OPGW anomaly monitoring based on fiber Bragg grating has been widely studied, but anomaly detection data is difficult to classify and identify whether it is a normal signal or a false signal. Aiming at the above problems, this paper studies random forest classification based on improved grid search method. Firstly, a classification and recognition model is built theoretically. The optimal classification solution is found iteratively by particle swarm optimization algorithm. Then, the monitoring data is divided into multi-frame training set, test set and verification set to realize classification and recognition of abnormal vibration by traditional random forest algorithm, grid search random forest algorithm and improved grid search random forest algorithm. Finally, the accuracy and precision of identifying abnormal signals are used to quantify the comparison results of the three algorithms. It is proved that the improved grid search random forest algorithm studied in this paper can achieve 98.56% accuracy of abnormal signal recognition in test set and 99.56% accuracy of abnormal signal recognition in verification set, which proves the effectiveness of the method and has practical significance for classification and recognition of OPGW optical cable abnormal vibration.