Earthquake Prediction System using Neuro-Fuzzy and Extreme Learning Machine

Abstract

Knowledge of earthquake predictions is very important, especially to recognize patterns of occurrence. This paper proposes an earthquake prediction system, in the form of b-value predictions as parameters that indicate the potential for earthquakes. The methods used are neuro-fuzzy with ANFIS structure and Extreme Learning Machine (ELM). From the experimental results, it shows that the ELM method has better performance than Neuro-fuzzy with ANFIS structure.

Application of Artificial Intelligence in Predicting Earthquakes: State-of-the-Art and Future Challenges

“…ANFIS [54] predicted the occurrence and non-occurrence with low relative performance. GFCV-based method [48] showed high performance for bvalue prediction for the Indonesian region.…”

“…The ANFIS based methods [55] and [47] provided relative RMSE of 17.96% and 24.95%. The ANFIS with ELM method [58] achieved relative RMSE of 19.8% and GFCV method [48] achieved relative RMSE of 0%. These methods are considered as high performing methods according to the criteria described previously.…”

“…Rahmat et al [48], compared extreme learning machine and neuro-fuzzy based ANFIS model in prediction of b-value in the Andaman-Nicobar area. The ANFIS had four inputs and one output with five layers.…”

“…[47] Time, logarithm of mean annual rate of exceedance [57] Location and logarithm of seismic moment [58] GFCV Seismograph [48] ANFIS and ELM b-value parameters of 45 earthquakes from the USA, Canada, Japan, Mexico, and Yugoslavia were selected as a dataset. In GP, a uniformly portioned grid with defined membership functions and parameters were generated to produce an initial FIS.…”

“…Rahmat et al [48] tried to predict b-value using AN-FIS and ELM algorithm. For training, the ANFIS achieved RMSE of 0.003, and ELM achieved RMSE of 6.33 × 10 −7 .…”