16:30 〜 16:45
[HRE12-10] Multi-Component Analysis for Silica Scale Prediction Using Neural Network Architecture
キーワード：Geothermal, Silica scaling, Machine Learning, Neural Network
Scaling is a persistent challenge in geothermal energy extraction, affecting production wells, separators, pipelines, reinjection wells, and even the formation around the reinjection wells. Due to the complex mechanisms governing scale formation, a universal prediction and mitigation method has not been established. This study aims to develop a predictive model for silica scale formation rates using a multi-component approach that integrates geothermal water chemistry, polymerization kinetics of silicic acid, and the saturation indices of relevant minerals.
The research was conducted in three stages: (1) onsite experiments at geothermal power plants, (2) laboratory analyses to quantify polymerization of silicic acid and its adsorption behavior, and (3) the development of a predictive model using a Dense Feed-Forward Neural Network (DFFN). Onsite experiments included studies for polymerization of silicic acid and its adsorption on a silica gel surface, and metal plate immersion tests. Spectrophotometry and ICP-OES analyses determined monomeric and total dissolved silicic acid concentrations, while XRF and LA-ICP-MS analyses quantified scale deposition on metal surfaces.
The developed machine learning model achieved an RMSE of <15%, demonstrating strong predictive performance. Additionally, the model provided quantitative insights into the contributions of various input parameters, offering a valuable tool for scale mitigation strategies. This study is expected to contribute to improving geothermal energy utilization by enhancing predictive capabilities and supporting operational efficiency.
The research was conducted in three stages: (1) onsite experiments at geothermal power plants, (2) laboratory analyses to quantify polymerization of silicic acid and its adsorption behavior, and (3) the development of a predictive model using a Dense Feed-Forward Neural Network (DFFN). Onsite experiments included studies for polymerization of silicic acid and its adsorption on a silica gel surface, and metal plate immersion tests. Spectrophotometry and ICP-OES analyses determined monomeric and total dissolved silicic acid concentrations, while XRF and LA-ICP-MS analyses quantified scale deposition on metal surfaces.
The developed machine learning model achieved an RMSE of <15%, demonstrating strong predictive performance. Additionally, the model provided quantitative insights into the contributions of various input parameters, offering a valuable tool for scale mitigation strategies. This study is expected to contribute to improving geothermal energy utilization by enhancing predictive capabilities and supporting operational efficiency.