Research On Short-term Power Load Forecasting Method Based On Artificial Intelligence

Electricity is an important source of energy for national development,and electricity is not only relevant to people's lives,but also to the economy of a country.Rapid economic development has improved people's standard of living and comfort,and the demand for electricity is growing.High-quality electric energy provides an important guarantee for the efficient and stable development of the whole country.In the power system,short-term power load forecasts guide the daily operations within the system.However,due to the difficulty of large-scale storage of electrical energy,the three main processes of power generation,transmission and consumption must be highly coupled together,and in recent years the country's electricity consumption has gradually increased,resulting in huge electricity consumption that can lead to power outages.The relevant components must strike a balance between the quality of electricity supply and economic efficiency,and the quality of forecasts directly affects the economic viability and reliability of each power company.In this paper,the exploration of the model is carried out based on the relevant needs of the power system and the research of previous scholars.The two-branch LSTM model and the combined model with better results are presented in this paper.In this paper,we first study the characteristics of electrical load data,and based on the data characteristics,we study how to perform data preprocessing including outlier processing,normalize the data,and construct the training set and test set using the sliding window strategy for the next step of model training and testing.Subsequently,the two-tier LSTM model,the LSTM model with the Attention mechanism and the two-branch LSTM model were proposed and validated by combining the real electric load data of a power plant in Slovakia and the real electric load data of a region in China.The experimental results of the two data sets show that the best fit to the load data and the most accurate prediction results are obtained due to the highest model complexity of the two-branch LSTM model.The LSTM model that incorporates the Attention mechanism also performs better than the two-layer LSTM by re-optimizing the parameters of the first layer LSTM model.The two-branch LSTM model and the Attention-LSTM model were 4.19% and 0.62% lower on the MAPE metric than the two-branch LSTM model,respectively,and 26.70% and 17.50% lower on Data Set 1 and Data Set 2,respectively.Finally,combining the two data sets,an average weight model and a combination model based on the Stacking idea are proposed and validated based on the implementation of the XGBoost model,LightGBM model,and CatBoost model for combining these three single models.The experimental results show that a single gradient lift tree model performs better than the LSTM-based model.On top of this,the combined model can further improve the predictive accuracy of the model and overcome the instability of a single model.The combined model and the average weighting model are 23.96% and 10.24% lower on the MAPE metric,respectively,than the simple XGBoost,and the combined model is more accurate and stable than the single model.