王瑞，李虹锐，逯静，等.基于VMD-LILGWO-LSSVM短期风电功率预测［J］.河南理工大学学报（自然科学版），2024，43（2）：128-136.

WANG R，LI H R，LU J，et al.Short-term wind power prediction based on VMD-LILGWO-LSSVM［J］.Journal of Henan Polytechnic University（Natural Science），2024，43（2）：128-136.

基于VMD-LILGWO-LSSVM短期风电功率预测

王瑞1，李虹锐1，逯静1，卜旭辉2

1.河南理工大学 计算机科学与技术学院，河南 焦作 454000；2.河南理工大学 电气工程与自动化学院，河南 焦作 454000

摘要: 目的 为了减小风电功率并入国家电网时产生的频率波动，提高风电功率预测精度，方法 提出一种结合变分模态分解（VMD）、改进灰狼算法（LILGWO）和最小二乘支持向量机（LSSVM）的风电功率短期预测方法。首先通过VMD方法将风电功率序列分解重构成3个复杂程度性不同的模态分量，降低风电功率的波动性；其次使用LSSVM挖掘各分量的特征信息，对各分量分别进行预测，针对LSSVM模型中重要参数的选取对预测精度影响较大问题，引入LILGWO对参数进行寻优；最后将各分量预测结果叠加重构，得到最终预测风电功率。结果 以宁夏回族自治区某地区风电站实际数据为例，对未来三天分别进行预测取平均值，本文方法的预测平均绝对误差（mean absolute error，MAE）为2.706 8 kW，均方根误差（root mean square error，RMSE）为2.021 1，拟合程度决定系数（R-Square，R2）为0.976 9，与对比方法3~6相比，RMSE分别降低了40.93%，25.21%，14.7%，6.24%；MAE分别降低了42.34%，28.04%，16.97%，7.77%；R2分别提升了4.21%，1.78%，0.82%，0.28%。预测时长方面，BP和LSSVM平均训练时间分别是10，138 s，虽然LSSVM预测时间较长但效果最好，采用PSO、GWO、LILGWO对LSSVM进行寻优后训练时间分别平均缩短了39，44，58 s。结论 仿真验证了所提方法在短期风电功率预测方面的有效性。

关键词: 风电功率；短期预测；变分模态分解；近似熵；改进灰狼算法；最小二乘支持向量机

doi:10.16186/j.cnki.1673-9787.2021110135

基金项目:国家自然科学基金资助项目（U1804147）；河南省科技攻关项目（222102210120）

收稿日期: 2022/12/05

修回日期: 2023/05/26

出版日期:2024/03/25

Short-term wind power prediction based on VMD-LILGWO-LSSVM

WANG Rui1，LI Hongrui1，LU Jing1，BU Xuhui2

1.School of Computer Science and Technology，Henan Polytechnic University，Jiaozuo 454000，Henan，China；2.School of Electrical Engineering and Automation，Henan Polytechnic University，Jiaozuo 454000，Henan，China

Abstract: Objectives In order to reduce frequency fluctuations caused by wind power integration into the national grid and improve the accuracy of wind power prediction， Methods a short-term wind power prediction method combining variational modal decomposition（VMD），lens imaging learning grey wolf optimizer（LILGWO）and least squares support vector machine（LSSVM）was proposed.Firstly，the wind power sequence was decomposed into three modal components with different complexity by VMD method to reduce the fluctuation of wind power.Then LSSVM was used to mine the feature information of each component，which was predicted separately.For the problem that the selection of important parameters in LSSVM model had a large impact on the prediction accuracy，LILGWO was introduced to optimize the parameters.Finally，the prediction results of each component were superimposed and reconstructed to obtain the final predicted wind power. Results Taking the actual wind power station data in Ningxia Hui Autonomous Region as an example， average the predictions for the next three days. The prediction mean absolute error（MAE） of the proposed method in this article was 2.706 8 kW，the root mean square error （RMSE） was 2.0211，and the coefficient of fit determination（R-Square，R2） was 0.976 9.Compared with the comparison methods 3~6 mentioned in the article， the RMSE of the method decreased by 40.93%，25.21%，14.7%，and 6.24%，respectively；MAE decreased by 42.34%，28.04%，16.97%，and 7.77%，respectively；R2 increased by 4.21%，1.78%，0.82%，and 0.28%，respectively.In terms of prediction time，the average training time for BP and LSSVM was 10 seconds and 138 seconds，respectively.Although LSSVM had a longer prediction time，it performed the best.PSO，GWO and LILGWO were used to optimize LSSVM，and the training time was shortened by an average of 39 seconds，44 seconds，and 58 seconds，respectively. Conclusions The effectiveness of the decomposition algorithm proposed in this paper had been verified through comparative experiments.The lens imaging learning grey wolf optimizer algorithm proposed in this paper had excellent optimization ability for the key parameters of least squares support vector machines.The effectiveness of the proposed method in short-term wind power prediction was verified by simulation.

Key words: wind power；short-term prediction；variational mode decomposition；approximate entropy；lens imaging learning grey wolf optimizer algorithm；least squares support vector machine