A Study On Medium-term Power Load Forecasting In Regional Distribution Networks Based On Load Clustering And Improved XGBoost

Accurate medium-term load forecasting of regional distribution networks is an important reference value for the planning,construction and scheduling of regional distribution networks,and plays an important role in the safe and stable operation of power systems.As the structure of the regional distribution network becomes more and more complex,the changes in the distribution network load are full of variables and lack of regularity,which makes it difficult to adapt the traditional load forecasting tools to the new development trend.The stochastic nature of the load of electricity users in the regional distribution network makes it timeconsuming and resource-intensive to build a separate prediction model for each user;if a prediction model is built for the total load data of the regional distribution network,it will weaken the correlation between various types of load and time and other relevant factors,resulting in a lower accuracy of load prediction.To address the above problems,this paper proposes a medium-term load forecasting method for regional distribution networks based on a combination of k-means++ clustering and BO-TPE-XGBoost algorithms after reviewing relevant domestic and international data.The work of the paper mainly includes.1.To address the problem of noise in the electricity load data,the noise present in the load data is removed by singular spectrum analysis(SSA).Secondly,the features of the load data are selected,and finally,the regional distribution network load data are visualised to demonstrate that the time factor affects different categories of users to different degrees,making the regional distribution network load data can be better applied to the prediction model of electricity load.2.To study the rational classification of different types of electrical loads in the regional distribution network,this paper uses the K-means++clustering algorithm to classify the loads.The clustering experiments of the load data demonstrate the good clustering effect of the K-means++algorithm and the low inter-cluster similarity.3.For the difficulty in determining the parameters of the XGBoost prediction model,this paper proposes to use the BO-TPE algorithm for hyperparameter optimisation to give the BO-TPE-XGBoost prediction model.The simulation results show that the BO-TPE-XGBoost model has higher accuracy compared with other commonly used algorithms.4.In order to make up for the shortcomings of previous models that cannot build prediction models separately according to the gaps in load curves and to improve the accuracy of medium-term load forecasting of regional distribution networks,this paper proposes to use k-means++cluster analysis and BO-TPE-XGBoost to combine to build the underlying medium-term power load forecasting model of regional distribution networks,and finally to sum up the load forecasts of different categories afterwards to obtain The complete regional distribution network mediumterm load forecasting results are obtained.Taking the actual data of a distribution network in Yangzhou Hightech Zone as an example,the combined K-means++ and BO-TPEXGBoost prediction model proposed in this paper is verified to be highly accurate in the medium-term load forecasting of the regional distribution network through experimental simulation.