Short-term Power Load Forecasting Method Based On Modal Decomposition And Ensemble Learning

Short-term power load forecasting is an important part of the operation and maintenance,planning and dispatching of the power industry,and provides guarantee for the safe and stable operation of the power grid and the scientific power consumption of users.Aiming at the problem of insufficient prediction accuracy of short-term power load forecasting models,by studying the characteristics of power load itself,sequence decomposition methods,and model integration methods,a short-term power load forecasting model based on modal decomposition and integrated learning is proposed to achieve accurate load forecasting.The main work content and innovations of the thesis are as follows:(1)Power load characteristic analysis and predictive modeling.The characteristics of power load under different power users and power consumption conditions are studied.From the aspects of user power consumption behavior,climate change,and grid macro-control,analyze the impact of external factors on power load fluctuations,and establish a mathematical model for short-term power load forecasting to provide theoretical support for load forecasting.(2)Research on short-term power load forecasting models based on modal decomposition.Study the load sequence decomposition method,adopt the ensemble empirical mode decomposition method,add noise signals to the original load sequence as an aid,decompose the sequence into modal components in different frequency bands,and decompose the component sequence into high frequency by judging the zero-crossing rate of the sequence The LSTM network model and the ELM network model are used to predict the high-frequency component and the low-frequency component respectively.The EEMD-LSTM-ELM prediction model based on modal decomposition is constructed,and the load data of a certain manufacturing user is selected to verify the decomposition.Predict the effectiveness of the model.(3)Research on short-term power load forecasting models based on ensemble learning.Research the basic principles of integrated learning,use elastic network regression model to enhance learning of nonlinear basis learning model,select nonlinear stochastic gradient boosting tree model to enhance learning of linear basis learning model,and use elastic network regression model output results to improve stochastic gradient The initial function of the boost tree is to integrate the two local enhancement models,build an Ensemble prediction model based on integrated learning,and perform simulation experiments.The experimental results show that the prediction accuracy of the proposed model is higher than the prediction results of the single model and the local integrated model.And it has better prediction performance than other integrated models.(4)Model comparison experiment.The proposed EEMD-LSTM-ELM model and the Ensemble model are compared with prediction experiments.The prediction results of the two models under different sample sizes are comprehensively analyzed,and the performance and advantages of the two models under different conditions are compared.