摘要
为了进一步提升区间预测的综合效果,提出一种基于竞争学习机制的LSTM风电多目标区间预测方法。提出基于LSTM的上下界估计模型来构建风电区间多目标预测模型,研究多目标系统中的估计误差与预测区间平均宽度的关系;进一步考虑预测误差,引入一种新的偏最小二乘评价指标;通过引入竞争学习机制,提出一种改进的非支配快速排序遗传算法,有效实现了多目标优化。最后通过两个实际风电数据集进行实验验证,结果表明提出方法具有较高的预测精度。
In order to further improve the comprehensive effect of interval prediction,a multiple objective interval prediction method of LSTM wind power based on competitive learning mechanism is proposed.The upper and lower bounds estimation model based on LSTM was proposed to construct the multiple objective prediction model of wind power interval,and the relationship between the estimation error and the average width of prediction interval in the multiple objective system was studied.Further considering the prediction error,a new partial least squares evaluation index was introduced.In addition,by introducing competitive learning mechanism,an improved non dominated quick sort genetic algorithm was proposed,which effectively realized multiple objective optimization.Two real wind power data sets were used to verify the proposed method.The results show that the proposed method has high prediction accuracy.
作者
任鹏
付文杰
申洪涛
陶鹏
张洋瑞
Ren Peng;Fu Wenjie;Shen Hongtao;Tao Peng;Zhang Yangrui(State Grid Hebei Marketing Service Center,Shijiazhuang 050000,Hebei,China)
出处
《计算机应用与软件》
北大核心
2024年第6期305-311,349,共8页
Computer Applications and Software
基金
国网河北省电力有限公司科技项目(kj2020-088)。
关键词
风电预测
长短期记忆网络
区间预测
遗传算法
Wind power forecasting
Long and short-term memory network
Interval forecasting
Genetic algorithm
