Research On Ultra-short-term Load Prediction Of Micro-grid

Nowadays,human society comes into an era of industrialization with the evolution of technology,which consumes a massive consumption of fossil energy and has led to the continuous deterioration of the environment.In order to meet the urgent needs of these problems,developing renewable energy sources and optimize the energy structure,has become a task which could not be delayed.However,efficiency of new energy use is very and it is difficult to integrate new energy into power grids.The most important way to make full use of new energy such as wind power and solar energy is to build a micro-grid vigorously and optimize the energy structure in an efficient and reasonable manner.Ultra-short-term load prediction for micro-grid facilitates efficient dispatching of micro-grid and enables optimization of micro-grid energy management.Currently,the load prediction model is generally single-point model,while the load of micro-grid has strong randomness and serious fluctuations,which make the load data of micro-grid a random non-stationary sequence that contains a large amount of uncertainties.In that case,single-point prediction cannot accurately reflect the uncertainties in micro-grid load.Therefore,it is difficult to meet the needs of the stable operation of micro-grid.Therefore,this paper focuses on studies of the accuracy and reliability of ultra-short-term load intervals prediction for micro-grid and carries out research on the accuracy and reliability of the proposed interval prediction.Which have important theoretical and practical values for the optimal operation of the micro-grid system.In order to improve the accuracy of short-term load prediction of the micro-grid,data preprocessing is needed on the original load sequence to filter out the components of the load sequence that interfere with the prediction.By analyzing the morphological theory and the constitution of the micro-grid load sequence,a preprocessing method of micro-grid load sequence based on generalized morphology filter is proposed.According to the micro-grid load sequence,the paper designs a generalized morphological filter to filter the load of the micro-grid by selecting suitable morphological structural elements and widths of morphological elements.Then a new micro-grid load interval prediction method based on recurrent neural network is introduced.Artificial bee colony algorithm（ABC）is improved by using population feedback strategy based on information feedback and optimal guidance elimination strategy,and the modified artificial bee colony algorithm is used to update the weight threshold of recurrent neural network.Finally,a micro-grid load prediction model based on MABC-RNN is established.When the load of micro-grid fluctuates greatly,the situation that the prediction interval will be too wide,which decrease the prediction accuracy.In order to further improve the accuracy of load prediction,a new criterion _MOis proposed to build a new model based on mean offset to improve the prediction interval model in Chapter 3.The traditional Coral Reefs Optimization algorithm（CRO）is improved through the improved elimination mechanism to avoid the algorithm getting into the local optimum and improve the prediction accuracy.The sample micro-grid demonstrates the MCRO-RNN model has a good performance in Ultra-Short-Term micro-grid load prediction.At last,in order to avoid the problem that penalty coefficient is difficult to choose in the single point prediction,multi-objective optimization prediction interval method for micro-grid load is proposed.Technique for order preference by similarity to an ideal solution and the grid selection strategy are introduced to modify multi-objective artificial bee colony algorithm（MMOABC）,which optimizes the RNN prediction model,improving the accuracy and reliability of micro-grid short-term load intervals prediction.