Research On Load Forecasting And Spatial Characteristics Of Distribution Network In Big Data Environment

In order to achieve higher planning,management,and investment requirements of the distribution network,high-precision load forecasting of distribution networks combined with big data technology has become a new development trend.On the one hand,it can provide useful guidance for load scheduling daily or weekly;on the other hand,the study of the spatial characteristics of loads can help implement the differential planning,as well as arrange power generation and maintenance.First of all,this paper proposes a dimension reduction and improved Kmeans clustering algorithm for load data.By establishing daily load indicators,the original high-dimensional load data can be converted into low-dimensional load indicators.Then the improved K-means algorithm based on entropy weight is used to cluster the load indicators.After obtaining several types of typical loads,the power consumption rules are analyzed to provide high-quality data for high-precision prediction.Secondly,this paper proposes a short-term load prediction method based on LSTM and SVR models.The LSTM model pays attention to the change law of historical load,and the SVR model takes factors such as weather and day types into consideration.By weighting the two prediction results,a higher-precision load forecast is achieved.Finally,this paper proposes a spatial load density estimation model and a medium and long-term prediction method.Through the KDE model,the typical load density of each type in the unit is obtained.Then the similarity and development level of each power supply unit are evaluated by using hierarchical clustering and entropy weight methods;The S type space load forecasting model can be used to predict the load density of various types,and then the space load evolution of the current year,transition year,and prospect year can be inferred.In the case analysis part,not only the performance of the proposed dimensionality reduction and clustering algorithm is analyzed,but also the combined prediction model is compared with some single models,which proves that the accuracy and robustness are improved.It also combines mid and long-term forecasts with geographic information to achieve evolutionary for spatial loads.