Research On Regional Residents Load Forecasting Models Based On Clustering Analysis And Their Applications

Load forecasting is playing an increasingly important role in grid management.Especially as a large number of residential user-level load sequences are collected and stored,accurate load forecasting will help to further optimize the unit scheduling and reduce start-up and shutdown costs.At present,many studies use clustering analysis as a preprocessing step for regional load forecasting to improve accuracy.However,it mainly focuses on point load forecasting rather than interval load forecasting or probabilistic load forecasting.Interval load forecasting or probabilistic load forecasting can provide a more comprehensive forecast value distribution,which is of great significance and value to scheduling decisions.In this study,the clustering-based regional load forecasting framework is extended to interval load forecasting and probabilistic load forecasting.The effects of different clustering methods,similarity measurements,and the number of clusters on interval load forecasting and probabilistic load forecasting are explored.The significance of social type grouping is demonstrated.Also,this study applies the multi-series forecasting method Deep AR in combination with clustering tags to predict the probability distribution of load.The main research contents and innovation of this article are as follows:First,to cooperate with the subsequent interval load forecasting,this study adopts interval clustering to group the residents,and at the same time delves into the impact of the technical details of the interval clustering algorithm on group profiles.For cooperating with the subsequent probabilistic load forecasting,this study applies point clustering to group the residents.In addition to comparing the details of the clustering algorithm,the experiment creatively analyzes the similarities and differences between the residential social type grouping and the load profiles grouping,which helps to understand the electricity consumption habits of residents.The results show that compared with other methods,the k-means clustering framework has more prominent profile characteristics after grouping and is more sensitive to the slight differences between the profiles.Besides,the comparison between interval load clustering and point load clustering has confirmed that intervalization can generate new features,thereby changing the grouping.Secondly,based on the grouping results of interval clustering,this study proposes a clustering-based interval load forecasting framework for regional residents.The smart meter datasets in Ireland and London are used as prediction targets.The impact of different forecasting and clustering methods on the effectiveness of the framework are explored.Finally,based on the grouping results of point clustering,this study proposes a clustering-based probabilistic load forecasting framework for regional residents and uses the smart meter datasets in Ireland and London as prediction objects.Experiments compare different probabilistic load forecasting and clustering methods to claim the validity of the framework,especially the effect of social type grouping to improve the final accuracy.Besides,this study combines cluster tags with Deep AR to construct a multi-series load forecasting model.