Research On Clustering Analysis Method Of Power Load Data Based On Improved K-means Algorithm

With the development of intelligent power grid,a large amount of load data appears,and there is a lot of information contained in the load data,which is beneficial to the work of demand side management and load forecasting.The traditional load clustering method according to user type cannot meet the needs of power grid control.Through research,it is found that clustering the load curve can divide users reasonably.An efficient load clustering method can extract useful information to help power supply The company discovered the rules of electricity consumption and formulated corresponding control strategies to achieve safe and efficient operation of the power system.Aiming at the problem of load clustering research,there is a deficiency that clustering can not be used to analyze the load curve.Thesis studies the user load clustering method,and the main work is as follows:(1)Because the k-means algorithm needs to manually specify the number of clusters during the clustering process of the power load data,the clustering results fall into the local minimum solution.Thesis proposes a dynamic time warping histogram about k-means algorithm.Principal component analysis was used to reduce the dimensionality of the high-dimensional power load data,followed by the introduction of the histogram method to determine the number of initial clusters of load data,and then k-means calculated the distance between load curves by DTW to divide the load curves into K categories.Finally,the k-means algorithm of the DTW histogram and the classic k-memas algorithm are selected for clustering comparison.The algorithm operation time,the number of iterations,the operation efficiency,and the cluster evaluation index are compared.Based on the clustering results,the user load in the area is compared.Features are described.(2)Aiming at the influence of the initial center point on the load clustering results,a k-means algorithm based on logarithmic adaptive gravitational search is introduced in thesis Firstly,the attenuation coefficient of the gravitational coefficient in the gravitational search algorithm is improved.The parameter ? changes from small to large,so that the gravitational coefficient G changes non-linearly from large to small.Second,it is applied to the K-means algorithm to achieve the optimal clustering center position.The search makes the initial clustering center closer to the actual clustering center.Then,the k-means algorithm based on LAGSA and the traditional k-means algorithm are compared with the actual load data.Finally,the clustering results show that the k-means based on LAGSA When the algorithm performs load clustering,the number of iterations is small,the convergence speed is fast,the clustering accuracy is good,the noise immunity is strong,and the robustness is good.Experiments show that the improved algorithm not only improves the load curve clustering when processing load data clustering.It is efficient and can mine the typical daily load curves of different users in detail,reflecting the user's electricity consumption rules and characteristics.