| Title | Application of machine-learning for the prediction of formation rate of silica scale from geothermal water |
|---|---|
| Authors | S. Juhri, K. Yonezu, T. Ryunosuke, K. Manaya, E. Watanabe, K. Mori, S. Sato, N. Inoue, H.E. Wibowo, T. Yokoyama |
| Year | 2024 |
| Conference | New Zealand Geothermal Workshop |
| Keywords | machine learning, artificial intelligence, silica, scaling, geothermal |
| Abstract | Utilization and development of geothermal energy are still hindered by scaling and corrosion to this day. Scaling is particularly troublesome as it can occur in almost all locations in geothermal power plants, i.e., in production wells, on surface facilities, in reinjection wells, and even in the reservoir rock formation around the reinjection wells. However, there is no single and universally applicable mitigation method to date, due to the unique characteristics of each geothermal water and the complexity of scale formation. This study aims to better understand and quantify silica scaling assisted by artificial intelligence (AI), i.e., supervised machine learning (SML). The SML was used to predict the formation rate of silica scale based on the physicochemical characteristics of geothermal water and the kinetics of polymerization of silicic acid in geothermal water and adsorption of silicic acid on the surface of scale substances. These data were used as input parameters in the training data. The data was obtained from onsite batch experiments in several geothermal power plants in Japan. The rate of silica scale formation from onsite plate immersion experiments using stainless-steel plate was used as the output parameters. This experiment was conducted for up to 5 hours in the corresponding geothermal power plants. The produced models are evaluated based on their percent relative root means squared error (%RRMSE) value when used to predict unknown data, i.e., non-training data. Our study showed that the model can achieve %RRMSE value of 15 which is in the good category, i.e., the model can accurately predict the formation rate of silica scale within 5 hours. Furthermore, the model will be trained to predict the formation rate of silica scale for up to 5 days and beyond. This study is expected to aid geothermal power plants to mitigate the silica scale problem more efficiently. |