Classification Method Of Power Load Users Based On Piecewise Aggregation Approximation

Electric power resources are an indispensable resource in daily life and industrial production.With the increase of power demand from all parties in the society,the complexity of the power load data in the power system has gradually increased,and the daily load peak and valley differences have continued to increase.Power consumption is concentrated in the peak time period,while the power demand in the flat valley period is small,and the contradiction between power supply and demand is sharp.A good load classification method helps ensure a balanced supply of the power system and reduces the burden of managing too many power users.Therefore,considering the characteristics of the power system at the current stage,effective classification based on the characteristics of the load curve,and carrying out load control and management work are of great significance to improve the efficiency of power supply and optimize resource allocation.The main tasks are as follows:(1)First part analyzes the characteristics of power system loads,introduces the composition of the power system loads and the load characteristics of some typical industries,and makes a preliminary discussion on the basic problems about cluster analysis.Then,the pre-processing process of load data is introduced,including wavelet threshold denoising and normalization.(2)Because the load data has high dimensionality and massiveness,the dimension reduction processing is performed on it.The piecewise aggregation approximation(PAA)is a common approximation method in the field of time series data mining.For the PAA algorithm,the same compression ratio is selected for each sample,which is the disadvantage of this method.Therefore,a novel method based on the particle swarm optimization piecewise aggregate approximation was proposed.The fitting error between the approximately expressed load and the original load curve set as the objective function,which is optimized by using particle swarm algorithm,and the minimum objective function and the position of the current optimal solution are obtained,which is the current time series the best number of segments.Experimental results show that PSO-PAA has a smaller fitting error when approximating the load curve,and can more accurately achieve the trade-off between data accuracy and dimensionality reduction than PAA algorithm.(3)The final part focuses on the difficulty of determining the number of clusters in the FCM algorithm.Based on the limitation of the similarity measure with a single distance,this paper improves the classification and proposes a classification method based on curve similarity(CS-FCM).First,we propose a clustering validity function based on improved partition coefficients to determine the optimal number of clusters for a sample.Next,for the problem of similarity measurement,a two-scale similarity measurement based on the mean and standard deviation was proposed,which comprehensively considered the distance and morphological characteristics of the load curve,and improved the shortcomings in morphological similarity.The classification of the daily load curve of Liaoning Provincial Network Company shows that the CS-FCM algorithm proposed in this paper is superior to the traditional FCM algorithm.