Research On Short-Term Load Forecasting Method Based On TCN-TPA

Electricity is the lifeblood of modern social development and the foundation of national economic development,people’s life and social stability.Electricity is not easy to store in large quantities,and to ensure the stable operation of the grid,it is necessary to keep the real-time power balance between the power generation side and the load side of the grid,which means accurate prediction of future power demand.However,as distributed power sources are integrated into the grid in large quantities,the complexity and uncertainty of the grid are increasing.To this end,this thesis proposes a "decomposition-feature selection-model prediction-integration" approach to load forecasting,and the main research is as follows:(1)A feature extraction method based on MIC-PCA is proposed.In the feature selection,not all load-related factors are strongly correlated and there is a large redundancy between different feature factors,which will increase the prediction time and decrease the prediction accuracy.To address this problem,this thesis adds the maximum mutual information coefficient(MIC)to the common principal component analysis(PCA)processing method for feature processing,which not only reflects most of the information contained in all original feature variables,but also eliminates redundant and co-linear information between feature variables,thus reducing the risk of overfitting.In addition,considering the influence of real-time electricity price and historical load on the current load value,this thesis will add these two factors to the original features for in-depth analysis,and finally a MIC-PCA-TCN model is established for validation.(2)A data decomposition method based on SSA-VMD-SE is proposed.To improve the prediction accuracy,we need to accurately grasp the various characteristics of the power load series: volatility,periodicity,randomness and nonlinearity.In order to make full use of these characteristics and improve the modal confounding and endpoint effects that occur in the decomposition of load data by traditional methods,this thesis adopts the sparrow search algorithm optimized variational modal decomposition(SSAVMD)to smooth the load series and decompose the original load into several relatively smooth modal components.Meanwhile,in order to reduce the model input complexity as well as improve the model training efficiency,the complexity of each modal component is measured by using the sample entropy(SE),the modalities with similar complexity are reconstructed,and the SSA-VMD-SE-TCN model is established for validation.(3)In order to obtain a more effective and stable prediction model,this thesis combines the advantages of each of the above two optimization algorithms and constructs a combined SSA-VMD-SE-MIC-PCA-TCN-TPA model for short-term load prediction,which not only solves the influence between different frequency series,but also completes the feature extraction better and improves the prediction performance effectively.Finally,simulation experiments are conducted with real data from a region in Australia,and the experimental results are compared with other prediction models to prove that the model proposed in this thesis has significant advantages in terms of prediction accuracy,robustness and generalization.