Short-term load forecasting method for power system based on key feature optimization

Abstract

Accurate forecasting of short-term power load is an important condition for the safe and economic operation of the power system. To improve the accuracy of short-term load forecasting for the power system， a short-term load forecasting method based on key feature optimization is proposed. Firstly， the construction method of the meteorological features， daily type features and historical load features affecting the short-term load of the power system is optimized， which can provide more prior knowledge for the load forecasting model. Then， considering the characteristics of the input features and the output prediction vector， a short-term power load forecasting model combining the convolutional neural network and the fully connected layer is constructed. Finally， the effect of the short-term load forecasting method for the power system based on the key feature optimization in the actual load forecasting task is validated by a numerical example. The example result shows that the key feature optimization of meteorological features， daily type features and historical load features is conducive to improving the accuracy of the short-term load forecasting for the power system.

Keywords

• feature optimization
• load forecasting
• convolutional neural network
• fully connected layer