Research And Implementation Of Load Combination Forecasting Method For Integrated Energy System Based On Data-driven

Under the current situation of the rapid development of the world economy,energy has always been the focus of attention of all countries.In the process of its use,problems are constantly exposed.Non-renewable energy is decreasing day by day,and the problem of low energy utilization rate is increasingly acute.To solve this problem,the world began to build integrated energy systems.The integrated energy system is closely connected with various energy forms to realize complementarity and mutual benefit among different energy forms,so as to improve the comprehensive utilization efficiency of energy.Accurate load prediction is indispensable to the dispatching and operation optimization of different energy forms in integrated energy system.In this paper,the integrated energy system as the research object,the data driven method to analyze the characteristics of multiple loads,comprehensive consideration of the coupling between multiple loads,temperature,humidity,wind speed and calendar information and other factors,the integrated energy system of multiple loads short-term forecast method is carried out.The main research contents are as follows:(1)In view of the complex original sequence of electric load in integrated energy system,which affects the prediction effect,and a single prediction algorithm can not get accurate prediction results,an electric load prediction method based on CEEMDANSE-GM-LSTM was designed.Firstly,CEEMDAN-SE was used to decompose the original sequence of electrical load and recompose multiple subsequences with obvious differences in complexity and linearity.Then GM or LSTM is selected for prediction of these sub-sequences,and the predicted values of decomposed and reconstituted subsequences are superimposed through the BP neural network layer to obtain the final predicted values.The obvious superiority of CEEMDAN-SE-GM-LSTM model in predicting electrical load is proved by an example.(2)There is a strong coupling between multiple loads in a integrated energy system,so it is necessary to analyze the characteristics of multiple loads when making multiple loads prediction.In this paper,the maximum information coefficient(MIC)was used to analyze the correlation between multivariate load and influencing factors,and the multivariate load prediction model based on CEEMDAN-SE-Prophet-LSTM was built.Finally,by comparing the prediction results with other combination models through specific examples,it is proved that the CEEMDAN-SE-Prophet-LSTM model constructed in this paper is more stable in the multivariate load prediction process,and the prediction accuracy is improved by about 20% compared with CEEMDAN-SE-GMLSTM.(3)On this basis,a set of multi-load short-term forecasting application system for integrated energy system is developed.The application system implements the modules of user authority management,data preprocessing and load forecasting,and completes the functions of algorithm packaging and encryption,off-line training and on-line forecasting.Finally,the software test results show that the design function and performance requirements are achieved.