A Reduced Order Model for Arc Heat Transfer in Oil Based on POD-LSTM

In real-time simulation applications, the full-order model often necessitates lengthy computation times, rendering it inadequate to meet the demands of real-time performance. In pursuit of establishing a digital twin for predicting the arc temperature field within transformer oil, this paper introduces a reduced-order model capable of rapidly solving the temperature distribution. Initially, a full-order model of arc heat transfer within transformer oil is developed employing the finite volume method. Subsequently, Proper Orthogonal Decomposition (POD) is employed for feature analysis, followed by dimensionality reduction based on an energy ratio criterion to capture the predominant modes. Finally, a prediction model for each primary modal coefficient is constructed using a Long Short-Term Memory (LSTM) network. The effectiveness of this approach is demonstrated through numerical simulations of a two-dimensional arc heat transfer multiphase flow model. The results indicate that the calculation outputs from the reduced-order model meet the specified error criteria. Concurrently, the computational time is dramatically reduced from hours to seconds. This research underscores the accuracy and timeliness of the reduced-order model, enhancing computational efficiency while maintaining the precision of the digital twin model.