Research On Short-term Load Forecasting Based On Similar Day Clustering And Improved DRESN

Short-term load forecasting is an indispensable foundation for ensuring the safe and economic operation of power systems.It is used to formulate daily or weekly plans,including generator set control,load distribution,and hydropower scheduling.It plays a role in the energy distribution and management of modern power systems.Plays a vital role.However,with the development of the power system,the factors that affect the power load are becoming more and more complex,and the traditional prediction model is difficult to adapt to the characteristics of modern power load.In this paper,aiming at the change law and characteristics of modern power load,with the aim of improving the short-term power load forecasting accuracy,a load forecasting model based on similar day clustering and improved DRESN is established.First of all,based on the current problem of many factors affecting the power load and the huge amount of data,a similar daily cluster analysis method considering the influencing factors is established.The Pearson correlation coefficient is used to analyze the correlation of load impact factors.The selected influencing factors and load data are reduced by dimensional analysis using principal component analysis,and then the new data set is processed by fuzzy C-means clustering to load The data is classified.At the same time,its effectiveness is verified by comparison with traditional fuzzy C-means clustering.The results show that the improved algorithm can accurately select the number of load categories and has a faster convergence speed.Secondly,for the problem of unstable performance of the ESN in load forecasting,a load forecasting model of the improved beetle whisker algorithm to optimize the dual reserve pool echo state network was established.First,the structure of the reserve pool is optimized as a double-reservoir echo state network(DRESN).Secondly,elite strategy,Levi’s flight and adaptive strategy are introduced into the longhorn search(BAS)algorithm,and the improved longhorn search algorithm is used to optimize the dual The network parameters of the echo state of the reserve pool have been verified by examples to verify the superiority of the model.Finally,this paper combines the above methods to establish a complete short-term power load forecasting model,that is,a similar daily clustering with consideration of influencing factors combined with an improved beetle whisker algorithm to optimize the dual reserve pool state network forecasting model.According to the actual data in the past three years in Xinyang area,an example simulation was carried out in MTALAB and compared with the four models of BP,ESN,BAS-BP and BAS-ESN respectively.The results prove that the prediction scheme proposed in this paper has good prediction performance.