基于ISSA与GRU分位数回归的风电功率概率密度预测

Wind power probability density prediction based on ISSA and GRU quantile regression

针对风电功率点预测无法对风电功率时序的不确定性进行风险评估的问题,本文提出了基于变分模态分解结合改进的松鼠算法优化门控循环单元(GRU)分位数回归的短期风电功率概率密度预测方法。采用变分模态分解降低数据样本训练的复杂程度,利用折射反向学习策略和加入惯性权重基于箕舌线调整的方法对松鼠算法进行改进,通过改进的松鼠算法对GRU分位数回归的超参数进行寻优,获取改进后的松鼠算法优化GRU分位数回归的风电功率概率密度模型。结果表明,得到改进的模型对比初始模型具有更高的精度和效率,均方根误差、平均绝对误差评价指标分别为0.775 7 MW和0.619 6 MW。用调优后的模型预测不同分位点下的风电功率,并通过核密度估计法获得风电功率的概率密度函数。最后,利用中国实际风电场的实际运行数据对提出的理论和方法进行了实验研究。

In response to the inability of wind power point forecasting to assess the risk of uncertainty in wind power time series,this paper proposes a short-term wind power probability density forecasting method based on variational modal decomposition combined with an improved squirrel algorithm to optimise GRU quantile regression.Variational modal decomposition is used to reduce the complexity of data sample training,the refractive reverse learning strategy and the method of adding inertia weights based on tongue line adjustment are used to improve the squirrel algorithm,and the hyperparameters of GRU quantile regression are optimized by the improved squirrel algorithm to obtain the wind power probability density model of ISSA-QRGRU.The results showed that the improved model has higher accuracy and efficiency,with the root mean square error and mean absolute error evaluation indexes of 0.7757 MW and 0.6196 MW,respectively.The tuned model is used to predict the wind power at different quantile points,and the probability density function of the wind power is obtained by the kernel density estimation method.Finally,empirical research was conducted on the proposed theory and method using actual operating data of wind farms in China.