基于深度时间卷积神经网络的风电功率预测

Wind Power Forecasting Based on Deep Temporal Convolutional Networks

为了提高风力发电预测的准确性,依据某近海地区风电场出力数据,提出基于深度时间卷积网络的风电功率组合预测模型;利用自适应集成经验模态分解对风电功率序列进行特征提取,得到若干本征模态分量,通过排列熵相关理论计算各模态分量的复杂度,根据复杂度进行序列重构,并输入至改进余弦退火算法优化的深度时间卷积网络中进行风电功率分析与预测。结果表明,该模型与其他模型相比具有较好的预测效果,能够有效提高超短期风电功率预测精度。

To improve accuracy of wind power forecasting,a combined wind power forecasting model based on deep temporal convolutional networks was proposed according to wind farm generation output data of a nearshore area.Complete ensem ble empirical mode decomposition with adaptive noise was used to extract characteristics of wind power sequences,and several intrinsic modal components were obtained.Complexity of each modal component was calculated according to cor relation theory of permutation entropy,and the sequence was reconstructed according to the complexity,which was input to deep temporal convolutional networks optimized by using improved cosine annealing algorithm for wind power analysis and forecasting.The results show that this model has better forecasting effect than other models,and can effectively improve forecasting accuracy of ultra-short-term wind power.