Published: Građevinar 78 (2026) 8
Paper type: Professional paper
Download article (Croatian): PDF
Download article (English): PDF

Prognostic model of radial displacements of concrete arch dams

This paper presents a prognostic model for the radial displacement of a concrete arch dam based on long-term monitoring data. The model is developed using artificial neural networks (ANNs), with reservoir water level, air temperature, and groundwater level employed as input variables. The dataset covers 48 years of operational monitoring of the Piva Dam. The model was implemented in the Python environment using the TensorFlow and Keras libraries. Model performance was evaluated using the Mean Absolute Percentage Error (MAPE). The optimal model (ANN 23) achieved a MAPE of 8.34% for the training dataset and 7.73% for the testing dataset. The results demonstrate that the proposed model provides reliable predictions of the structural behaviour of the dam and represents an effective tool for supporting dam monitoring systems.

LICENCE:

This paper is licensed under a Creative Commons Attribution 4.0 International License.