Power Angle Prediction with Embedded Physics Knowledge Network
Wenjin Jiang, Xianan Huang, Wei Lin, Xiaodong Yang, Mingfu Chen, Lingzhe Zhang · 2024
With the increase of power demand and the advancement of grid interconnection, the security and stability of the power system become especially critical. Traditional power angle prediction methods, such as ultra-real-time simulation and trend extrapolation, have limitations. In order to improve the accuracy and reliability of the prediction and enhance the interpretability of the model, this paper proposes a prediction method that combines a Long Short-Term Memory (LSTM) with an embedded physical knowledge full connectivity layer. The LSTM efficiently handles the time-series data through its gating mechanism, while the embedded physical knowledge full connectivity layer integrates the physical information. It enhances the model's understanding of the dynamic characteristics of the power system. This approach enables the model to reflect the physical constraints of the power system. It also improves the accuracy and reliability of the prediction. The LSTM network provides high-quality inputs to the physical fully connected layer by capturing long-term dependencies in the sequential data. The physical fully-connected layer then reflects the changes in the power angle in the power system through the time sequence, forming a dynamic learning model that conforms to the actual physical laws. This method of combining physical information makes each layer of the model not just a mathematical transformation, but an intelligent unit that can reflect the physical characteristics of the power system. Ultimately, this prediction model combining LSTM and fully connected layers with embedded physical knowledge improves the accuracy and reliability of the prediction. It also enhances the explanatory and transparent nature of the model, making the model's output easier to understand and accept.