Research On Short-term Load Forecasting Of Microgrid

With the development of social economy and the improvement of people's living standards,energy demand has become more and more important,traditional energy sources will cause environmental pollution.Therefore,renewable energy sources such as wind,light and water are favored,and related distributed power research has been attracting more and more attention.In order to fully exploit and utilize renewable energy and distributed power sources,microgrid emerged as the times require.Load forecasting is an important part of microgrid research.Microgrid load forecasting is related to the safe economic operation of the microgrid and the regulation of stable systems.The microgrid load has the characteristics of small capacity,large volatility and strong nonlinearity compared with large regional grid load.It makes the short-term load forecasting of the microgrid more difficult.In view of this situation,this paper constructs a deep neural network model for short-term load forecasting of microgrid,extracts features by analyzing relevant factors affecting load,and then determines the input of the model,and uses the good nonlinear mapping ability of deep neural network to effectively improve the short-term load forecasting accuracy of microgrid.The number of microgrid is large,and a large number of models are generated during training,which takes up storage space and is not easy to manage.Therefore,this paper proposes a scheme to reduce the number of models,that is,multiple microgrid load data are trained to obtain a model,which can be used to predict all microgrid loads.Under the one model scheme,based on the deep neural network model,the microgrid load prediction accuracy is improved by adding the quantile feature at the input and cascading the residual block at the output.The experimental results show that the deep neural network model constructed in this paper has the lowest prediction error and the model performance is more stable than models in other literature,which indicates that the feature input and network structure selected in this paper are reasonable and more suitable for short-term load forecasting of microgrid.In this paper,nine different microgrids are used to verify the feasibility of the one model scheme.The prediction error is further reduced,the prediction result is more stable,and the prediction error does not exceed 5%,indicating that the one model scheme is feasible and has a certain meaning.