Research And Application Of Short-term Power Load Forecasting Based On Hybrid Forecasting Model

As the construction of smart grids advances in China,ensuring stable power dispatching in the distribution network has become a primary objective.Developing reasonable and accurate generation plans and distribution schemes is crucial to achieving this goal.In this thesis,a method based on the combination of many prediction models is suggested,which combines artificial neural network algorithms and intelligent optimization algorithms and applies them to short-term electricity load forecasting,after a thorough review of relevant literature and theoretical analysis at home and abroad.The main research contents of this thesis are as follows:(1)Data preprocessing.The data collected by the State Grid Corporation of China and local meteorological stations may contain missing or abnormal,and therefore require thorough cleaning and processing to enhance the quality of input information.Then,K-means++clustering and grey relational analysis methods are utilized to process data of similar days for short-term power load forecasting,aiming to reduce irrelevant data and improve prediction accuracy.The cluster analysis of similar day data is performed to determine similar day groups,followed by grey relational analysis to analyze the similarity between groups.The processed dataset is divided into training and testing sets in a specific proportion for model training and testing,with the goal of enhancing the accuracy of short-term power load forecasting.(2)Model Construction.A Bi LSTM model(CSSA-Bi LSTM-TPA)based on chaotic sparrow search algorithm and temporal attention mechanism is proposed.Firstly,the Bidirectional Long short-term memory(Bi LSTM)model is used to consider the forward and backward information of historical data to better capture the dependencies in the time series;then,through temporal pattern attention(Temporal Pattern The Attention,TPA)mechanism outputs attention weights to the hidden layer vectors trained by the Bi LSTM model,and weights the relevant variables at different time steps to strengthen the attention to important information in the time series;finally,using the Chaotic Sparrow Search Algorithm(Chaotic Sparrow Search Algorithm),CSSA)to find the optimal model hyperparameters.(3)Experimental Applications.This thesis utilizes two provincial-level datasets from the National Power Grid and local meteorological stations to establish short-term power load forecasting models and compare the performance of different prediction models.According to the experimental results,the root mean square error(RMSE)of the constructed CSSABi LSTM-TPA model is 5.287 and 1.117,which are significantly lower than other forecasting models,indicating that the CSSA-Bi LSTM-TPA model has higher accuracy in short-term power load forecasting and can Better solve the actual power load forecasting problem.