首页|基于INGO-SWGMN混合模型的超短期风速预测研究

基于INGO-SWGMN混合模型的超短期风速预测研究

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
为提高超短期风速预测的精度,提出一种融合变分模态分解(VMD)、相空间重构、改进的北方苍鹰优化算法(INGO)和共享权重门控记忆网络(SWGMN)的超短期风速混合预测模型.首先,考虑到风速的强波动性会对预测带来不利影响,采用VMD对风速时间序列进行分解,得到一系列相对平稳的子序列.然后对各子序列分量进行相空间重构,得到相应的相空间矩阵.接着针对长短期记忆网络(LSTM)训练时间较长和权重参数较多的问题,提出一种SWGMN对各子序列分量建立预测模型.同时,为提高模型预测性能,提出一种INGO对SWGMN模型的两个超参数进行寻优,得到最优参数组合.最后累加各子序列预测值,得到最终风速预测结果.实验结果表明,在单步预测和多步预测中,所提方法的平均绝对误差(MAE)、均方根误差(RMSE)、平均绝对百分比误差(MAPE)、决定系数R2分别为0.1828 m/s、0.2263 m/s、4.5481%、0.987和0.2429 m/s、0.3107 m/s、6.1113%、0.976,相较于传统方法具有更高的预测精度和预测效率.
ULTRA-SHORT-TERM WIND SPEED PREDICTION BASED ON INGO-SWGMN HYBRID MODEL
In order to improve the accuracy of ultra-short-term wind speed forecasting,a hybrid ultra-short-term wind speed prediction model that incorporates variational modal decomposition(VMD),phase space reconstruction,improved northern goshawk optimization algorithm(INGO)and shared weight gated memory network(SWGMN)is proposed.First,considering that the strong volatility of wind speed can adversely affect the prediction,the wind speed time series are decomposed by VMD to obtain a series of relatively smooth subseries.Then the phase space reconstruction is performed for each subsequence component to obtain the corresponding phase space matrix.Subsequently,a shared weight gated memory network(SWGMN)is proposed for the problems of long training time and many weight parameters of long short-term memory network(LSTM),and the SWGMN is used to build a prediction model for each subseries component.Meanwhile,to improve the prediction performance of the model,an improved northern goshawk optimization algorithm(INGO)is proposed to find the optimal combination of the two hyperparameters of the SWGMN model.Finally,the predicted values of each subseries are superimposed to obtain the final wind speed prediction results.The experimental results show that the proposed method has higher prediction accuracy and efficiency compared with the traditional methods.

wind speedforecastingdeep learningvariational mode decompositionshared weight gated memory networkimproved northern goshawk optimization algorithm

付文龙、章轩瑞、张海荣、傅雨晨、刘兴韬

展开 >

三峡大学电气与新能源学院,宜昌 443002

三峡大学梯级水电站运行与控制湖北省重点实验室,宜昌 443002

中国长江电力股份有限公司,宜昌 443133

风速 预测 深度学习 变分模态分解 共享权重门控记忆网络 改进的北方苍鹰优化算法

国家自然科学基金湖北省自然科学基金梯级水电站运行与控制湖北省重点实验室开放基金

519090102022CFD1702202KJX10

2024

10.19912/j.0254-0096.tynxb.2022-1975

太阳能学报
中国可再生能源学会

太阳能学报

CSTPCD北大核心
影响因子：0.392
ISSN：0254-0096
年,卷(期)：2024.45(5)
  • 17