Short-Term Electrical Load Prediction Based On Deep Learning Methods

Under the background of the construction of the national intelligent electrical network,a large number of renewable energy and new energy electrical generation projects have been put into the electrical system together.However,there are interruptions and instabilities in new energy electrical generation,and this discontinuity and instability may cause the generator set to be unable to meet the stable electrical supply demand,which greatly increases the difficulty of electrical load forecasting.Electrical load forecasting is not only an important cornerstone to ensure the smooth operation of the electrical system,but also an important prerequisite for the long-term operation planning of the electrical system.The dispatching of the electrical supply system and the maintenance arrangement of the unit are inseparable from the short-term electrical load prediction of the electricity.Improving the accuracy of short-term electrical supply load forecasting can not only lay a solid foundation for the stable operation of the electrical supply system,but also lay a theoretical foundation for electrical supply planning and dispatch planning,and promote the more stable and economical operation of the electrical supply system.Therefore,it is necessary and meaningful to carry out research on how to improve the accuracy of short-term electrical load forecasting.Firstly,the basic principle and classification of short-term electrical load forecasting were introduced.The inevitable errors in the load forecasting process were analyzed,and the evaluation indicators of the forecasting model were introduced.Analyze the practical application and research status of various load prediction algorithm models at home and abroad,combined with practical applications,the advantages and disadvantages of various forecasting methods in the field of electrical load forecasting were analyzed.Secondly,the application of deep learning in short-term electrical load prediction was analyzed.The structural concepts of CNN and LSTM of typical deep learning algorithms and their operation principles were introduced.In this paper,the attention mechanism was added to the combined algorithm model constructed,and the attention mechanism can highlight the characteristics of important influencing factors to solve the weight allocation problem between the input sequences of the algorithm model.The basic idea of attention mechanism and the calculation process of weight allocation are introduced,and the main influencing factors of load prediction and the preprocessing of data were analyzed.Then,the construction of short-term electrical load forecasting model was studied.We proposed to construct a combined neural network prediction model based on CNN,LSTM,CNNLSTM and attention mechanism.Based on the real data,the results showed that the MAPE of the short-term electrical load prediction model based on CNN-LSTM was 0.6255%,the RMSE was0.1524,the MAE was 0.1166,and the R2 was 0.9954.The model has advantages in curve coherence and detailed prediction points,and the prediction results are the most accurate.Finally,a visual interaction platform for electrical data was established.A brief introduction to the basic features of Tableau and the implications of Python’s integration with Tableau.The design of the three major steps of Python’s integration with Tableau and the design of the time interaction button and its application process were described in detail.A visualization worksheet for predicting the electrical load for the next 5 days was established,and a real total data and four model prediction data visualization worksheets were made in Tableau.Finally,through the dashboard,the historical data,model forecast data,future load forecast and other information were gathered and displayed to electrical users and uploaded to the server to achieve multi-person sharing,the platform helps users to implement fine management of electrical load data.