Research On Short-term Load Forecasting Of Hybrid Deep Learning Model Based On AdamW Optimization

Efficient and accurate load forecasting is of great significance for improving power supply system stability,power dispatching economy and long-term power planning.Based on the research of short-term load forecasting,this paper focuses on the power data,the forecasting model and the optimization process with the help of deep learning model:(1)Power data processing and multi-feature filtering.First,the mean filling method is used to fill the missing values of power data,and the outliers are corrected by calculating the deviation rate.Then,the data were correlated by Pirsson’s product moment method,and the key features that affect load changes,such as temperature,humidity and wind speed,were selected to participate in the construction of the feature matrix.Finally,feature projects are expanded to include factors such as electricity prices,working days,holidays and urban populations.(2)The prediction framework of deep learning model.The load forecasting framework based on Adam W optimization is studied,which includes feature extraction module and Adam W optimization depth learning module.The Adam W optimized deep learning module is used to predict the nonlinear trend of load change.The experimental results show that,compared with the traditional optimizer,the loss of the framework decreases faster,the error loss is lower and the prediction accuracy is higher under the Adam W optimizer.(3)Improved Bi LSTM hybrid depth learning model.The model consists of a Temporal Convolutional Network(TCN),an improved Bidirectional Long Short-Term Memory(Bi LSTM)and Attention.Among them,TCN module is used to extract the input time series data,Bi LSTM module is used to forecast by bidirectional data mining,and Attention module pays Attention to important features.The experimental results show that the model has higher prediction performance than the traditional deep learning model,and achieves the lowest prediction error on weekdays and holidays.The results show that the hybrid depth learning model based on Adam W optimization in this paper has some practical value and application prospect for short-term load forecasting,it provides a new solution for the research of short-term load forecasting.