Short-term Load Forecasting Based On Machine Learning Algorithm

Short-term load forecasting is the basis of power system operation and planning.Accurate load forecasting can ensure the safe and stable operation of the power system,reduce power generation costs,and improve economic efficiency.With the development of the power industry and the increase of distributed energy sources,short-term load forecasting becomes more and more important.Due to the characteristics of load data,such as non-linearity,heteroscedasticity,and non-stationarity,the difficulty of short-term load forecasting has also increased significantly.Therefore,the machinelearning algorithms with strong nonlinear learning capabilities can provide good technical means for this research.Among many machine-learning algorithms,artificial neural networks have the advantages of self-learning ability and generalization ability.They have been widely used in short-term load forecasting and have achieved relatively satisfactory results.In the field of load forecasting,the BP neural network method is currently a more mature load forecasting method.Thus,we mainly use the BP neural network method for short-term load forecasting research in this paper.Generally,Power load data has some periodicities such as weekly periodicity,monthly periodicity,and annual periodicity.Meanwhile,it also shows some randomness caused by various external factors such as temperature,weather,holidays,and user behavior.The uncertainty caused by such randomness greatly increases the difficulty of short-term load forecasting.Considering the load's own change law and the influence of external factors,we establish the point prediction model and the interval prediction model starting from the aspects of forecast accuracy and reliability in this paper.The optimized point prediction model improves the accuracy of the load prediction effectively.In this paper,we focus on the point forecast and interval forecast of power load,and study the related aspects of short-term load forecasting problem with real data.The main research contents include:(1)Considering external factors such as temperature,weather,and date type,two types load point prediction models of BP neural network with different structures are established,which are a multiple-input multiple-output model and a multiple-input single-output model,respectively.(2)To solve the problem that BP neural network is slow in convergence and easy to fall into local minimum,we propose two methods to optimize the above two models by using wavelet transform(WT)and improved particle swarm optimization(IPSO)algorithm.The two established model called the WT-BP neural network point prediction model and IPSO-BP neural network point prediction model,respectively.The simulation results show that the two optimized prediction models can improve the accuracy of load point prediction.(3)Aiming at the problem of uncertainty in load forecasting,we simplify the prediction interval satisfaction indicators(PISI),and use IPSO algorithm to search for the optimal scale factors in this paper.By combining the scale method(SM)with the BP neural network,we establish the IPSO-SM-BP short-term load interval prediction model.The example results show the proposed model can achieve better performance of interval prediction.