Research On Short-Term Power Load Forecasting Based On Cluster Analysis

Short-term power load forecasting plays an indispensable role in the energy management system(EMS).It plays a different role in ensuring the safety,stability and economy of the power system.This paper studies short-term power load forecasting.This article first briefly introduced some factors that affect the characteristics of power loads,the existing power load prediction methods,and the basic principles of power load prediction.Then,this paper analyzed and compared BP neural network,self-organizing competitive neural network and SOFM neural network.At the same time,in order to address the shortcomings of BP neural network's weak generalization,this paper introduced Bayesian and premature termination methods to improve BP neural network.Due to the strong regional nature of the power load,this article focused on the power load of Hohhot in 2018 and analyzed the basic characteristics of the load and used the horizontal processing method and vertical processing method to repair the missing historical power load data,This article quantified and normalized the load and related factors affecting the power load.The relevant factors include the maximum and minimum temperatures,wind,weather,humidity,and date type on the forecast day.This article quantified text information into digital information that can be processed.All data is proportionally compressed between zero and one.To prevent false saturation of neurons,this measure eliminates the effects of differences in dimensions and range of change.Finally,in order to improve the accuracy of BP network prediction,a method based on clustering network combined with BP neural network to predict short-term power load was proposed.The sorted data were input into the Self-organizing competition network and the SOFM network.In this paper,a learning mode combining unsupervised training with clustering neural networks and similar learning rules was used to perform network learning training autonomously without the need to expect output data.This article used its characteristics and rules to adjust network weights.Load classification was based on similar day characteristics.After comparing the classification results,the SOFM network was selected.Each type of SOFM network classified separately was modeled by inputting BP network prediction,Bayesian BP network,and early termination BP network prediction.This article uses MATLAB as the platform for simulation verification.By analyzing the simulation data,the conclusion of this article is obtained.The BP neural network can initially achieve the purpose of predicting the load.However,due to the fixed defects inherent in the network itself,the errors in the final network prediction are relatively large and cannot be achieved.To achieve a higher accuracy prediction effect,therefore,intending to improve the prediction accuracy,this paper improves based on the BP neural network prediction model and finds that the method of clustering and prediction is higher than that of using BP neural network to predict.And the prediction method combining SOFM and early termination BP neural network has the best prediction effect.