基于混合神经网络的多时间尺度负荷预测

Multi-time scales load forecasting based on hybrid neural network

精准的负荷预测对于电力系统保持经济、可靠运行有十分重要的意义,电力系统规划、运行、电力市场竞价系统的设计等都需要不同时间尺度的负荷预测结果,然而现有研究大多围绕一个时间尺度展开,单一模型难以满足实际工程需求。造成这一现象的原因在于模型预测结果的误差会随着预测时间尺度的延长而出现不同程度的增加,预测难度大,并且影响负荷的因素大多分布在不同时间尺度的数据上,难以充分利用。针对以上问题,本文在考虑负荷曲线的定积分与对应时间内用电量之间约束关系的前提下,提出融合多时间尺度数据的混合神经网络模型。该模型的损失函数同时考虑了点预测结果的误差以及负荷曲线定积分的物理意义,增强了负荷时间序列中各个元素之间的几何相关性。并且利用神经网络将短尺度数据提取为抽象的综合数据后,与长尺度数据拼接组成新的特征向量,用于预测不同时间尺度的负荷值。算例结果表明,本文提出的模型在实际的变压器负荷数据上能够实现多个时间尺度的预测并且有效提高预测精度。

Accurate load forecasting is a crucial foundational element for maintaining the safe and reliable operation of power system.For power system planning,operation,and design of power market bidding systems,among other things,results of load forecasts at various time scales are required.On the other hand,the bulk of current studies are conducted using a single time scale.It is challenging to match the actual engineering needs with a single model.The key reason for this is that the prediction error grows as the time scale expands,and the majority of the load-af-fecting components are dispersed throughout data at various time scales,making it challenging to effectively utilize them.In order to address the aforementioned issues,this paper suggests a hybrid network model that takes into ac-count of data from multiple time scales while also taking into account of the constrained relationship between the definite integral of the load curve and the amount of electricity consumed corresponding time.The loss function of the model accounts for both the physical significance of the definite integral of the load curve as well as the error of the point prediction.The geometric correlation among the elements in the time series is improved by the loss func-tion.A novel feature vector for forecasting load values at various time scales is created by using networks to extract small-scale data from abstract integrated data and then combining it with long-scale data.The suggested model can accurately forecast data from actual transformer load measurements at several time intervals.