Data-driven modeling of solar tower power plants based on a hybrid xLSTM-Informer model with improved convolutional neural network

Solar tower power plants have attracted considerable attention for its potential to address energy sustainability challenges. However, concentrated solar power systems are characterized by significant intermittency and variability due to strong weather dependencies. This not only complicates grid scheduling but also potentially impacts the supply-demand balance of the power system. Therefore, establishing an accurate model of a concentrated solar power system is crucial for maintaining stable grid operation and improving the efficiency of renewable energy utilization. To this end, a hybrid Informer framework is designed, combined with an extended long short-term memory framework to capture long-term features. Furthermore, we propose a multi-scale dilated convolutional network to mitigate information degradation within convolutional neural networks and strengthen the modeling of complex nonlinear dependencies among multiple input variables. Experimental results demonstrate the model's good performance, achieving normalized mean absolute error of 9.140E-02 and 1.433E-01 for the output variables, receiver heat flux and power generation, respectively. These results confirm the model's superior predictive performance and adaptability, offering valuable contributions to advancing renewable energy forecasting methodologies. Future work will focus on studying the application of the proposed model to the heliostat tracking control framework of tower solar thermal power plants.