摘要
针对风电出力非线性、不稳定且用传统方法难以准确预测的问题,提出了一种基于对深层混合核极限学习机(DHKELM)参数进行优化的短期风电功率预测。利用核主成分分析(KPCA)方法进行特征优选得到的最优特征集,既能表达风电功率的有效信息,也能避免冗余信息的出现,有利于DHKELM模型的学习与训练,同时也降低了模型的复杂度。针对DHKELM超参数难确定的问题,利用改进的野犬优化算法(IDOA)对DHKELM的8个超参数进行寻优,可以发掘原始序列特征信息,从而使模型能够充分掌握数值天气预报(NWP)与风电功率之间的非线性关系。以国外某风电场真实数据为算例,结果表明:提出的预测模型相较于野犬算法、差分进化算法和粒子群优化算法的平均绝对百分比误差(MAPE)分别降低了0.979 3%、2.342 1%、3.383 2%,有效提高了风电功率的预测精度。
Aiming at the problem that wind power output is nonlinear, unstable and difficult to be accurately predicted by traditional methods, this paper proposes a short-term wind power prediction based on the optimization of parameters of the deep hybrid kernel extreme learning machine(DHKELM). The kernel principal component analysis(KPCA) method is used to well select the features to form an optimal feature set, which can not only express the effective information of wind power, but also avoid the appearance of redundant information, and is thus conducive to facilitating the learning and training of the DHKELM model and reducing the complexity of the model. In view of the problem that it is difficult to determine the hyperparameters of DHKELM, the improved dingo optimization algorithm(IDOA) is used to find the eight optimal hyperparameters of DHKELM and explore the original sequence feature information, so that the model can fully grasp the nonlinear relationship between numerical weather prediction(NWP) and wind power. Taking the real data of a foreign wind farm as an example, the results show that the proposed prediction model effectively improves the accuracy of wind power prediction, with the mean absolute percentage error(MAPE) 0.979 3%, 2.332 1% and 3.383 2% lower than that of the dingo optimization algorithm, the differential evolution optimization algorithm and the particle swarm optimization algorithm respectively.
作者
商立群
黄辰浩
侯亚东
李洪波
惠泽
张建涛
SHANG Liqun;HUANG Chenhao;HOU Yadong;LI Hongbo;HUI Ze;ZHANG Jiantao(School of Electrical and Control Engineering,Xi’an University of Science and Technology,Xi’an 710054,China)
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2023年第1期66-77,共12页
Journal of Xi'an Jiaotong University
基金
陕西省自然科学基础研究计(2021JM-393)。
关键词
短期风电功率预测
深层混合核极限学习机
改进的野犬优化算法
特征优选
核主成分分析
short-term wind power prediction
deep hybrid kernel extreme learning machine
improved dingo optimization algorithm
feature optimization
kernel principal component analysis
