Research On Short-Term Power Load Forecasting Based On LSTM And ResNet

Load forecasting is the basis for ensuring the balance of power supply and demand,and provides decision-making basis for the planning,dispatching and smooth operation of the power system.In 2021,there have been power outages in many places across the country,further highlighting the importance of power load forecasting for the smooth operation of the national economy.Therefore,it is of great significance to study new technologies and new methods of power load forecasting for the construction of smart grids,energy Internet of things,and new energy storage systems.According to the literature survey,this thesis divides the power load forecasting methods into traditional forecasting methods and intelligent forecasting methods.The traditional prediction method has a simple model and fast running speed,but the performance of the model depends greatly on the experience of experts,which is difficult to achieve in practical engineering applications;the intelligent prediction method is based on the new generation of artificial intelligence technology,which can automatically learn the operating characteristics and characteristics of power loads.Therefore,it can better fit the historical load curve and more accurately predict future load changes.According to the actual project requirements of the on-the-job work of the master of engineering,this paper focuses on a new method of short-term power load forecasting based on deep learning.First,in view of the nonlinearity,diversity,and time-varying characteristics of power load time series,this paper selects the Long Short Term Memory Network(LSTM)as the basic model of this research.According to the engineering practice,the data set required for power load forecasting is constructed,and the data set is analyzed and preprocessed.The short-term power load forecasting model based on LSTM is designed by using the Tensorflow framework,and the transmission state of information is controlled through the special gated structure of the LSTM network.Learn short-term features of power load data,and further optimize model and experimental parameters through comparative experiments.The results show that the LSTM-based model can automatically complete the feature extraction,classification and prediction of time series data,and can meet the demand for short-term power load forecasting.However,it is not perfect to only consider the time series characteristics of power load.In order to more accurately obtain the deep-level characteristics of power load data under the influence of multiple factors,this paper studies and designs a short-term power load forecasting model based on ResNet-LSTM.This model introduces Residual Network(Residual Network,ResNet for short),on the basis of the aforementioned LSTM-based prediction model,the ResNet feature extraction module is added,and the ResNet-LSTMbased prediction model is implemented using PyTorch framework programming.Three representative data sets are constructed according to the power load data in engineering applications.By comparing the experimental optimization model parameters,it is proved that the power load forecasting model based on ResNet-LSTM is better than the power load forecasting model based on LSTM.On the basis of the above research,the influence of temperature,rainfall,season,climate,holidays and other factors on power load forecasting is analyzed,and the power load forecasting considering multi-factor fusion is studied and designed.The prediction model of ResNet-LSTM,forming a new method of multivariate data fusion input.This method uses principal component analysis method PCA to analyze the main influencing factors of power load,realizes data dimension reduction and reduces model running time.The experimental results show that both the prediction model based on LSTM and the prediction model based on ResNet-LSTM,the performance of short-term power load forecasting has been improved to varying degrees after comprehensive consideration of multiple factors.Finally,the content and methods of this paper are summarized,the problems existing in short-term power load forecasting in this paper are analyzed,and the future research work and research directions are prospected.