海底电缆作为各类海上平台能源供给的生命线,一旦发生故障将产生巨大的经济及战略影响,准确预测海底电缆运行状态有助于提前把握其运行风险,从而实现预防性维护.在充分挖掘海底电缆运维数据中的动、静态特征的基础上,提出了一种基于注意力机制和卷积神经网络-门控循环神经网络(convolutional neural networks-gated recurrent unit,CNN-GRU)海底电缆运行状态预测方法.首先,考虑在线监测、巡检指标、静态试验三类关键影响因素,建立海底电缆运行状态评估指标体系;然后,基于改进层次分析法及多层次变权评估思想构建海底电缆运行状态评估模型;最后,建立基于注意力机制和CNN-GRU 组合神经网络模型,将历史运行参数及状态量化结果作为输入特征参量,实现海底电缆运行状态的演化趋势预测.算例分析表明,所提方法可有效预测海底电缆的运行状态,平均百分数误差低至1.04％,与全连接神经网络、CNN、CNN-长短期记忆神经网络(long short term memory,LSTM)等算法相比均具备更优的预测精度.
Prediction Method of Submarine Cable Operation State Based on Attention Mechanism and CNN-GRU Composite Network
As the lifeline of energy supply for various offshore platforms,submarine cables will have a huge economic and strategic impact in case of failure.Accurate prediction of submarine cable operation status will help to grasp its operation risks in advance,so as to achieve preventive maintenance.On the basis of fully mining the dynamic and static characteristics of submarine cable operation and maintenance data,a method for predicting the operation state of submarine cables based on attention mechanism and convolutional neural network gated cyclic neural network(CNN-GRU)was proposed.Firstly,considering the three key influencing factors of online monitoring,patrol inspection index and static test,the evaluation index system of submarine cable operation status was established.Then,based on the improved analytic hierarchy process(AHP)and the idea of multi-level variable weight evaluation,the evaluation model of submarine cable operation status was constructed.Finally,a combined neural network model based on attention mechanism and CNN-GRU were established,and historical operation parameters and quantitative results of status were taken as input characteristic parameters to realize the evolution trend prediction of submarine cable operation status.The analysis of numerical examples shows that the proposed method can effectively predict the operation state of submarine cables,and the average percentage error is as low as 1.04％.Compared with full connection neural network,CNN,CNN long short memory neural network(LSTM),and other algorithms,the proposed method has better prediction accuracy.
万方数据
维普
