Short-term Load Forecasting Based On Mixed Decomposition Of Load Series And Deep Learning Model

As one of the basic industries of the country,the development level of the power industry to a certain extent reflects the degree of national economic development.With the continuous consumption of fossil fuels and the development and utilization of renewable energy,the power grid system has become more complex.In order to keep the production and use of electric energy in a dynamic equilibrium state,it is important to predict the power load in advance.Moreover,accurate short-term load forecasting is the key to ensuring efficient operation of the power system and power market scheduling planning.However,the nonlinear,non-stationary,and temporal characteristics of load series make it difficult for traditional single load forecasting models to accurately fit and predict load series.In response to the above issues,the main research content of this thesis is as follows:(1)This thesis first proposes a short-term load forecasting model based on load series mixed decomposition and deep learning techniques(A Model Ensemble patch transformation,Variational modal decomposition,Temporary convolutional networks,and Temporary pattern attention,EPT-VMD-TCN-TPA).This model proposes an Ensemble patch transformation and Variational modal decomposition(EPT-VMD)method for the non-stationary and volatile characteristics of load series.The use of Ensemble patch transformation(EPT)can accurately extract the basic trend components and residual fluctuation components of the load series;The residual wave components obtained from EPT are decomposed into relatively stable sub components using variational modal decomposition(VMD)for secondary decomposition.This model proposes a Temporal Convolutional Networks and Temporal Pattern Attention(TCN-TPA)prediction model based on the temporal and multiple influencing factors characteristics of load series prediction.In Temporal Convolutional Networks(TCN),the use of extended causal convolution and residual block structures allows for a more stable and balanced learning of nonlinear temporal relationships between load data;The temporal pattern attention mechanism(TPA)enhances the temporal attributes between multiple influencing factors and decomposed load sequences,and can add different weight coefficients to different elements in the input data sequence,thereby making the prediction model more focused on important influencing factors.The results of ablation experiments and classical comparative experiments on the regional dataset indicate that the hybrid decomposition module and prediction network module in the EPT-VMD-TCN-TPA model have a favorable impact,and the goodness of fit(R~2)of the model reaches0.9955.(2)In order to improve the operational efficiency of the model,this thesis introduces the concept of fuzzy entropy(FE)on the basis of the EPT-VMD-TCN-TPA model to calculate the complexity of the sub components obtained from VMD decomposition,Thus,a short-term load forecasting model(A Model Ensemble patch transformation,Variable modal decomposition,Fuzzy entropy,Temporary convolutional networks,and Temporary pattern attention,EPT-VMD-FE-TCN-TPA)integrating fuzzy entropy recombination component technology is proposed.This model considers that although the subcomponents are different,they have similar characteristics.By merging subcomponents with similar fuzzy entropy values,the number of subcomponents is reduced.The experimental results of region one show that the EPT-VMD-FE-TCN-TPA model has a 33.19%improvement in operating efficiency compared to the EPT-VMD-TCN-TPA model,greatly reducing the model’s running time.