基于数据驱动的风电机组功率曲线建模方法

Data-driven Wind Turbine Power Curve Modeling Method

针对现有风电机组功率曲线建模存在非线性拟合能力不足,且不能很好的捕捉风速与风功率之间的复杂关系,提出了一种基于数据驱动的风电机组功率曲线建模的方法(mELM-CA-LSTM)。该方法利用多个极限机器学习机(Extreme Learning Machine,short for ELM)将单个的风速变量映射到多维特征空间中,组成多个特征图,通过通道注意力机制(Channel Attention,short for CA)减少高维空间特征图的冗余性,最后将长短时记忆网络(Long short-term memory network,short for LSTM)拟合风速与相应风功率之间非线性关系。对比分析了其他功率曲线建模的方法,所提的mELM-CA-LSTM方法在三个数据集上获得的最高的精度,验证了所提方法的有效性。

In view of the lack of nonlinear fitting ability of existing wind turbine power curve modeling and the inability to capture the complex relationship between wind speed and wind power,a data-driven wind turbine power curve modeling method(mELM-CA-LSTM)was proposed.This method uses multiple Extreme Learning machines(short for ELM)to map a single wind speed variable into a multidimensional feature space to form multiple feature maps.The Channel Attention mechanism(short for CA)is used to reduce the redundancy of high-dimensional spatial feature graphs.Finally,the Long short-term memory network(short for LSTM)is used to fit the nonlinear relationship between wind speed and corresponding wind power.Compared with other power curve modeling methods,the proposed mELM-CA-LSTM method obtained the highest accuracy on the three data sets,and verified the effectiveness of the proposed method.