摘要
为了提高电力负荷预测精度和速度,针对极限学习机在电力负荷预测中需要人工设定隐层节点数目的不足,提出一种采用增量优化极限学习机的电力负荷预测方法。增量优化极限学习机算法是通过一个快速增量输出权值更新方法,对其进行增量优化改进,提高极限学习机算法的泛化性。最后,采用Mackey-Glass混沌时间序列和真实电力负荷时间序列数据进行算法性能测试。此外将提出的算法与主流电力负荷预测算法——支持向量回归算法比较预测性能,实验结果表明增量改进的极限学习机学习算法在具有较快学习速度的前提下,依然能够获得较好的预测性能和泛化性能。
In order to improve the load prediction accuracy and speed of power system,for the shortage of needing to manually set the hidden layer node number of extreme learning machine,this paper applies a power load prediction method based on incremental optimization extreme learning machine. The incremental optimization improvement is conducted through a fast incremental output weights update method,to improve the generalization of El M algorithm.The algorithm performance is tested by using the data of Mackey-Glass chaotic time series and real power load time series. In addition,the prediction performance of the online ELM sequence optimization algorithm is compared with the support vector regression electric load prediction algorithm. The experimental results show that the proposed algorithm has a fast learning speed,and can achieve good prediction performance and generalization performance.
出处
《计算机仿真》
CSCD
北大核心
2016年第2期163-166,377,共5页
Computer Simulation
关键词
时间序列预测
极限学习机
增量优化
电力负荷预测
Time series forecasting
Extreme learning machine(ELM)
Incremental optimization
Power load forecasting