| Popis: |
Accurate prediction of corrosion rates is crucial for preventing infrastructure failures, reducing maintenance costs, and ensuring operational safety. Traditional models often struggle to account for the complex, non-linear interactions between environmental factors and material properties. This study presents a novel approach integrating Simulated Annealing (SA) with an Adaptive Network-based Fuzzy Inference System (ANFIS) to improve corrosion rate predictions for pipeline steels. The SA-ANFIS model features six input neurons representing temperature, H₂S pressure, CO₂ pressure, salinity, moisture content, and material type. These factors influence corrosion rates, represented by a single output neuron. The SA algorithm optimizes the ANFIS model's parameters, enhancing its ability to handle non-linear relationships. Historical corrosion data for P110SS, L80, and 2205 Duplex steel were used, incorporating environmental variables such as temperature, pH, and gas pressures. The SA-ANFIS model achieved superior accuracy, with a maximum error of 2.8424 % and an average error of 1.2536 %, outperforming the GA-ANFIS model and conventional ANFIS and SVR models. The SA-ANFIS model offers a robust, optimized tool for predicting corrosion in petroleum pipelines, significantly improving prediction accuracy under harsh conditions. |
