Short-Term Load Forecasting Based On Improved Spectral Clustering Algorithm And TTLSSA-LSSVM

With the continuous development of smart grid,various countries and regions are actively assembling smart meters and using data collected by smart meters to carry out a lot of research work.Short-term load forecasting is an important content of these studies,and it plays an important role in the efficient,safe and economic operation of the power systems.The load of a single user is highly random,and establishing a prediction model for each user separately will consume a lot of computational cost.It is difficult to directly establish a forecast model for the aggregate load of all users to meet the accuracy requirements,because the load of different users has different responses to factors such as weather,electricity prices,and holidays.The aggregate load will weaken the influence of these factors.This is the difficulty of user-side load forecasting.In response to the above problems,after investigating relevant domestic and foreign data,a short-term load forecasting method based on the combination of an improved spectral clustering algorithm and a least squares support vector machine model optimized by the Two types of leaders salp swarm algorithm is proposed.The following work is mainly done:1.In order to reduce the adverse effect of outlier load data on the prediction accuracy,a detection method combining Mahalanobis distance and statistical knowledge is proposed.First,the Mahalanobis distance between the daily load and the average load of each user is calculated,and then it is adaptively determined through the statistical knowledge threshold.The data in the Mahalanobis distance that exceeds k times the standard deviation of the mean is judged as outlier data.Examples show that different types of outlier load data can be effectively identified by the proposed method.2.Clustering first and then modeling is a common method for short-term power load forecasting on the user side.In order to improve the clustering quality of load data and enhance the predictability of the load sequence,an improved spectral clustering algorithm is proposed.A part of the average Hausdorff distance is proposed,which is introduced into the spectral clustering algorithm to construct the similarity matrix.The improved spectral clustering algorithm is used to cluster user load data.The results show that the clustering results of this method have better DBI(Davies-Bouldin index)and SC(Silhouette coefficient)indexes.3.In order to improve the prediction accuracy of Least squares support vector machine(LSSVM),an improved salp swarm algorithm—Two types of leaders salp swarm algorithm(TTLSSA)was proposed.TTLSSA is used to solve the problem of selecting the core parameters and regularization coefficients of LSSVM,and a short-term power load forecasting model is established.Examples show that TTLSSA-LSSVM has higher prediction accuracy than LSSVM models optimized by some common algorithms.The real electricity consumption data collected by the Portuguese power company Elergone was used to compare and test the prediction methods of this Thesis in Matlab.The results show that:(1)The forecasting method of "clustering-modeling-sum" is better than direct modeling of aggregate load.(2)The TTLSSA-LSSVM prediction model has good generalization ability and stability.(3)The prediction method in this thesis has high accuracy.