Research On Short-Term Power Load Forecasting Based On Functional Data Analysis

As one of the important tasks of power system planning,load forecasting is the basis of economical and safe operation of power system.Among them,the short-term load forecasting is even related to the organization of power system daily scheduling plan,open and shut down plan and load distribution plan.The accuracy of short-term load forecasting directly affects the economic benefits of grid construction and system operation.There has been a large number of studies on short-term power load forecasting methods and good results have been achieved.However,from the perspective of traditional data analysis,these methods treat the load data as discrete observations or time series,which not only loses a lot of information,but also ignores the change characteristics of the load itself.Functional data analysis has been a hot spot in recent years.It can be used to fit discrete load data into a continuously changing curve,that is,functional data.The derivative curve or differential curve is used to further consideration to mine the information contained therein deeply.Therefore,this paper attempts to predict short-term power load from the perspective of functional data analysis.In the short-term load forecasting,the consistency of the sample data is a key factor influencing the prediction accuracy.Only the sample used for modeling and prediction is sufficiently similar to the daily load to be predicted to ensure accurate and reliable prediction results.In addition,the short-term load is vulnerable to many external factors,such as climate,temperature,electricity habits,showing a typical nonlinear characteristics,so the general linear regression model for short-term load forecasting less than ideal.In this regard,this paper presents a predictive model based on functional data analysis,fitting the daily load data into functional data,and mining the inherent laws of load changes through the functional SOM and k-means clustering combination algorithm to ensure that the consistency of input sample and establish a functional non-parametric regression prediction model based on the clustering results,can well capture the dynamic and nonlinear characteristics of short-term load.In addition,when carrying out functional clustering analysis,this paper introduces the derivative distance into the similarity measure,taking full account of the shape characteristics of the load curve.The empirical results show that the proposed method can achieve better prediction results.