摘要
售电量预测的精度决定了售电公司的运营收益。传统售电量预测方法存在未计及偏差电量考核机制的差异、缺少时序相关性与长程依赖性等问题。为此,提出一种计及偏差电量考核机制的人工神经网络售电量预测模型。首先,根据购售电交易时序特点重构特征向量。其次,建立基于季节分解的加权模型(SDW)与双向长短期记忆神经网络(Bi-LSTM)分别对年度双边协商月度分解售电量和月度集中竞价售电量进行预测,基于月度偏差考核规则定义非对称损失函数(ALF),关联反向传播过程与整体收益,使网络输出趋向收益最大化。最后,通过数据集进行算例仿真并比较各项性能指标,验证了该模型相比于传统预测模型经济实用性强,准确度高且稳定。
The accuracy of electricity consumption forecast determines the operating income of an electricity sales company.Traditional electricity consumption forecasting method has problems such as ignoring the difference of deviation electricity assessment mechanism,the lack of timing correlation and long-range dependence.To this end,an artificial neural network electricity sales forecast model that takes into account the deviation electricity assessment mechanism is proposed.Firstly,the feature vector is reconstructed according to the timing characteristics of electricity transactions.Secondly,the Seasonal Decomposition Weighted(SDW)model and Bidirectional Long Short-Term Memory(Bi-LSTM)networks are established to forecast the annual bilateral consultation monthly decomposition of electricity and monthly centralized bidding of electricity respectively.Based on the monthly deviation assessment rules,the asymmetric loss function(ALF)is defined.The back propagation process and the overall revenue correlation system are constructed to maximize the output of the network prediction.Finally,a numerical simulation is conducted in the data set and various performance indexes are compared,which verifies that the model is more economical,accurate and stable than traditional prediction models.
作者
白登辉
BAI Deng-hui(College of Electrical Engineering, Sichuan University, Chengdu 610065, China;Sichuan Provincial Architectural Design and Research Institute, Chengdu 610000, China)
出处
《电工电能新技术》
CSCD
北大核心
2020年第6期58-66,共9页
Advanced Technology of Electrical Engineering and Energy
基金
国家重点研发计划项目(2018YFC0704700)。
关键词
偏差电量考核
非对称损失
售电量预测
双向长短期记忆神经网络
季节分解
deviation electricity assessment
asymmetric loss
electricity consumption forecast
bidirectional long short-term memory networks
seasonal decomposition
