Research On Short-term Load Forecasting Of Enterprise Dc Distribution Network

In the new energy era,the traditional AC distribution network cannot meet the requirements of enterprises to use clean energy,and more new enterprises are beginning to consider using the DC distribution network for power transmission.Compared with the traditional AC transmission grid,the DC distribution network has good controllability,high transmission power quality,and low power loss during power transmission,making the DC distribution network a new direction for the development of enterprise distribution networks in the future.At present,how to predict the short-term load situation of the enterprise,realize the planned feeding from the busbar to ensure the stable operation of the enterprise distribution network,and ensure the stable operation of the enterprise production is the key technology for the enterprise to use the DC distribution network on a large scale.Therefore,The research on short-term load forecasting of enterprise DC distribution network has practical significance.First of all,this article takes the actual production situation of the enterprise’s DC distribution network as the research object,and improves its short-term load forecast accuracy as the main research purpose,reduces the interference of many factors produced by the enterprise during production on the forecast results,and improves the forecast accuracy.In order to accurately find out other factors that affect the accuracy of forecasting besides historical load data factors,power spectrum analysis and other research methods are used to conclude that similar daily characteristics have a strong influence on the forecast;at the same time,the production power consumption is standardized and modeled,using Pearson The correlation coefficient method analyzes the AC load,and it is concluded that there is a strong correlation between the use time of the motor and the load.Secondly,historical load data has missing values,abnormal values,and some noise effects.In order to reduce the error of historical load data series on the prediction results,it is necessary to repair and clean the historical load data.Repair missing data and outliers,and then perform Wavelet Threshold smoothing processing on historical load data series to reduce the random error of historical data on the prediction results.Then,use Comprehensive Learning Particle Swarm Optimization-Support Vector Regression load forecasting algorithm is proposed to predict enterprise load usage.The algorithm uses the Support Vector Regression algorithm as the prediction algorithm,and uses the Particle Swarm Optimization algorithm to adjust the parameters of the Support Vector Regression algorithm to avoid the SVR algorithm falling into the local optimum.The SVR parameters are selected in advance and no longer updated,which improves the prediction accuracy of the prediction algorithm.Finally,the simulation verification of the CLPSO-SVR algorithm shows that the MAPE value of the result is 3.92%,which is an decrease of 0.14% to 1.24% compared with other algorithms;the RMSE value is 4.18%,which is an decrease of 0.81% to 2.38% compared with other algorithms.At the same time,the load data of a small sample of enterprises has a good performance,verifying that the CLPSO-SVR algorithm can basically complete the task of load forecasting in the enterprise’s DC distribution network.