Research On Short-Term Power Load Forecasting Algorithm Based On Machine Learning

With the rapid development of the economy and the continuous improvement of people’s living standards in China,the demand for electric energy is increasing in both industrial production and daily life.Electricity load forecasting has been a hot research topic in the academic field.How to build an accurate load forecasting model and improve the forecasting performance has been the focus of research.Nowadays,with the rapid development of artificial intelligence,the e Xtreme Gradient Boosting(XGboost)and Light Gradient Boosting Machine(Light GBM)in machine learning algorithms are widely used in power load forecasting modeling,which can better handle load forecasting modeling problems with their powerful fitting and information mining capabilities.In addition,in order to improve the predictive performance of power load forecasting models,the selection of training data,parameter optimization,and the construction of composite models are mainly used to improve the predictive performance of the models.The main research content of this thesis is as follows:Firstly,the MLK(MIC-Light GBM-Kmeans)and MLXK(MIC-Light GBM-XGboostKmeans)models based on feature optimization strategy are proposed.It aims to solve the impact of a large amount of input data on the model’s learning burden and efficiency.To this end,starting from the factors that affect the load in the region,Maximum Information Coefficient(MIC)is used for the first screening of each influencing factor to select the factors that affect the load changes in the region.Then,these factors and their corresponding load data are input into the Light GBM model to construct a feature importance ranking based on the Light GBM algorithm.In order to obtain the optimal feature set,Light GBM and XGboost models are introduced to perform regression analysis on the feature variables selected by MIC and the variables ranked by importance based on the Light GBM algorithm.Meanwhile,the Kmeans clustering algorithm is used to solve the ambiguity problem of time variables,and its clustering results are used as feature variables of short-term power load instead of basic time variables to improve the learning efficiency and performance of the prediction model.Secondly,a short-term load forecasting method using the Sparrow Search Algorithm(SSA)optimization of Light GBM-XGboost model parameters is proposed.The SSA algorithm has good global exploration and local development capabilities,considering most factors in the population,and can effectively optimize model parameters.The algorithm reduces the loss of computing resources,lowers the randomness of experience settings,and can better determine the optimal parameters to improve the predictive performance and accuracy of the model.Finally,an EC-MAPE(Error Correction-Mean Absolute Percentage Error)algorithm based on error correction is proposed.The algorithm considers the influence of temperature,weather type,and basic time variables on the prediction model error,and realizes fast and accurate prediction of the electricity load data in the region by minimizing the gap between the predictioned value and the true value.At the same time,the EC-MAPE algorithm efficiently combines Light GBM and XGboost in non-linear models to solve the problem of insufficient generalization ability of a single prediction model.The algorithm can fully utilize the implicit information contained in the error,reduce the inherent error of the model,make the model more integrated and unified,reduce the interference of influencing factors on load prediction,and greatly improve the accuracy of prediction.In this thesis,the performance of the proposed model is investigated by comparing the models and metrics through various simulations,and the simulation results show that the improved Light GBM-XGboost combined model has high prediction accuracy and stability.