Research On Forecasting Loads Of A Regional Integrated Energy System Based On Improved Multi-task Learning

Load forecasting is the premise of energy system precision scheduling,Accurate load forecasting can make the integrated energy system run smoothly and economically.Given the excellent performance in processing high-dimensional data,Deep learning methods have been widely used in energy systems to improve prediction accuracy.However,when forecasting the load of multi-energy in integrated energy system,there are different degrees of coupling relationship between different energy,which has a great impact on the accuracy of multi-energy load forecasting.Based on the idea of multi-task learning sharing mechanism,multi-task learning sharing layer is used to simulate the coupling relationship between different energy sources,which can improve the accuracy of multi-energy load prediction.However,the existing methods are still insufficient to explore the coupling relationship between different energy,it does not achieve high multi-energy load prediction accruacy.In view of this,this thesis studies the load forecasting based on improved multi-task learning,which mainly includes the following two contents:(1)In order to solve the coupling characteristics and load timing characteristics between different energy,a mutli-task learning method combining encoder-decoder model and sparse sharing mechanism is proposed for multi-energy load forcasting of regional integrated energy system.Firstly,grey correlation analysis is used to evaluate the coupling degree between different types of loads.Then,the encoder-decoder model based on long-term and short-term memory network is used to mine the hidden time features of multi-energy time series data.Finally,a multi-task learning method based on sparse sharing mechanism is proposed for multi-energy load forecasting.The proposed method is applied to the University of Texas and Arizona State University,and compared with existing methods.Experiments show that the proposed multi-task learning method can effectively mine the coupling relationship between different types of loads and improve the accuracy of load forecasting.(2)In the case of abnormal data in the regional integrated energy system,a multi-task ensemble method based on time pattern mechanism is applied to multi-energy load forcasting of integrated energy system.Firstly,the time pattern attention mechanism is used to assign weights to the features at different times.Then,the assigned temporal features are used as input to the multi-task learning model,a time-pattern multi task comprehensive energy load forecasting model is established.Finally,xgboost ensemble learning strategy is adopted to integrate a variety of forecasting models with different parameters to obtain the final load forecasting results.Taking the integrated energy system of the University of Texas and Arizona State University as the application object,experiments show that the proposed method can achieve high-precision prediction results in the case of abnormal data,indicating the superior robust performance of the method.The thesis has 19 pictures,12 tables,and 85 references.