Research On Short-term Load Forecasting Under Smart Distribution Network

With the rapid development of smart grid,the gradual access of large-scale new energy electric vehicles and the continuous reform of power market,it makes the random factors affecting the load change increase and the load volatility become more and more complex.Short-term load forecasting is of great significance for the safe,stable and economic operation of the power grid.Facing the new characteristics of load in smart grid,the traditional load forecasting methods are no longer fully applicable,and it is necessary to explore new methods for improving the accuracy and reliability of short-term load forecasting.Therefore,combined with the load characteristics of smart grid,this paper studies the short-term load forecasting of smart distribution network from two aspects of load forecasting influencing factors and load forecasting algorithm.Firstly,the load characteristics of smart distribution network and the factors influencing load variation are analyzed.Then,the load and meteorological data used in the prediction are identified and preprocessed,and the time feature sequence is established according to the time information.Finally,Pearson correlation coefficient and maximum information coefficient are used to analyze the correlation between time and meteorological factors,and external factors with strong correlation for load are selected to participate in training for improving the accuracy of load forecasting.Then,a load forecasting model based on EMD-GRU is proposed to address the problem of low accuracy in single prediction for the fluctuating electric load.Firstly,the change of power load is regarded as the superposition of various basic characteristics,and the original load data are decomposed by empirical mode decomposition(EMD)to obtain finite components with intrinsic mode function(IMF).Then,a multilayer gated recurrent unit(GRU)is established for predicting these components,and the attention mechanism is added to the network to assign larger weights to the critical information features in the historical data.Finally,the predicted results employed to the proposed method are compared and analyzed with those of the traditional direct prediction methods using the actual load data in Guizhou,the results show that the proposed method has better prediction accuracy.Finally,an EMD-CNN-GRU deep learning model is proposed that takes into account multiple factors affecting the load variation of smart distribution networks.Firstly,the EMDCNN-GRU model is improved by adding convolutional neural network to fully extract features from a large amount of input historical data.Secondly,the conventional factors including time and weather,real-time tariff factors under demand-side response are considered respectively.Actual load data are used for example analysis.Finally,through the error comparison and impact analysis of component prediction,daily load prediction,improved model prediction and load superposition prediction,the results show that the accuracy of load prediction taking into account the external conventional factors and tariff factors is better than that considering only the load itself.Moreover,the influence of electric vehicle load forecasting results on distribution network load is analyzed,and thus the corresponding measures can be taken in advance to reduce the peak-valley difference of distribution network.