摘要
电力负荷预测是电网规划、运行、节能管理的重要基础,较高的波动性和不确定性的个体层级的电力负荷预测是一个难点问题。根据电力负荷固有的周期特性,提出一种将电力时间序列嵌入向量空间的模型——周期自动编码器PAE(Periodic Autoencoder)。通过聚合来减低电力负荷序列的不确定性和波动性;在嵌入空间中采用多种深度神经网络模型作为预测器,实现个体层级的电力负荷的准确预测。实验结果表明,PAE生成的电力时间序列嵌入能够捕捉电力负荷固有的周期特性,有效地降低其波动性和不确定。与传统方法相比,该电力负荷预测方法具有更高的预测精度。
Power load forecasting is an important basis for power grid planning,operation and energy-saving management.It is a difficult problem to forecast power load at individual level with high volatility and uncertainty.According to the inherent periodicity of electricity loads,we proposed a model named periodic autoencoder(PAE),which embedded electricity time series into a vector space.Aggregation was used to reduce the uncertainty and volatility of power load series.A variety of deep neural network models were applied as predictors in the embedded space to achieve accurate power load forecasting at individual level.Experiment results show that the electricity time series embedding generated by PAE can capture the inherent periodicity of electricity loads and effectively reduce the volatility and uncertainty.Compared with the traditional methods,the proposed electricity load forecasting method has higher prediction accuracy.
作者
田英杰
苏运
郭乃网
姚博
庞悦
周向东
Tian Yingjie;Su Yun;Guo Naiwang;Yao Bo;Pang Yue;Zhou Xiangdong(State Grid Shanghai Municipal Electric Power Company,Shanghai 200437,China;School of Computer Science,Fudan University,Shanghai 200433,China)
出处
《计算机应用与软件》
北大核心
2018年第11期55-60,73,共7页
Computer Applications and Software
基金
国家自然科学基金项目(61370157)
国家电网公司总部科技项目(52094016001Z)