摘要
以极限学习分析模型为基础,对子网负荷变化进行了预测,同时比较了经过改进得到的核极限学习机模型与神经网络方法各自计算得到的结果。研究结果表明:利用聚类方法进行改进的KELM模型可以获得更低的Max-AE,使模型获得更优的拟合性能;采用聚类方法试试改进的KELM模型还可以达到比直接预测更高的效率,表明聚类得到的组合预测模型具备有效性,能够显著降低模型的预测误差;极限学习机模型不必对算法实施迭代,整体运行效率较高,最后可以获得一个最优解;核极限学习机方法可以达到最优状态,获得理想的泛化能力。
Subnet load change was predicted based on an extreme learning analysis model,and the results of respective calculation of the improved kern extreme learning machine(KELM)model and the neural network method were compared.Research results showed that the KELM model improved by the clustering method could achieve still lower Max-AE and give better fitting performance to the model;it could achieve even higher efficiency than direct prediction.That indicated that the combination prediction model obtained through clustering was effective and could greatly reduce prediction error of the model.The extreme learning machine model did not need to implement algorithm iteration;it could achieve high overall operation efficiency and finally obtain an optimal solution.The kern extreme learning machine method could reach its optimal state and obtain ideal generalization ability.
作者
朱勇
陶用伟
王常沛
李泽群
黄琼
杨键
Zhu Yong;Tao Yongwei;Wang Changpei;Li Zequn;Huang Qiong;Yang Jian(Guizhou Power Grid Co.,Ltd.Kaili Power Supply Bureau,Kaili Guizhou 556000,China)
出处
《电气自动化》
2020年第5期42-44,共3页
Electrical Automation
关键词
短期负荷
预测模型
改进核极限学习机
聚类方法
KELM模型
最优解
泛化能力
神经网络
short-term load
prediction model
improved kern extreme learning machine(KELM)
clustering method
KELM model
optimal solution
generalization ability
neural network