Investigating a hybrid extreme learning machine coupled with Dingo Optimization Algorithm for modeling liquefaction...

Area of Science:

• Geotechnical Engineering
• Earthquake Engineering
• Computational Intelligence

Background:

• Soil liquefaction, a significant earthquake hazard, causes catastrophic ground failure in loose, saturated soils.
• Accurate prediction of liquefaction resistance is vital for hazard mitigation, risk assessment, and geotechnical engineering advancements.

Purpose of the Study:

• To introduce a novel hybrid model, Extreme Learning Machine with Dingo Optimization Algorithm (ELM-DOA), for estimating strain energy-based liquefaction resistance.
• To compare the performance of the ELM-DOA model against conventional methods like ELM, ANFIS-FCM, and ANFIS-Sub.
• To evaluate the impact of data pre-processing techniques (linear vs. non-linear normalization) on model prediction accuracy.

Main Methods:

• Development of a hybrid ELM-DOA model for liquefaction resistance prediction.
• Comparative analysis with existing models: Extreme Learning Machine (ELM), Adaptive Neuro-Fuzzy Inference System with Fuzzy C-Means (ANFIS-FCM), and ANFIS-Sub-clustering.
• Application of two data pre-processing methods: traditional linear and non-linear normalization.

Main Results:

• Non-linear normalization significantly improved prediction performance across all models by approximately 25% compared to linear normalization.
• The ELM-DOA model demonstrated superior accuracy, achieving the lowest RMSE (484.286 J/m³), MAPE (24.900%), MAE (404.416 J/m³), and the highest R² (0.935).
• A Graphical User Interface (GUI) was developed for the ELM-DOA model to enhance practical applicability for engineers and researchers.

Conclusions:

• The proposed hybrid ELM-DOA model, enhanced by non-linear normalization, offers a highly accurate and effective tool for assessing soil liquefaction resistance.
• The developed GUI facilitates user-friendly access to the model's predictions, improving its practical utility in geotechnical engineering.
• The study underscores the potential of hybrid intelligent models and advanced data pre-processing for mitigating earthquake-induced liquefaction hazards.