摘要
精确的电力负荷预测具有很大的经济和社会效益。本文基于深层神经网络研究负荷预测。文章首先分析了负荷预测中用到的关键特征,接着描述了深层神经网络和有监督的判别式预训练方法,以及文中使用的三种激活函数。最后,在一个较大的电力负荷数据集上比较了不同神经网络模型的预测效果。实验结果表明,使用有监督的预训练的深层神经网络具有最好的预测精度。
Accurate electrical load forecast has great economic and social value. In this paper, we study load forecast methods based on deep neural networks. We first analyze the critical features in load forecast. Then we present details of deep neural networks, supervised discriminative pre-training method, and the three activation functions used in this paper. We compare the performances of different neural network models and show the advantages of the proposed methods using a rather large data set of loads.
出处
《环境与可持续发展》
2016年第1期83-87,共5页
Environment and Sustainable Development
关键词
负荷预测
深层神经网络
预训练
激活函数
load forecast
deep neural networks
pre-training
activation function
