Short-term Load Forecasting Based On Machine Learning

Power system load forecasting is an important foundation for ensuring safe and stable economic operation of power systems.According to the time scale,it can often be divided into long-term load forecasting,mid-long-term load forecasting,short-term load forecasting and ultra-short-term load forecasting.Load forecasting at different time scales has different effects on the power system.Short-term load forecasting is helpful to improve the utilization rate of power generation equipment and the effectiveness of economic dispatch.It is also helpful to help power system operators formulate more reasonable dispatch plans to maintain supply and demand.Balance and ensure the safety of the power grid,reduce resource waste and electricity costs.As the load connected to the power grid continues to increase and the types of electrical equipment become more and more numerous,especially the access of electric vehicles brings great challenges to the short-term load forecast of the power system.In this paper,in order to solve the problems of massive load data,increased load volatility,low efficiency of traditional methods and large resources occupied by recurrent neural networks,the JANET network is introduced for short-term load forecasting.The significance of this is that the JANET network ensures the accuracy and stability of the forecast.At the same time,it can greatly improve the training and prediction speed of the model.This article will fully prove its superiority through the single model and combined model based on JANET.In this thesis,first of all,through the comparative analysis of the advantages and disadvantages of traditional forecasting methods and artificial intelligence forecasting methods,the artificial intelligence method more suitable for short-term load forecasting is selected.Further through the comparative analysis of the advantages and disadvantages of various neural networks and applicable occasions,it is determined that this article will use the JANET network.When conducting short-term load forecasting,a large amount of raw load data and meteorological data are first collected and pre-processed to correct the abnormal data in these data,and the vacancy data are completed and normalized using manual completion methods,use the One-Hot Encoding technology to encode the week type;then,based on mutual information theory,through relevant Analysis to determine the degree of correlation between various influencing factors and load,expressed by correlation coefficients,and selecting influencing factors according to the size of the coefficient;secondly,using variational modal decomposition,the historical load data of the power system is regarded as time The sequence information is decomposed to obtain several eigenmode functions.Different eigenmode functions are used to build load forecasting models.Finally,based on the actual load data of a certain area of China Southern Power Grid,a single JANET network and a combined model based on JANET network were built and compared with BP,GRU,LSTM networks.Convergence speed,higher prediction accuracy and stronger stability.