Research Of Load Curve Data Cleaning Based On Spline And Kernel Function

Load curve data refers to the electric energy consumption recorded by meters atcertain time intervals in transmitter points or user terminals. Load curve data is the“heart beat” of the power system, and also the data base of power system analysis andapplications. Two important features of smart grid--the system self-healing fromdisturbance events and the user demand side management interaction all need loadcurve data. Effective load curve data playing a key supporting role in smart gridoperation and management.Therefore, analysis of the load curve will greatly improvethe power system’s daily operation and management, system analysis, system reliability,application of the visualization, system reliability, grid energy-saving and planning level.In this paper, a novel statistical method was used to detect adverse load data, theconfidence interval was used as the bad data detection standard, and the data correctionwas based on the similar day-load curve clustering. The main contents are as follows:the identification method of nonparametric regression completely based on loadcurve data itself, which has the widespread application prospect. At first, This paperintroduces the content of identification algorithm based on B-spline function. In thelight of the phenomenon of literature method has low efficiency in dealing with massload data,This paper proposes the node vector formation strategy based on the endpointand local extremum point. That is choose the load curve part feature data points insteadof full load data to construct the vector, which can substantially reduce the size of thedata without affecting the identification effect.in order to improve the identification sensitivity on the less volatile corrected data,multiple smoothing parameter was the introduced in B-spline identification method, toimprove the bad data identification effect by weighing the fitting and smoothing degree.Secondly, this paper introduced content of identification algorithm based on Kernelfunction. This paper based on kernel density adaptive principle, introduced the partialwindow bandwidth control action, so as to improve phenomenon of incorrectidentification.In the two part of the data identification method, according to choosing geometriccharacteristics of load curve as feature points, through energy minimization methodreverse control points and curve fitting, and the introducing of confidence intervalsbetter finished identification function. The method has the ability of strong adaptability, good stability, error request and data distribution is relatively low, and can be varied bychanging the parameters to adjust the shape of the curve has a good local optimizationfunction.finally, this paper presents the data correction method based on clustering analysis,through the cluster of similar days load curve to extract normal data in similar loadcurves. Using this part of data to improve the correction effect, so as to improve theoverall quality of the load curve data.