To address the limitations of existing fiber-optic sensing technologies in capturing weak-amplitude, high-precision forward-looking vibration data streams, this study proposes a novel prediction method based on a tornado optimizer with Coriolis force optimization. This method predicts abnormal vibration signals in optical-fiber composite overhead ground wire lines. The processing flow for multiple reference signals within the TOC-CNN-LSTM algorithm is theoretically derived. The hyperparameters are initialized, followed by fitness evaluation, algorithm state updates, and iterative optimization to output the global optimal solution. Experiments are conducted on a 220-kV transmission line in eastern Inner Mongolia, with six sensors providing multi-point reference signal inputs. The experimental results indicate that the predicted outcomes for both training and test sets closely align with actual measurements, exhibiting high accuracy. Compared to traditional algorithms, the proposed method demonstrates smaller prediction errors, superior stability, and enhanced reliability in error distribution analysis. Finally, ten datasets are processed using five algorithms, with runtime, accuracy, root mean square error, and relative percent difference metrics compared. The GA-BP algorithm achieved the shortest runtime with 14.892 s, while BOA-CNN-LSTM required the longest with 85.281 s. In terms of accuracy, the algorithms achieved 82.136 %, 90.652 %, 94.901 %, 95.633 %, and 97.846 %, respectively. The TOC-CNN-LSTM algorithm yielded optimal performance for 97.846 % accuracy, with an root mean square error of 0.420 % and relative percent difference of 5.156 %. Its comprehensive superiority confirms significant engineering application potential for distributed multi-point abnormal vibration monitoring.
扫码关注我们
求助内容:
应助结果提醒方式:
