Icing thickness prediction model of transmission line based on KPCA-INGO-BiLSTM

A transmission line icing prediction model based on an optimised bidirectional long short-term memory network (BiLSTM) and the enhanced northern Goshawk algorithm (INGO) is developed in an effort to increase the prediction accuracy of transmission line icing. The classic northern Goshawk algorithm (NGO) is enhanced by employing a hybrid sine cosine algorithm approach and a random opposition-based learning strategy in order to increase the optimisation performance of the INGO method. Based on this, the icing properties of the wire are extracted using the kernel principal component analysis (KPCA), which serves as the input for the icing prediction. Simultaneously, a two-way long short-term memory network (BiLSTM) is utilised to build an icing prediction model that takes into consideration the time-varying nature of icing and thoroughly takes into account the past and future climate features of various times. The bidirectional long short-term memory network's hyperparameter selection is optimised using the INGO method, and a transmission line icing prediction model based on INGO-BiLSTM is created. From the perspective of MAPE, MAE and RMSE, the MAPE, MAE and RMSE of the INGO-BiLSTM model are smaller than those of several other commonly used icing prediction models. The simulation results show that the INGO-BiLSTM model effectively improves the prediction accuracy compared with the BiLSTM icing prediction model, GWO-SVR icing prediction model and NGO-BiLSTM icing prediction model.