CC BY 4.0Kraft, ThomasThomasKraftKhan, Mohammad HaziqMohammad HaziqKhanBern, GregorGregorBernPlatzer, WernerWernerPlatzer2024-11-192024-11-192024Note-ID: 00008E22https://doi.org/10.24406/publica-3761https://publica.fraunhofer.de/handle/publica/47905710.52825/solarpaces.v2i.93010.24406/publica-3761Artificial intelligence offers the opportunity to use the large amounts of data from commercial CSP power plants to supplement the experience of operations personnel through accurate predictions to optimize predictive maintenance and operations management. As a constant high outlet temperature of the solar field even under fluctuating environmental conditions is a relevant factor for the efficiency of commercial CSP power plants, the focus of this work is on the prediction of solar field outlet temperature. The analysis of this work is based on operating data of the commercial CSP power plant Andasol III in Spain with a temporal resolution of 5 minutes over a period of 5 consecutive years. To optimize the prediction, the three models random forest, feed forward artificial neural network - also known as multiple layer perceptron (MLP) - and long short-term memory (LSTM) network were compared in their performance and optimized separately by means of hyperparameter variation. The best results were achieved with the LSTM model with a mean absolute error of 6.78 K averaged over the prediction period of one year. By using AI models, future deviating outlet temperatures can be predicted at an early stage. These predictions offer the possibility to keep the outlet temperature more constant by predictive adjustment of the mass flow and thus increase the efficiency of the solar field and the whole CSP plant.enartificial intelligenceConcentrated Solar PowerLong Short-Term MemoryNeural NetworkOptimizationOutlet-temperature predictionSolar Fieldconference paper