Electric power regulation and modeling of a central tower receiver power plant based on artificial neural network technique

This paper addresses an intelligent control scheme using the artificial neural network (ANN) technique for modeling and simulating a central tower receiver (CTR) plant with thermal energy storage (TES). The multilayer perceptron neural network (MLPNN) was implemented with two input parameters, one output parameter, and one hidden layer. The inputs comprise the receiver inlet temperature and the receiver thermal power. The output includes the mass flow rate of heat transfer fluid. A total of 888 datasets and three types of training algorithms, Levenberg-Marquardt (LM), Quasi-Newton, and Scaled Conjugate Gradient, are applied for training the proposed MLPNN model. Among all the investigated learning algorithms, the adopted LM with 40 neurons in the first hidden layer displayed superior ability to efficiently estimate the required performance compared to other learning algorithms. Based on the statistical evaluations, the optimized LM delivered preferred values for the root mean square error (0.0037), the coefficient of variance (0.0011), and the coefficient of determination (0.9999) during the training process. Hence, the MLPNN with LM-40 is an effective technique to precisely control the mass flow rate and subsequently the receiver outlet temperature regardless of the variations in the direct solar radiation and receiver inlet temperature. Furthermore, the generation strategy was modified to regulate CTR/TES output through electricity generation according to the hot storage system conditions. The performance of the adopted CTR-ANN model is presented and compared with the System Advisor Model (SAM) results. The results demonstrate better compatibility between the adopted CTR-ANN model and SAM outputs. The simplicity and minimum required input data of our proposed model make it appropriate for evaluating the power system reliability by using the Monte Carlo method.