摘要
交直流(AC/DC)配电系统中的负荷类型更加多样化和复杂化,因此负荷变化规律也更加难以掌握,精确的负荷预测对AC/DC配电系统的调度非常重要。针对神经网络、灰色理论和支持向量机等传统的短期负荷预测方法存在的预测精度不高的问题,本文采用改进集成学习算法对传统的预测方法进行改进,提出了一种基于浅层神经网络的梯度提升算法(GBSNN)以及基于长短期记忆网络的极度梯度提升(XGBLSTM)算法。同时,本文采用Huber函数作为预测模型的损失函数,该函数对异常的负荷数据具有很强的鲁棒性,可以有效减小模型的泛化误差。最后通过仿真分析证明本文提出的基于GBSNN和XGBLSTM的短期负荷预测方法比其他方法具有更高的预测精度,在AC/DC配电系统负荷预测中具有更好的效果。
There are more diverse types of loads in an AC/DC distribution system,so it is much more difficult to grasp the change rules.Accurate load forecasting is of great significance for the optimal scheduling of AC/DC distribution system.Aiming at the precision problem of traditional short-term load forecasting methods,such as neural network,grey theory and support vector machine,this study uses the ensemble learning to improve the traditional forecasting methods,and proposes a gradient boosting method based on shallow neural network(GBSNN)as a base learner.Meanwhile,by using the Huber function as the loss function,it is robust to abnormal load data and can reduce the generalization error.Through simulation results and comparison analysis,the proposed short-term load forecasting method based on GBSNN has higher precision than other methods and better performance in load forecasting of AC/DC distribution system.
作者
姜世公
王云飞
吴志力
崔凯
陈庆
Jiang Shigong;Wang Yunfei;Wu Zhili;Cui Kai;Chen Qing(State Grid Economic and Technological Research Institute Co.,Ltd,Beijing 102209,China)
出处
《科技通报》
2021年第7期68-73,79,共7页
Bulletin of Science and Technology
基金
国家重点研发计划项目(2018YFB0904700)
关键词
负荷预测
AC/DC配电系统
集成学习
浅层神经网络
load forecasting
AC/DC hybrid distribution System
ensemble learning
shallow neural network
