Research Of Short-term Load Forecasting Under Smart Grid Environment

Under the background of the rapid development of smart grid technology and the reform of electricity marketization,users' power consumption patterns will also change dramatically.One of the change is that users can change their power consumption behavior based on their own power demand combined with real-time electricity prices or peak-valley electricity prices,which greatly increases the complexity of short-term load fluctuations.In addition,with the addition of a large number of self-used distributed power sources and electric vehicles,more and more random factors affecting short-term load forecasting.Short-term load forecasting is not only an important basis for the safety and stability of the power system and economic operation,but also an important part of power grid production planning and operation scheduling.Accurate short-term load forecasting results have important guiding significance for the start-up and shutdown of electric generating units,coordination of hydropower and thermal power and other power generation equipment,arranging short-term scheduling plans and responding to emergencies.The randomness of load fluctuations brings huge challenges to short-term load forecasting,so it is necessary to study new methods for short-term load forecasting to improve its accuracy and reliability.Based on the characteristics of short-term power load,this thesis considers the external factors that affect the load change fully on the basis of digging the underlying variation rules of load data.The main work is as follows:(1)The construction of input features in short-term load forecasting model are studied.Firstly,the processing methods of abnormal load data and missing load data and the standardized methods of improving load data are introduced;Secondly,the characteristics of load fluctuation laws are analyzed;Finally,two correlation analysis methods are introduced,and these two correlation analysis methods are used to quantitatively analyze the continuous influence factors that affect short-term load fluctuations and the correlation between historical load and current load.(2)A short-term load forecasting method based on improved particle swarm optimization to optimize LASSO and SVM coupling model is proposed.Firstly,the lag load in the SVM prediction model was screened by LASSO regression,and the input characteristics of the SVM prediction model were determined by combining the remaining influencing factors,and the LASSO and SVM coupling model was established;Then the parameters in LSVM prediction model are optimized by improved particle swarm optimization algorithm to improve the accuracy and stability of prediction results;Finally,the validity of the method is verified by actual load data of a city in southern China.(3)A short-term load forecasting method based on Attention-LSTM model is proposed.This method takes into account the advantages of long short-term memory neural networks in time series prediction,and introduces an attention mechanism to highlight input sequences that have a key impact on the prediction result.A short-term load forecasting model based on Attention-LSTM is established.The model is verified on the actual load data set of a city in southern China,and the results show that the model has high prediction accuracy and robustness.(4)The prediction accuracy and efficiency of the two short-term load forecasting models are compared in the case of different amount of training data.The results show that the IPSO-LSVM model is more suitable for the case of less training data,and the Attention-LSTM model can obtain higher efficiency and prediction accuracy in the case of large training data.