Research On Short Term Power Load Forecdiction Improving Deep Learning Based On Self-Attention Mechanism

In In recent years,the quality of China’s socio-economic development has steadily improved.With the increasing complexity of the power grid,improving the efficiency and safety of power operation has become an important research that needs to be carried out for high-quality socioeconomic development.At the same time,in the context of "carbon peaking,carbon neutrality"and Low-carbon economy under the requirements of the national development plan,short-term load forecasting of power system bears more important responsibilities.The main work content and innovation points of this article are as follows:This article first explains the key role of power system load forecasting in the power grid,and compares the current research status of load forecasting at home and abroad.Next,the basic concepts and deep learning related technologies of power load forecasting were analyzed.Deep learning technology focused on analyzing typical models and found that classical models were unable to effectively explore the relationship between historical power load data and weather data,providing a theoretical basis for the subsequent model proposed in this article.Secondly,the influencing factors of short-term power load such as temperature and date were analyzed,as well as the preprocessing methods for power load data.Standardize and normalize the load in different situations to prepare for subsequent training of the model.Species reintroduction Self Attention Mechanism,which has a good effect in Machine translation,to deal with short-term power load forecasting.At the same time,deeply analyze the structure and principle of this mechanism.Self Attention Mechanism can effectively dig out major factors such as weather and date,so as to discover better load laws.Then,according to the characteristics of load data and the advantages and disadvantages of the above mentioned deep learning model,a short-term power load forecasting model based on self attention mechanism and deep learning is proposed.First,Convolutional neural network is used to extract the features of power load data,and then the extracted feature vector is input to the Long short-term memory neural network LSTM to complete the learning of load sequence dependency,To improve the model’s ability to learn power load patterns,and then use self attention mechanism to relearn LSTM output data to obtain more power load data patterns.Finally,by comparing the designed model CNN-LSTM Self Attention with the CNN-LSTM and LSTM benchmark models through experiments on three datasets,the experimental results show that the neural network with self attention mechanism is superior to the ordinary neural network model in terms of prediction accuracy and other aspects.When the prediction error of dataset 1 is divided,the Mean absolute error(MAE)and Root-mean-square deviation(RMSE)are reduced to 12.86 and 17.43 respectively,High accuracy.