基于Hadoop架构的多重分布式BP神经网络的短期负荷预测方法

发布时间：2018-10-05 11:22

【摘要】：随着智能电网、通信网络技术和传感器技术的发展,电力负荷数据规模呈现指数形式增长、且复杂程度增大,逐步构成了电力负荷大数据,传统负荷预测方法已无法满足海量负荷大数据分析的要求。提出一种基于Hadoop架构的多重分布式BP神经网络的短期负荷预测方法。该方法首先在从BP神经网络原理层对其输入信号的正向传递、误差信号的反向传播过程予以剖析的基础上,研究并建立基于Hadoop架构中Map Reduce框架的BP神经网络负荷分布式预测模型;其次,为弱化其"过拟合"问题,在引入"多重"概念的基础上,提出基于灰色关联度和最短距离法聚类的方式择取多重分布式BP神经网络预测模型初始重数和成员集的方法,并定义衡量聚类优劣的有效指标,以确定合理重数。实验结果表明,多重分布式BP神经网络预测方法相比传统BP神经网络,预测精度更高。
[Abstract]:With the development of smart grid, communication network technology and sensor technology, the scale of power load data increases exponentially and the complexity increases. Traditional load forecasting method can not meet the requirements of mass load big data analysis. This paper presents a short term load forecasting method for multiple distributed BP neural networks based on Hadoop architecture. The method is based on the analysis of the forward transmission of the input signal and the backward propagation of the error signal from the principle layer of the BP neural network. The distributed load forecasting model of BP neural network based on Map Reduce framework in Hadoop architecture is studied and established. Secondly, in order to weaken the problem of "overfitting", the concept of "multiple" is introduced. This paper presents a method of selecting the initial multiplicity and membership set of multiple distributed BP neural network prediction model based on grey correlation degree and shortest distance clustering method, and defines the effective index to evaluate the clustering quality in order to determine the reasonable multiplicity. Experimental results show that the multiple distributed BP neural network prediction method is more accurate than the traditional BP neural network.
【作者单位】：四川大学电气信息学院;国网信通产业集团北京中电普华信息技术有限公司;
【分类号】：TM715;TP183

【相似文献】