Day-ahead Load Forecasting Of Microgrid Based On GRU Network

The accurate forecasting of microgrid load is significant for ensuring the system safety,operation stability and lower cost.Because the load volatility and randomness of the micro-grid system are greater than that of the traditional grid,it is more difficult to accurately predict the microgrid load.According to the load characteristics of the micro-grid system,this paper studies the prediction of the day-ahead power load of the microgrid based on the GRU neural network:(1)An online algorithm for day-ahead load forecasting of microgrid with GRU neural network is proposed: due to the over-reliance of traditional machine learning algorithms on the selection of samples of similar days and the inability to update the constructed model in time,the prediction accuracy is difficult to be reached.In view of the advantages of GRU in dealing with time series data,we propose an online accurate load prediction method with GRU neural network.Firstly,based on GRU network,the framework of micro-grid load forecasting is constructed.Then,the prediction model of GRU network is established,and the method of online updating the model is given.The experimental results show that the algorithm has a good prediction accuracy and can also performs well in tracking and forecsting the changed load.(2)Fuzzy similar days and multiple GRUs based forecasting on micorgrid day-ahead load with intervals is further presented: Considering the fuzzy impact of weather on load,we further put forward a forecasing method with fuzzy similar days and multiple GRUs based on work(1).Firstly,this study analyzes the volatility and the meteorological correlations of the micro-grid load,and selects those most important factors that have a great influence on the load changes.The fuzzy quantitative representation of the meteorological factors is then presented,and the fuzzy similar day selection strategy is developed.Finally,a load interval prediction model with multi-GRU neural network and fuzzy similar days is constructed.The simulation results show that the algorithm performs well in the predicted interval with larger area coverage and smaller average width of microgrid load prediction.This thesis contains 26 figures,8 tables and 82 references.