Research On Load Clustering And Load Probability Forecasting Method Based On Hierarchical Load Data

In recent years,the scale of power users in China has been expanding.The load data of those users have also gradually collected.Using these user load data and network topology data from different levels to deeply mine the change characteristics of power load and accurately predict the load change at all levels is highly appraciated.Thus,based on the hierarchical load data in power system,load clustering and load forecasting of power system are investigated in this paper,and mainly completes the following work:1.Aiming at the problem of substation load clustering,a global multi-objective clustering model with standard deviation and fixed weight based on pso-k-means is proposed in this paper.Firstly,the k-means algorithm is used to cluster the 10 kV user load belonging to the substation to obtain the substation load composition,and then the substation is clustered according to the load curve and load composition to obtain the cluster center and cluster class number of the substation and its subordinate user load.Then,taking the number of clustering centers of the substation and its subordinate users as the optimal number of clustering centers,and comprehensively considering the three characteristics of substation load curve,load composition and load curve of users,a global multi-objective clustering model with standard deviation weight based on pso-k-means is established to cluster the global 10 kV user load and substation load at the same time.Then the PSO algorithm with gravity factor is used to solve the clustering problem to improve the convergence of the model.Both test and practical examples show that the multi-objective clustering model can integrate the load information of substation and its subordinate users,so it can get more accurate classification.2.Aiming at the problem of substation load forecasting,a bottom-up substation probabilistic load forecasting method considering load correlation is proposed.Firstly,one-dimensional CNN network is used to predict the load of medium voltage distribution transformer to obtain the mean and variance of medium voltage load distribution,and then the mean and variance of medium voltage load are accumulated to obtain the mean and variance of 220 kV high voltage load.When accumulating the variance of medium voltage load,considering the influence of the correlation of each medium voltage load on the accumulation of square difference,a correction term is introduced to eliminate the interference of the covariance of load between two lines and the accumulation of square difference,so as to obtain a narrower load range.Practical examples show that the proposed algorithm has higher accuracy and stability than the traditional algorithm,and can get a narrower probability interval.3.On the basis of the previous work,considering the consistency of layered load accumulation,a layered probability prediction method based on layered weighted accumulation consistency is further proposed.Firstly,the orthogonal-input-GRU network is used for the modeling and prediction of 220 kV and 110 kV voltage loads,and the one-dimensional CNN modeling and prediction of 10 kV load is still used in the previous chapter.In order to make the load forecasting results of each layer meet the cumulative consistency,the correction quantity is introduced into the load forecasting results of each layer,and then the optimization model is established with the cumulative consistency of the load forecasting results of each layer as the constraint and the weighted sum of the load correction quantity of each layer as the objective function.According to the correction value calculated from optimization model,the load forecasting results of are revised.The results shows that the proposed method can improve the accuracy of each layer’s load.