地区378 d的风力数据集,采用DenseNet160网络以及27种算法对未来一天的风力发电情况进行预测。结果表明:DenseNet160网络的平均绝对误差、均方误差以及平均绝对百分误差比其他算法分别降低了至少10.89%,4.98%,8.68%;同时,与使用相同数据集的混合经济模型相比,DenseNet160网络的MAE值小了25.56%。说明该模型能精准地拟合风力发电数据,获得可靠的风力预测结果。
Wind energy,as a clean and renewable energy source,plays a crucial role in the energy transformation.And accurate prediction on wind power output is important for the safe and efficient operation of the power system.However,the volatility and randomness of wind speed challenges the wind power prediction.To improve the accuracy of the prediction,a short-term wind power prediction model based on DenseNet convolutional neural network is proposed.The DenseNet160 network obtained from a simplified DenseNet160 network is of an outstanding densely connected structure,and proper depth and width,capable of solving vanishing gradient in the training process and realizing deep supervision by sending information from a upper layer to a deeper layer.Based on the 378-day wind power dataset collected from Natal in Brazil,the wind power output of the next day was predicted by DenseNet160 network and other 27 algorithms.The prediction results show that the mean absolute error(MAE),mean squared error(MSE)and mean absolute percentage error(MAPE)of the DenseNet160 network is 10.89%,4.98%and 8.68%smaller than that of the second best algorithm,respectively.Meanwhile,the MAE of the DenseNet160 network is 25.56%smaller than that of the hybrid economy model using the same dataset.This results indicate that the proposed prediction model can fit the wind power data more accurately and obtain more reliable wind power prediction results.
作者
殷林飞
蒙雨洁
YIN Linfei;MENG Yujie(School of Electrical Engineering,Guangxi University,Nanning 510004,China)
出处
《综合智慧能源》
CAS
2024年第7期12-20,共9页
Integrated Intelligent Energy
基金
国家自然科学基金项目(52107081)。
