基于LSTM神经网络的输电线路脱冰跳跃动力响应预测模型研究

Research on the Dynamic Response Prediction Model of Transmission Line Shedding Jumps Based on LSTM Neural Network

中文关键词:输电线路  长短期记忆神经网络  深度学习  脱冰跳跃

keywords:transmission lines  LSTM  deep learning  ice-shedding jump

中文摘要:为了有效预测输电线路在覆冰脱落状态下的动态响应，并可靠评估输电线路的稳定性与安全性，结合有限元模拟与深度学习技术，构建了一种能够准确预测导线脱冰跳跃位移时程的长短期记忆（long short-term memory，LSTM）神经网络模型。该模型利用时序数据训练，实现了输入工况参数与输出动态响应之间的非线性映射，进一步结合仿真验证了模型在预测精度和泛化能力上的优势。首先，基于正交实验设计与有限元仿真生成多因素数据集，并对数据进行标准化预处理，然后，通过LSTM深度学习模型建立导线跳跃位移时程的预测模型。结果表明，该模型能够有效预测实际工程中多因素耦合作用下的位移时程，为输电线路覆冰脱落灾害的安全评估和风险预警提供了一种高效可靠的解决方案。

Abstract:To effectively predict the dynamic response of transmission lines during ice shedding and reliably assess their stability and safety， this paper integrates finite element simulation with deep learning technology， developing a long short-term memory （LSTM） neural network model capable of precisely forecasting the displacement time-history of conductor ice-shedding jumps. Initially， a multi-factor simulation dataset is generated based on orthogonal experimental design and finite element analysis， followed by standardized data preprocessing. Subsequently， an LSTM deep learning model establishes a predictive framework for conductor jump displacement time-history. This model leverages time-series data for training， achieving a robust nonlinear mapping between input operating conditions and output dynamic responses. Furthermore， simulation results validate the model’s superior prediction accuracy and generalization capability. The findings demonstrate that this model effectively forecasts displacement time-histories under the complex coupling effects of multiple factors in practical engineering scenarios， delivering an efficient and reliable solution for safety assessment and risk warning of transmission lines against ice-shedding disasters.