Power Load Prediction Method Based On Modal Number Estimation VMD And Improved Neural Network

Power load forecasting has an irreplaceable role in energy management and rational planning of power grids.It not only provides reference for the efficient use of electric energy,but also plays an irreplaceable role in real-time scheduling and control of the power grid.Accurate load forecasting plays an important role in promoting the commercialization and marketization of the power industry.Based on the analysis of the classification of electric load and its intrinsic and extrinsic characteristics and the evaluation index of electric load forecasting,this paper proposes a method based on modal estimation VMD and improved neural network.Firstly,in order to reduce the complexity of the original load data,the Variational Mode Decomposition(VMD)technique is introduced to decompose the load data.Compared with Empirical Mode Decomposition(EMD),it is verified that the filtering characteristics and anti-modal aliasing of VMD are better than EMD.In the selection of the decomposition mode number k,the Fourier analysis method is used to analyze the amplitude-frequency characteristics of the signal,so as to estimate the k.Building a predictive model based on VMD,and the measured data is used to test the effectiveness of the method in load forecasting.Then,for the BP neural network which is easy to fall into the local minimum value during the training process,the Firefly Algorithm(FA)is introduced to optimize the initial weight threshold of the BP neural network,and the optimized prediction model is established by combining the VMD method.The measured data of a certain area in the southern United States is simulated and compared with other methods to verify the effectiveness and superiority of the proposed method.Secondly,for the gradient disappearance problem of RNN neural network,the Long Short-Term Memory(LSTM)is introduced,and the load prediction model is established by combining with the VMD method.The measured data of a certain area in the southern United States is simulated and analyzed.It is proved that the prediction performance of the LSTM deep neural network is further improved for the feature data of the VMD decomposition.Finally,multivariate meteorological factors are introduced and subjected to dimensionality reduction screening using Kernel Principal Component Analysis(KPCA).The KPCA-VMD-FABP,KPCA-VMD-LSTM,VMD-FABP and VMD-LSTM are compared and analyzed.The comparative analysis of the examples verifies the effectiveness of the two models in this paper.And KPCA-VMD-LSTM is more effective in considering multi-factor load forecasting.