Research On User Short-term Load Forecasting Under The Open Sales Enviroment

Accurate short-term load forecasting can ensure the normal social life and production,effectively reduce the cost of power generation,and increase economic and social benefits.With the development of society,the power sales environment has gradually been liberalized,and electricity prices have become the dominant factor affecting load changes.In addition,the power supply company hopes to interaction with customers and understand customer's power usage behavior.However,the traditional load forecasting method is no longer applicable.How to find a method that can improve the accuracy of load forecasting and reflect the user's power consumption behavior should be a direction of current short-term power load forecasting research.This paper starts from two aspects,the one is the impact analysis of the short-term electric load forecasting accuracy under the open electricity sales environment,and the other is the relationship between the short-term electric load forecasting accuracy and the user's electricity behavior characteristics classification.For the first aspect,firstly,analyzing the factors affecting the change of power load,and then the same prediction algorithm is used to calculate the load separately when considering the price of electricity and without considering the price of electricity,and the accuracy of load prediction in the two cases is compared.For the other aspect,firstly,the user's electricity behavior analysis is used to construct the user's electricity behavior sequence,then the user's typical load curve is extracted,the user is classified according to the typical load curve,and the multi-class load prediction accuracy is compared under the same prediction algorithm.Finally,using the smart meter data of Huludao area to test,based on the influence of clustering and prediction algorithm on load prediction accuracy,The k-means and BP neural networks,k-means and OS-ELM neural networks,Kohonen and BP neural networks,Kohonen and OS-ELM neural networks were established respectively.The four models were used for load prediction.The experimental results show that the method can deeply mine the user's electricity usage behavior and reveal the relationship between the number of user clusters and the system load prediction accuracy.Meet the accuracy requirements of system short-term load forecasting.