Research On Short-Term Power Load Combination Forecasting Method Based On VMD And Improved CNN-LSTM

Accurate power load forecasting is the basis for ensuring the safe and reliable operation of the power system.With the continuous development of the power market,the power system is becoming more and more unstable,resulting in intermittent and non-linear changes in the power load,which makes the forecasting of the power load more difficult.At present,the challenge of power load forecasting is how to extract useful information from complex power load data and how to build accurate forecasting models.Aiming at the two major problems of fluctuating power load data and the difficulty of capturing the load variation law with a single model,this paper proposes a combined prediction and analysis method that combines data decomposition and deep learning network.The main research work is as follows:Aiming at the situation of power load forecasting,the characteristics and influencing factors of power load are expounded.First,analyze the daily,weekly,and seasonal periodicity of the power load;then,analyze the influence of various factors on the power load,including meteorological factors,economic factors,date types,etc.;finally,on this basis,fully consider various influencing factors to construct the feature set.Aiming at the problem of insufficient predictive ability of a single deep learning model,this paper combines temporal attention,Convolutional Neural Networks(CNN),Long Short-Term Memory Networks(LSTM),residual connection and Feature Attention,a deep learning-based power load prediction model is proposed,which introduces dual attention mechanism and residual connections on the basis of CNN-LSTM.First,the hybrid network makes full use of feature attention and CNN to perform feature selection and feature extraction on power load data,and at the same time,it uses LSTM and temporal attention to capture short-term and long-term temporal dependencies in load data,respectively;In addition,the model also utilizes residual connections to skip any layer in the network structure,providing adaptive network depth and complexity to adapt to a wider range of scenarios;finally,experiments are conducted on real datasets,the proposed combined model is compared with the single model without residual connection or attention.The results show that attention and residual connections improve the performance of the model.Aiming at the problem of random fluctuation of the power load due to various factors,this paper combines the data decomposition method to filter out the noise and random interference in the load data,and introduces the Variational Mode Decomposition(VMD)algorithm to decompose the original load sequence into simple subsequences with different frequencies.On this basis,a combined prediction method based on VMD and improved CNN-LSTM is proposed.First,the method uses VMD to decompose the load sequence to obtain subsequences of different periods;then,uses the CNN-LSTM combined network to predict each single component,and obtains the prediction result of the single component;Finally,the forecast results of each component are directly summed to obtain the final load forecast result.In order to verify the effectiveness of the proposed combined model,experiments are carried out on a real power load data set,and the combined model combined with VMD is compared with the combined model combined with EMD and the single model without considering data decomposition.The results show that the combined model proposed in this paper has higher prediction accuracy.