基于IDANN的新能源基地风光时序功率曲线生成方法

A Wind-Solar Time Series Power Curve Generation Method for New Energy Bases Based on Improved Domain Adversarial Network

中文关键词:风光时序功率  迁移学习  改进域对抗网络  跨站点建模

keywords:wind-solar time series power  transfer learning  improved domain adversarial network  cross-site modeling

中文摘要:针对新建大型风电和光伏发电基地缺乏可用历史功率数据这一问题，提出基于改进域对抗网络的新能源基地风光时序功率曲线生成方法。以气象与功率历史数据完备的新能源场站为源域，以仅具备气象数据的新建基地为目标域，将源域中习得的气象信息到风光功率输出的非线性映射知识迁移至目标域，联合条件域对抗网络与最大均值差异损失并动态调整权重，以降低目标域泛化误差。结果验证了该模型的有效性。

Abstract:To address the lack of available historical power data for newly built large-scale wind and photovoltaic power bases，this paper proposes a wind-solar time series power curve generation method for new energy bases based on improved domain adversarial neural network （IDANN）. Taking new energy stations with complete meteorological and power historical data as the source domain and newly built bases with only meteorological data as the target domain，the nonlinear mapping relationship from meteorological information to wind-solar power output learned in the source domain is transferred to the target domain. Combined with the conditional domain adversarial network and maximum mean discrepancy loss，the weights are dynamically adjusted to reduce the generalization error of the target domain. The experimental results verify the effectiveness of the proposed model.