Application Of Improved Sparrow Algorithm To Optimize BiLSTM-RF Method In Short Term Power Load Forecasting

High precision short-term power load forecasting can not only promote the coordination of power supply,distribution,and consumption,but also improve the economy of the power market and ensure the safe and stable development of the entire economy and society.However,short-term power load forecasting models are affected by human operations,environment,weather,and other factors.The forecasting results of traditional forecasting models are not ideal enough to meet the current demand for high accuracy.In order to improve the accuracy of shortterm power load forecasting models,a two-way short-term and short-term memory networkrandom forest combined forecasting model based on improved sparrow optimization algorithm is proposed.The main research contents are as follows:(1)Aiming at the problem of low accuracy of a single prediction model,a bi directional long short term memory Random Forest(Bi LSTM-RF)combined prediction model was constructed.This model combines the advantages of bidirectional short-term and shortterm memory networks that can extract deep level features from power load data from both past and future directions,as well as the characteristics of stochastic forest networks such as high prediction accuracy and fast training speed.Through comparative experiments with several commonly used single power load forecasting models,it is verified that the Bi LSTM-RF combined forecasting model has better performance in short-term power load forecasting.(2)An improved sparrow search algorithm is proposed to solve the problems of uneven group initialization,slow convergence speed,and easy to fall into local optimum in sparrow search algorithms.Firstly,in the initialization phase of the sparrow algorithm,sparrow individuals are randomly distributed in various dimensional locations,which is prone to cause local aggregation of sparrow individuals.By introducing a Circle chaotic map to initialize the sparrow population,the randomness and diversity of the algorithm can be increased;Secondly,the frequency fluctuation characteristics of the sine cosine model are used to update the position of the enrollee to maintain the diversity of the enrollee and avoid falling into local optimization;Then,the firefly algorithm is introduced,using the photosensitive characteristics of the firefly,adding random perturbation terms,and further updating the position of the sparrow.By expanding the search area,the global search ability of the sparrow search algorithm is improved;Finally,using different benchmark test functions,a comparative experiment with five other optimization algorithms verifies that the improved sparrow search algorithm has better optimization accuracy,robustness,and stability.(3)Aiming at the difficulty of artificially setting the super parameters in the combined forecasting model,an improved sparrow search algorithm was used to optimize the super parameters in the combined forecasting model,and a short-term power load forecasting model based on the improved sparrow algorithm optimized bidirectional short-term and short-term memory network-random forest combination was constructed.In order to reduce the non periodicity and non stationarity of power load data,an improved sparrow search algorithm is first used to analyze the super parameter combination in the variational mode decomposition method [K,α] Combined optimization,which decomposes the original power load data into several harmonics,can filter out more random disturbances and maintain the characteristics of the original power load data;Then,an improved sparrow search algorithm is used to combine and optimize the internal parameters of the bidirectional short-term memory network-random forest combination model,to avoid the impact of human factors on the prediction accuracy of the combination prediction model;Finally,through comparative experiments with other three optimization algorithms and the combined forecasting model constructed by Bi LSTM-RF,it is verified that the improved sparrow search algorithm based Bi LSTM-RF combined forecasting model has better performance in short-term power load forecasting.