基于聚类经验模态分解-样本熵和优化极限学习机的风电功率多步区间预测

Wind Power Multi-Step Interval Prediction Based on Ensemble Empirical Mode Decomposition-Sample Entropy and Optimized Extreme Learning Machine

针对风电功率序列的不确定性和随机性特征,提出一种基于聚类经验模态分解-样本熵和优化极限学习机的多步区间预测模型。首先,利用聚类经验模态分解-样本熵方法将原始风电功率序列分解为一系列复杂度差异明显的子序列。然后,分别对各子序列建立基于上下界直接估量的区间预测模型。为分析不同区间构造的差异,提出一种体现训练目标值偏离区间范围影响的新型区间预测评估指标作为目标函数,并采用基于混沌萤火虫结合多策略融合自适应差分进化的优化算法寻求其最优解,以提高模型预测性能。最后,以某一风电场实际功率数据为算例,验证了所提模型能获得可靠优良的多步区间预测结果,可为风电功率多步不确定性预测提供一种新的有效途径。

In allusion to uncertainty and randomness of wind power, a multi-step interval prediction model based on ensemble empirical mode decomposition-sample entropy（EEMD-SE） and optimized extreme learning machine is proposed. Firstly, EEMD-SE technique is adopted to decompose original wind power series into a number of subsequences with obvious complexity differences. Then interval prediction model based on lower-upper bound estimation method is established for each subsequence. In order to analyze differences among different interval structures, this paper proposes a new interval prediction evaluation index, referred to as objective function to indicate the influence when targets deviate from constructed intervals. In addition, a multi-strategy self-adaptive differential evolution combined with chaotic firefly algorithm is implemented to seek optimal solution of the problem, aiming to improve prediction performance of the proposed model. Finally, taking real power data of a wind farm as example, simulation results demonstrate that the proposed model can obtain reliable and excellent multi-step interval forecast, providing a new effective way for multi-step uncertainty prediction of wind power.