Research On Load Forecasting Method Based On Stacking Ensemble Learning

Load forecasting is an important basis for making power system planning,and its forecast accuracy is closely related to the safe and stable operation of the power system.Support vector machine(SVM)is the mainstream method for load forecasting.However,due to the influence of many factors such as climate and electricity consumption habits,the power load situation becomes complicated and difficult to analyze,which makes it difficult to select the kernel function of support vector machine.Not even applicable to the SVM with any single kernel function in a complex power system.It is proposed to apply the Stacking integrated learning method to load forecasting to provide a reliable reference for power system planning and operation.This paper summarizes the working principles of SVM,LightGBM,the improved artificial fish swarm algorithm and ensemble learning methods.The characteristics of SVMs with different kernel functions are analyzed,and the method based on Integration of SVR and Stacking is proposed.The stacking method of ensemble learning is selected to integrate SVMs with RBF,Linear,Sigmoid and Poly to construct a Stacking fusion model for load forecasting,which solves the difficulty of selecting kernel functions in complex power load forecasting.Then,the mutual harmony and complementation of based models in Stacking are studied,and the improved Stacking method of load forecasting is proposed,which further improves the forecasting accuracy of the Stacking fusion model.And the more relevant SVM with Poly is calculated through the cosine similarity algorithm.At the same time,LightGBM is used instead of SVM with Poly as the fourth based model of the Stacking fusion model by comparing from the different principle and structure of algorithms to promote the mutual harmony and complementation between the based models and effectively improve the load forecast accuracy.In the simulation process,the k-fold-cross validation method is used to train the stacking fusion model to improve its generalization ability;and the improved artificial fish swarm algorithm is used to optimize the parameters to improve the forecast accuracy of the Stacking fusion model,and the speed variable is introduced to replace the step In order to improve the convergence speed and search ability,the superiority of the proposed method is verified by comparing other stacking fusion models.From a longer-term trend,the proposed stacking fusion model is more suitable for the increasingly complex and changeable power load forecasting in the era of big data.