Study on Precursor Anomaly Data Recognition and Prediction of L-IAZPSO-SVM Algorithm

The traditional seismic data anomaly identification method usually adopts a simple threshold method to determine the data collected by the device outside the set normal monitoring threshold as abnormal data, and the abnormal data cannot be accurately identified. This paper adopts a method to improve the traditional anomaly recognition L-IAZPSO-SVM method to improve the accuracy of abnormal data recognition.In this paper, the earthquake precursor data [1] provided by China Seismic Network is used as the input index, and the improved PSO (Particle Swarm Optimization) [2] is used to optimize the SVM(support vector machine) [3] training to realize the identification and prediction of abnormal data. First Using Cluster analysis and PCA(Principal Component Analysis) [4] make the data preprocessed, and then Itô Lemma [5] is used to optimize the PSO initialization process to improve the particle variability and make the initial particles diverse. The adaptive speed update method is used to improve the problem that the algorithm is easy to fall into the local optimal solution in the later stage of the algorithm. The Ziggurat algorithm [6] is used to increase the number of particles and solve the problem of imbalance between the global search capability and the local search capability of the system. The improved algorithm solves the problem that the traditional SVM method cannot determine the number of support vectors, the system fitting fuzzy problem and improves the algorithm's abnormal recognition rate and prediction rate. Compared with the traditional PSO-SVM method, other improved PSO-SVM methods, and other anomaly data detection algorithms, the system's running speed, global search capability, operational stability, and abnormal data recognition rate prediction rate are improved.