Research On Short-term Load Forecasting Based On Mode Decomposition Method

Short-term load forecasting plays an important role in the stable and economic operation of power grids as an important element of research in power systems.As new energy generation continues to increase,the generation structure is being optimized and the grid structure is becoming more complex,which also poses new challenges for shortterm load forecasting.Accurate prediction of short-term power load is helpful to ensure the balance between supply and demand of load and avoid unnecessary energy waste.It has great economic benefits and social value.Due to system failure and human factors,load data and other related data usually have a certain amount of missing.In this paper the missing values are filled in by using a predicative mean method and then analysing the relationship between the influencing factors and the load variation.The input characteristics of the load forecasting model are selected by quantifying the filled data through the Spearman correlation coefficient.Due to the complex relationship between load series and the influencing factors,it is difficult to predict the load accurately by using a single model and manually adjusting the load.Therefore,based on the optimization algorithm,this paper constructs the ABCSVR prediction model,BES-SVR prediction model,COOT-ELM prediction model,and AVOA-ELM prediction model.This paper first analyses the influence of the kernel function of the support vector machine on the prediction performance of the model,and then analyses the model performance from various aspects such as evaluation index and error distribution.This paper also selects the excitation function of the extreme learning machine,and optimizes the thresholds and weights of extreme learning machine(ELM)models with different numbers of implied layer neurons by COOT and AVOA,resulting in a significant improvement in the prediction performance of the model,of which AVOA-ELM had the highest prediction accuracy,and MAPE reached 1.846%.In order to further improve the prediction performance of the model,this paper used modal decomposition combined with sample entropy to decompose and reconstruct the load series to achieve the separation of different time-scale features of the load series,and then used Spearman’s correlation coefficient to quantify the correlation between the reconstructed series and the influencing factors to determine the model inputs of the three series.The model with the decomposition using improved complete ensemble empirical mode decomposition with adaptive noise(ICEEMDAN)showed a significant improvement in prediction performance compared to the model without the mode decomposition,and the MAPE reached 1.61%.In this paper,the suitability of the other two modal decomposition methods and AVOA-ELM model was evaluated,and the reconstruction error and mode mixing occurred during decomposition both had an impact on the prediction accuracy of the model.It is confirmed through experiments that the proposed method has a good prediction effect on short-term load forecasting.Figure [49] Table [22] Reference [80]...