基于改进熵权法和SECEEMD的短期风电功率预测

为实现短期风电功率的高精度预测，综合考虑风电功率数据波动性以及多维气象数据对风电功率预测的影响，提出了一种基于改进熵权法和SECEEMD的短期风电功率组合预测方法。首先，提出了一种综合相关性分析模型，结合多种特征选择方法对多维气象特征实现综合评价，准确筛选与风电功率相关性较高的气象特征，提高预测精度。其次，针对CEEMD(Complementary Ensemble Empirical Mode Decomposition,互补集合经验模型模态分解)存在的分解分量过多，模态混叠程度加剧的问题，提出了SECEEMD分解算法，在降低分量数量，降低模态混叠程度的同时，提高模型的训练速度。然后，分别建立NWP-LSTM和SECEEMD-BP预测模型，并通过贝叶斯优化算法优化长短期记忆神经网络和BP神经网络结构；最后，通过改进熵权法寻找到最优权重组合进行加权组合。实验以内蒙古某风电场的风电功率数据和气象数据为实验数据，经验证，本文所提预测模型，能较大程度提高预测精度，相较于一般预测模型，R2-Score分别提高了4%和0.6%，MAE分别降低了44%和1.1%，证明本文所提风电功率预测方法具有更高的预测精度和更快的训练速度，更加适合进行风电功率预测。

Short term wind power prediction based on improved entropy weight method and SECEEMD

In order to achieve high-precision short-term wind power prediction, a short-term wind power combination prediction method based on improved entropy weight method and SECEEMD is proposed, taking into account the volatility of wind power data and the impact of multi-dimensional meteorological data on wind power prediction. First of all, a comprehensive correlation analysis model was proposed, which combined multiple feature selection methods to achieve comprehensive evaluation of multi-dimensional meteorological features, accurately screened meteorological features with high correlation with wind power, and improved prediction accuracy. Secondly, in view of the problem that CEEMD (Complementary Ensemble Empirical Mode Decomposition) has too many decomposition components and increases the degree of modal aliasing, the SECEEMD decomposition algorithm was proposed to reduce the number of components and the degree of modal aliasing, while improving the training speed of the model. Then, NWP-LSTM and SECEEMD-BP prediction models were established respectively, and the structure of long short-term memory neural network and BP neural network were optimized by Bayesian optimization algorithm. Finally, the optimal weight combination was found through the improved entropy weight method. The experiment takes the wind power data and meteorological data of a wind farm in Inner Mongolia as the experimental data. It has been verified that the prediction model proposed in this paper can greatly improve the prediction accuracy. Compared with the general prediction model, R2-Score has increased by 4% and 0.6% respectively, and MAE has decreased by 44% and 1.1% respectively. The experiment proves that the wind power prediction method proposed in this paper has higher prediction accuracy and faster training speed, and is more suitable for wind power prediction.