Prediction of Trash Rack Pollution in Run-of-River Power Plants Using Artificial Neural Networks

In run-of-river hydroelectric power plants, the level of trash rack pollution is commonly determined by pressure values measured through sensors installed before and after the intake structure. However, sensor calibration errors and external environmental factors may lead to inaccurate measurements, causing operational inefficiencies and reduced system reliability. Reducing or eliminating such inefficiencies through artificial intelligence methods presents a promising solution. This study focuses on the Karkamış Run-of-River Hydroelectric power Plant, where trash rack clogging is frequently observed. A comprehensive dataset was constructed using historical measurements retrieved from the SCADA system of the plant. The dataset includes flow rate, power output, net head, and indicators of trash rack pollution collected from relevant sensors. To predict trash rack pollution, the dataset was divided into training, validation, and test subsets. A training process was conducted using machine learning algorithms, with a particular focus on neural network models. Subsequently, the performance of these models in estimating rack pollution was evaluated and compared across different plant units using the test data. The results demonstrate that machine learning techniques, especially neural networks, are effective in identifying inaccurate sensor readings and predicting rack clogging levels. This contributes to enhanced operational efficiency and maintenance planning in run-of-river hydropower systems.