基于变分模态分解改进生成对抗网络的短期风电功率预测

Short-term wind power forecasting based on variational mode decomposition and generative adversarial networks

风电功率的可预测性和预测准确性取得了一定的研究成果,但风电数据中气象和功率的强非线性制约了短期预测精度的进一步提高,提高短期风电功率预测的精度已成为研究的热点与难点。针对风电数据非线性且非稳定的特点,基于分解思想提出一种基于变分模态分解改进生成对抗网络的短期风电功率预测方法。该方法使用变分模态分解分散风电数据中的非线性,将复杂序列的预测任务转化为多个较为简单序列的预测任务;设计了激活函数和损失函数,解决传统生成对抗网络模型不稳定问题,并对所设计激活函数的关键参数进行了分析。Bengaluru风电场某风机数据的算例测试表明,所提方法取得了较好的预测结果,其均方误差相比长短期记忆网络和变分模态分解-长短记忆网络方法分别下降了79.65%和51.83%。

Although certain progress has been achieved on wind power predictability and prediction accuracy,the shortterm prediction accuracy is restricted by the strong nonlinearity of meteorological and wind power data.Hence,the improvement on short-term wind power prediction is a research hotspot.To deal with the nonlinearity and instability of wind power data,a short-term wind power forecasting method based on variational model decomposition and generative adversarial networks is proposed.The method introduces variational mode decomposition to disperse the nonlinearity of wind power data and decompose the complex sequence of prediction task into several simple sequences.Activation function and loss function are designed to solve the instability of traditional generative adversarial network models,and the key parameters of the designed activation function are analysed.The proposed method is tested on the data from a Bengaluru wind farm,showing a decent prediction result.Its forecasting mean square error is 79.65%and 51.83%lower than that of long short-term memory and variational mode decomposition-long short-term memory,respectively.