Research On Short-term Load Forecasting In Wuwei Area

Short-term power load forecasting is the key to power system operation and planning.Its accuracy can ensure the safe and stable operation of the power system,reduce power generation costs,and improve economic benefits.As my country enters the "14th Five-Year Plan" period,and the demand for electricity continues to grow,accurate short-term load forecasting becomes more and more important.In order to further improve the accuracy of short-term load forecasting,this thesis studies a combined short-term load forecasting model based on Long Short Term Memory(LSTM)and Gradient Boosting With Categorical Features Support(CatBoost).Firstly,it analyzes the power load classification and short-term load characteristics in Wuwei area,and explores the periodicity,randomness and similarity of the load,as well as the time series and non-linear characteristics of the short-term load itself.Pearson’s correlation coefficient analysis was carried out on load and continuous features in Wuwei area,and CatBoost feature importance analysis was carried out on sub-type features.The two methods were combined to extract different load features to prepare for later load forecast modeling.Secondly,due to the time series characteristics of the power load data,the LSTM short-term load forecasting model is built;considering that the load characteristics have sub-type characteristics,the CatBoost algorithm is used to build the short-term load forecasting model;the complementarity of LSTM and CatBoost is analyzed,using different The length of the load sequence is used as the input of these two models for load forecasting analysis.It is learned that the LSTM algorithm can fully learn and mine the long-period time series information of the load data,and the CatBoost algorithm has poor perception of long-term time series data;These two models perform load forecasting analysis on working days,rest days,and statutory holidays respectively,and it is known that the CatBoost algorithm can fit the classification characteristics well,and the LSTM algorithm has poor ability to fit classification characteristics.Finally,in order to complement the LSTM and CatBoost models,a fixed-weightbased LSTM and CatBoost short-term load combination model is proposed.The optimal weighting method and the reciprocal variance method are used to determine the weights of the LSTM and CatBoost models,and two different fixed-weight combination models are obtained.;Propose a combined prediction model of LSTM and CatBoost based on RBF neural network,and use RBF global optimal approximation performance to non-linearly combine LSTM and CatBoost.Use actual load data in Wuwei area to verify the effectiveness of three different combination models under different date types.The results show that under any date type,the proposed combined model can effectively combine the advantages of a single model,and on the basis of fully mining long-period time series data,it can also improve the fitting ability of sub-type features.Compared with the LSTM and CatBoost models,the proposed combined model has further improved prediction accuracy.Among them,the combination method based on RBF neural network has the best prediction effect.