Optimisation of FGM Application in HVDC GIL Based on Deep Neural Network

High‐voltage direct‐current gas‐insulated transmission lines are employed due to their advantages considering safety, voltage capacity, self‐healing of the gas and reliability. However, under DC condition, high electric field stress and charge accumulation may occur, which can lead to partial discharges and therefore to insulation failures. The control of the electric field within the system is critical for ensuring operational reliability. Field control techniques are applied to regulate the electric field distribution. This work investigates the optimisation of the field control technique functional grading materials, where the spatial distribution of conductive filler particles within the spacer is adjusted to obtain a minimised and homogenised electric field distribution. Therefore, a deep neural network is utilised as a surrogate model to find the optimised distribution to keep the computational costs lower compared to conventional FEM simulations, since several simulations need to be performed for the optimisation. The results show that the DNN‐based surrogate model enables a computationally efficient optimisation of the HVDC GIL spacer.