Enterprise Electricity Load Forecasting Based On Deep Learning

Short term power load forecasting is an indispensable and important work in the current development process of the power industry.It provides strong guarantees for the safe,stable,and economic operation of modern power systems,and is also an important basis for decision-making in power grid scheduling and operation.How to improve the accuracy of short-term load forecasting is a hot research topic in the current power industry.With the development of intelligent technology,deep learning algorithms are widely applied in the field of time series prediction.This article studies deep learning methods,explores the advantages of deep learning,and conducts research on short-term power load forecasting problems.This paper first analyzes the characteristics of industrial power load data with actual data as an example,explores the external factors affecting the load,and filters out the environmental variables added to the load forecasting model through correlation coefficient and maximum Mutual information coefficient(MIC).Then this paper introduces the basic concepts of variational mode decomposition(VMD),Long shortterm memory networks(LSTM),and time series convolution networks(TCN),and uses LSTM and TCN models to initially perform training and prediction tasks for CNC machine tool(equipment 1)load data,and then evaluates the accuracy and applicability of the models.Experimental results show that TCN models are better at capturing shortterm dependence characteristics,and have better prediction effects than multivariable LSTM models.This article proposes a VMD-LSTM-TCN combined model based on Variational Mode Decomposition(VMD)to further improve the accuracy of load forecasting.This paper also uses the load data of equipment 1,determines the selection of initial parameters of VMD through multiple tests,and decomposes the original load sequence into six sub sequences using VMD decomposition.Then,the prediction effects of RNN,GRU.LSTM and TCN on each subsequence are compared respectively,and the VMDLSTM-TCN combination model with the best comprehensive prediction performance is established by comprehensively considering the factors such as model accuracy and training duration.In the short-term load prediction of equipment 1 data,the RMSE of this model reached 2.4548kW,and the SMAPE was only 0.2379%.both exceeding other models.Compared with directly using the TCN model,it decreased by 31.53%and 86.26%.respectively.At the same time,the prediction qualification rate was 100%,which is in line with practical application requirements.This article also selects some combination models proposed in recent years for testing and comparison on load data of different devices.The experimental results show that compared with some existing short-term load forecasting models,the VMD-LSTM-TCN model has higher prediction accuracy in short-term load forecasting.