基于深度学习融合多源数据的风电塔筒螺栓声时预测

Acoustic Time Prediction of Bolt for Wind Power Tower Based on Deep Learning and Multi-source Data Fusion

本文通过深度学习算法将多种影响螺栓状态的数据进行融合,并对螺栓声进行预测,为判断螺栓状态作准备。首先将风力机状态监测系统采集的塔筒振动、倾角、环境温度及声时信息进行数据处理,再把数据输入到CNN-BiLSTM模型中完成训练,最后通过监测数据对模型进行实验测试,并通过LSTM模型和BilSTM模型进行对比验证,结果表明:CNN-BiLSTM预测精度最高,可用于螺栓状态提前预警。

A variety of data which affect the bolt state were integrated by the deep learning algorithm,and the acoustic time of bolt was predicted.Firstly,the tower vibration,inclination,ambient temperature and acoustic time information collected by the wind turbine monitoring system were preprocessed.Then the data were input into the CNN-BiLSTM model for training.Finally,the model was experimentally tested with monitoring data,and the comparison and verification were carried out with the LSTM model and the BiLSTM model.The results show that the CNN-BiLSTM has the higher prediction accuracy and can be used for early warning of bolt failure.