摘要
采用引入附加动量和自适应学习率的BP(Back Propagation)神经网络来构建全社会用电量预测模型,此模型有效地解决了标准BP神经网络容易陷入局部极小点和收敛速度慢的问题,并且能够很好地解决全社会用电量与其影响因素之间复杂的非线性关系.利用MATLAB7.0对该模型进行了设计,并用设计好的模型对1986~2005年的全社会用电量及GDP数据进行了仿真,仿真结果表明该模型收敛速度快、拟合效果好、泛化能力强、预测精度高.运用该模型对2006年全社会用电量进行了预测,预测结果表明该模型具有一定的实用价值.
In this paper, an improved BP Neural Network with additional momentum and adaptive learning rate is applied to set up the predictive model to social power consumption. It aims to solve the two main weak points of standard BP Neural Network that the optimal procedure is easily stacked into local minimum value and the convergent speed is too slow, what is more, the complex nonlinear relationship between the social power consumption and its influence factors can be effectively solved by this model. Applying MATLAB to design the model, according to the designed model, the social power consumption and GDP from 1986 to 2005 are simulated. The results show that the model is of rapid convergent speed, high precision and better-fit effect. In the end, the model is applied to forecast the social power consumption of 2006 ; the forecasting result indicates the model is effective.
出处
《华北电力大学学报(自然科学版)》
CAS
北大核心
2007年第3期85-89,共5页
Journal of North China Electric Power University:Natural Science Edition
基金
国家自然科学基金资助项目(70572090)
关键词
BP神经网络
全社会用电量
预测
动量项
自适应学习速率
BP Neural Network
social power consumption
forecasting
momentum factor
adaptive learning rate