TY - JOUR
AU - Wang, Feng
AB - Ultra-weak fiber Bragg grating (UWFBG) array-based distributed acoustic sensor (DAS) is increasingly applied to vibration sensing in transmission lines, transportation, perimeter security and many other fields, it can be also combined with machine learning for mode recognition of vibration events. Random forest (RF) is a typical machine learning model, which can realize regression or classification prediction of events by constructing several decision trees. However, it is difficult to configure the optimal hyperparameters for RF model and it is very easy to overfit because of an imbalance in the data set, so we choose to use Bayesian optimization (BO) for hyperparameter optimization. This paper shows that BO can improve the recognition effect of the RF model when using UWFBG-array-DAS for vibration sensing. Specific steps are as follows: First, we set up the DAS system and collect several original vibration signals. Second, we preprocess the data, including IQ demodulation and eigenvalue extraction, after that we can get the characteristic values of vibration signals of total six vibration modes. Then we train the RF model and optimize its hyperparameters using BO method. In the end, we use the optimized RF model to recognize the vibration signals detected by the DAS and evaluate the various performance of the RF model through several index parameters. The final conclusion of our research is: Compared with the unoptimized RF model and other machine learning algorithms, the average recognition accuracy of the optimized RF model reaches 95.2%, making an increase of up to 17.6%.
TI - Vibration recognition of UWFBG-array-DAS based on Bayesian optimization random forest
JO - Proceedings of SPIE
DO - 10.1117/12.3055453
DA - 2025-02-25
UR - https://www.deepdyve.com/lp/spie/vibration-recognition-of-uwfbg-array-das-based-on-bayesian-Qy34JdXiKD
SP - 135420U
EP - 135420U-8
VL - 13542
IS - 
DP - DeepDyve
ER -