| With the rapid popularization of smart grid and the wide application of various intelligent measuring devices,a large number of power data have been generated on the user side.The application of big data technology to mine the load curve characteristics of load data,and the targeted construction of load forecasting model is of great significance to improve the accuracy of medium-term load forecasting,improve the power grid planning scheme,and ensure the safe and stable operation of the power grid.The load of the whole society mainly includes industrial load,agricultural load,commercial load and residential load.The load in different fields usually has different-load curve characteristics.It is an effective way to improve the overall load forecasting accuracy to build corresponding load forecasting model according to the load curve characteristics of users in various fields.Industrial load is the main component of the whole social load,so improving the prediction accuracy of industrial load can effectively improve the prediction accuracy of the whole social load,and then help to improve the utilization rate of power generation equipment and the effectiveness of economic scheduling.On the basis of summarizing and analyzing the characteristics of existing medium-term load forecasting methods,combined with the background of big data,this thesis comprehensively uses clustering analysis and machine learning algorithm to carry out the research of industrial user load morphological clustering based on Pearson correlation coefficient,characteristic engineering based on the characteristics of user load curve,medium-term load forecasting method based on morphological clustering and LightGBM,etc.The main work is done as follow in this thesis:(1)Based on the Pearson correlation coefficient clustering and the characteristics of industrial user load curve,an industrial user morphological clustering algorithm is proposed.This algorithm considers the influence of heavy load and light load industrial users on the total load,and the correlation of load curves among industrial users,which is conducive to improving the effectiveness of load curve clustering results.(2)This thesis proposes a feature construction method considering the law of holiday load.Combining the advantages of Filter and Embedded feature selection method,a hybrid Filter-Embedded feature selection method is proposed.This study is convenient for feature engineering processing of all kinds of user load curves after morphological clustering,which is conducive to improving the calculation efficiency and prediction accuracy of load forecasting model.(3)Combined with the above research of industrial user load form clustering based on Pearson correlation coefficient and characteristic engineering based on the characteristics of user load curve,a method of industrial user mid-term load forecasting based on form clustering and LightGBM is proposed.Firstly,the industrial user’s shape clustering algorithm is used to cluster the industrial user’s load curve,then the characteristic engineering processing is carried out for each kind of industrial user’s load after the shape clustering,and the LightGBM model corresponding to each kind of user is used to predict,finally the prediction results are obtained through the model fusion.In this thesis,1454 industrial users’ loads in a high-tech zone of Jiangsu Province are analyzed,which proves that the method proposed in this thesis has high prediction accuracy and calculation efficiency. |
