| Deep learning-based load forecasting with large size of historical data has been successfully used in integrated energy systems to effectively improve the prediction accuracy.However,when new users join in the system,their historical load data are often very rare,and deep learning is no longer applicable.At the same time,the reasonable distribution and utilization of coal,oil,natural gas and other auxiliary energy is the key to the operation optimization of integrated energy system.The feature recognition degree of multiple load time series data is different,and the prediction difficulty is also different.It is difficult to improve the comprehensive prediction accuracy of new users in a short period.In view of this,this paper studies the prediction strategy of the integrated energy system based on small sample data,and the research contents are as follows:(1)Small-sample Day-ahead Power Load Forecasting of Integrated Energy System Based on Feature Transfer Learning: Firstly,the feature extraction and classification models of source domain data are constructed through optimal clustering and Gated Recurrent Unit(GRU)training.Then,the trained GRU classification model is used to extract the features and category information of small samples in the target domain to be predicted.A feature fusion strategy based on feature similarity and time forgetting factor is proposed.Finally,according to the fusion characteristics,the load prediction based on transfer learning and feature input is given.The proposed algorithm is applied to the electricity load forecasting of high schools and buildings in Cardiff.The experimental results show the effectiveness of the algorithm in power load forecasting with small size of samples.(2)Small-sample Day-ahead Multiple Load Forecasting of Integrated Energy System Based on Coupling Relation Mining: Firstly,in order to solve the problem of feature extraction of small sample in multiple historical load data,the Variational Mode Decomposition(VMD)mechanism based on Particle Swarm Optimization(PSO)is constructed by taking the envelope entropy of the approximate correction of the center frequency of the eigenmode function as the fitness optimization function.Then,according to the decomposed eigenmode function with high characteristic identification degree,a mining strategy based on instantaneous frequency similarity and hysteresis factor is proposed.Finally,according to the coupling relationship between modes,the prediction models of common mode components and unique components are constructed to complete the multi-mode load forecasting of multi-element data.The proposed algorithm is applied to the comprehensive energy consumption forecasting of a region in Austin,and the experimental results show that the algorithm can effectively improve the accuracy of small sample day ahead multiple load forecasting.This paper focuses on solving load forecasting of integrated energy system with insufficient data.Firstly,this paper presents a forecasting model framework based on feature transfer learning and feature fusion for power system load forecasting.Then,in the face of system multi-modal load forecasting,a multi-mode forecasting strategy based on multi-modal coupling relationship mining is proposed.The experimental results show the feasibility and effectiveness of the algorithm.This thesis contains 20 figures,13 tables and 105 references. |
