摘要
针对目前国内外风功率预测精度普遍不高,预测效果难以为电力系统调度提供可靠保障的现象,提出了一种基于CNN的区域风功率预测方法。采用k阶最小二乘法曲线拟合和BN归一化对历史数据进行预处理,将总数据集划分为训练集、验证集和测试集,数据矩阵化作为网络的输入,训练结束后保存最优模型,通过滚动预测的方法得出最终预测结果。算例结果表明,上述方法适合对未来超短期风功率曲线进行准确预测,预测精度较传统方法和现有水平均有不同程度的提高。
In view of the fact that the accuracy of wind power forecasting is generally not high at home and abroad, and the forecasting effect is difficult to provide reliable guarantee for power system dispatching, a regional wind power forecasting method based on the CNN was proposed. The k-order least squares curve fitting and BN normalization were used to preprocess the historical data. The total data set was divided into three parts: training set, validation set, and test set. The data matrix was used as the input of the network. After training, the optimal model was saved and the final prediction results were obtained by the rolling prediction method. The numerical results show that the above methods are suitable for accurate prediction of future ultra-short-term wind power curves, and the prediction accuracy is improved in varying degrees compared with traditional methods and existing levels.
作者
查雯婷
杨帆
陈波
李亚龙
ZHA Wen-ting;YANG Fan;CHEN Bo;LI Ya-long(College of Mechanical and Electrical and Information Engineering,China university of mining and technology Beijing,Beijing 100083,China;Inner Mongolia Electric Power Research Institute,Hohhot Inner Mongolia 010020,China)
出处
《计算机仿真》
北大核心
2021年第5期318-323,共6页
Computer Simulation
基金
国家自然科学基金项目(61703405)。
关键词
风功率预测
卷积神经网络
超短期预测
季节性分析
误差分析
Wind power prediction
Convolutional neural network(CNN)
Ultra-short-term prediction
Seasonal analysis
Error analysis
