Abstract: The icing HHV transmission lines are high with large spans,line breakage,tower collapse and galloping are always appearing.However,few monitoring apparatuses on the towers cannot describe corresponding mechanical characteristics of all the members,it is therefore significant to improve the accuracy for pre warning of the icing lines.Taking a typical 500kV icing tower lines as a case study,a three-dimensional finite element model was established and verified through online monitoring information,and a represented model for calculating tower stress was established to quickly obtain corresponding displacement and stress of the tower under different icing conditions as the database of the mechanical response,and also,a BP neural network based on a new metaheuristic optimization algorithm named Coati Optimization Algorithm (COA) was proposed to and predict the mechanical response of the transmission towers.Four optimization algorithms were used for 23 benchmark functions to compare the evolutionary ability,fast convergence speed,and high convergence accuracy,showing the efficiency of COA to get the global optimizer for complex problems.Furthermore,the sine and cosine functions with different sample sizes were simulated,indicating that the simulating values based on the evolutionary BPNN were more accurate and had faster convergence to approach the theoretical values.And in addition,the main control factors affecting the mechanical response of the icing transmission lines and the top displacement as network inputs,and the maximum displacement,maximum tensile stress,and maximum compressive stress of the tower as network outputs,a dataset considering a wind direction angle of 45° was conducted to verify the learning generalization ability of the proposed model,and prediction comparisons with other four evolutionary neural network models showed its strong superiority.Finally,the proposed model was again applied to predict the mechanical response of the lines with a wind direction angle of 180°.The research results indicated that the predicted values were consistent with the displacement and stress of the actual icing tower lines,especially the mechanical response was in coincident with the practical variation of towers under different icing conditions,which once again verifies the rationality of the above prediction model.Therefore,the BP neural network based on a new metaheuristic optimization algorithm can realize the timeliness,efficiency and accuracy for predicting the mechanical response of the icing lines,providing a technical support for real-time health assessment and prediction of UHV towers.