A Data-Driven Fast Calculation Method of GIL Temperature Field Distribution for Real-Time Monitoring of the Thermal Faults

In order to achieve the online analysis of the status of the current carrying structure by using the limited number of external sensors in three-phase integrated gas insulated transmission line (GIL), this paper proposes a data-driven fast calculation method for the temperature distribution with deep-learning reduced-order model, to address the efficiency issue of finite element and other numerical methods in real-time applications. This method combines a proper orthogonal decomposition (POD) with the BP neural network (BPNN) and the deep convolutional neural network (DCNN) based on U-net structure, respectively, so that the accuracy and efficiency of temperature calculation in the solid and fluid domains can be well balanced. A lower-dimensional approximate system of the temperature in solid domains is constructed by POD so that the computational scale can be reduced. BPNN is introduced to map the external sensors data of the GIL to the feature coefficient obtained by POD nonlinearly. The DCNN based on U-net structure is developed to estimate the temperature of the fluid domains by learning the feature of the solid domains, so as to obtain the overall temperature distribution. The results show that the proposed framework can rapidly and accurately predict the thermal state of sliding contact section in three-phase integrated GIL with limited external data, where the maximum relative error is less than 1.0%. The proposed method achieves an acceleration factor of 5.6 × 10³ compared with the numerical simulation software, providing an available option for the real-time visualization and digital twin diagnosis of GIL temperature distribution.