Surrogate model for torsional behavior of bundle conductors and its application

By means of numerical models for torsional behavior of bundle conductors under distributed torque, parameter study of torsional behavior under various structural parameters, including conductor type, bundle number, bundle spacing, span length, height difference, initial tension in sub-conductors, spacer number and arrangement of spacers, is carried out. With the over twenty thousand numerically simulated samples a dataset is set up. A surrogate model for the prediction of torsional behavior of bundle conductor lines is constructed by means of the BP neural network, in which the structural parameters are set as the input variables and the collapse torque and torsional stiffness the output. The dataset is divided into two subsets, the training dataset and the testing one. High prediction precision of the surrogate model is achieved through training with the training dataset and verified by the testing dataset. Combining the surrogate model and the particle swarm optimization (PSO) method, optimizations of spacer arrangement to obtain maximum collapse torque and torsional stiffness of bundle conductor lines are carried out.