###
DOI:
电力大数据:2019,22(8):-
←前一篇   |   后一篇→
本文二维码信息
基于大数据和深度学习的全社会月用电总量预测
刘智
(国家能源集团泰州发电有限公司)
Forecast the total monthly electricity consumption of the whole society by exploting big data and deep learning
Liu Zhi
(National Energy Investment Group Taizhou Power Generation Co.,Ltd.)
摘要
图/表
参考文献
相似文献
本文已被:浏览 28次   下载 29
投稿时间:2019-05-30    修订日期:2019-06-06
中文摘要: 为解决经典方法预测全社会用电总量预测数值精度较低、模型结构参数过于复杂等技术难题,本文提出将电力大数据和人工智能领域深度学习算法相结合的研究方法。采用计算机建立具有阶层结构的深度神经网络,根据仿生学原理引入线性整流函数解决梯度消失及神经网络收敛速度减慢问题,采用梯度下降来进行优化模型,同时通过引入指数衰减法由神经网络模型自动设定学习率以提高模型预测精度并降低迭代次数。从数量场的梯度原理并结合泰勒公式,推导出梯度下降法背后数学原理。为解决过拟合问题引入早停算法以提高模型训练速度及泛化能力。最后深度学习算法预测数值与经典线性回归算法预测数值相比较,深度学习算法在对全社会月用电总量的预测精准度、稳定性指标上明显优于线性回归算法,深度神经网络模型对未来全社会电力需求的预测数值具有高度的可信性。
Abstract:In order to solve the technical problems such as the low numerical accuracy of the classical method to forecast the total electricity consumption of the whole society and the excessively complex structural parameters of the model, this paper proposes a research method combining the big data of electric power with the deep learning algorithm in the field of artificial intelligence. Using computer to build a deep neural network with hierarchical structure, According to bionics principle, Rectified Linear Unit is introduced to solve the problem of gradient disappearance and slow convergence rate of neural network, using gradient descent to optimize model, at the same time, by introducing exponential decay vector automatically set by the neural network model in order to improve the model prediction accuracy and reduce the number of iterations. From the gradient principle of quantity field and Taylor formula, the mathematical principle behind gradient descent method is deduced. In order to solve the overfitting problem, the early stopping algorithm is introduced to improve the training speed and generalization ability of the model. Values forecasted by deep learning algorithms compared with the values forecasted by linear regression algorithm, deep learning algorithm for forecasting of the total electricity consumption of the whole society is superior to linear regression algorithm on the precision and stability index, and the depth of the neural network model for the forecast of future demand for electricity in the whole society has a high degree of credibility.
文章编号:     中图分类号:    文献标志码:
基金项目:
引用文本: