The multivariable inverse artificial neural network combined with GA and PSO to improve the performance of solar parabolic trough collector

Abstract

This work focused on presenting a multivariate inverse artificial neural network (ANNim) by developing two functions coupled to metaheuristic algorithms to increase a parabolic trough collector (PTC). This work aims to provide a new method capable of improving the thermal efficiency of a PTC by determining multiple optimal input variables. At first, two ANN models carried out to predict the PTC thermal efficiency (ηt), validated, and compared in detail. For that, six input parameters rim-angle (φr), inlet-temperature (Tin), ambient-temperature (Tamb), water volumetric flow rate (Fw), direct-solar-radiation (Gb) and wind-speed (Vv) considered as variables in the input layer. Two non-linear transfer functions (TANSIG and LOGSIG) in the hidden layer, a linear function (PURELIN) in the output layer, and the Levenberg-Marquardt training algorithm were applied. The results showed that both ANN models achieved satisfactory results with a coefficient of determination of 0.9511 and a root mean square error of 0.0193. Then, to get the variable's optimal values: rim-angle, inlet-temperature, and water volumetric flow rate, both ANN models inverted to acquire the multivariable objective function that could be resolved with genetic-algorithms (GA) and particle-swarm-optimization (PSO). The TANSIG function demonstrated better adaptation to the ANNim model by finding all the input variables in a random test with an error of 3.96% with a computational time of 14.39 s applying PSO. The results showed that by using the ANNim methodology, it is feasible to improve the performance of the PTC by optimizing from one, two, and three variables at the same time. In optimizing one variable at a time, it was possible to increase a random test's performance up to 54.78%, 27.62%, and 51.92% by finding the rim-angle inlet-temperature and water volumetric flow rate, respectively. In optimizing two variables simultaneously, it was possible to increase a random test's performance up to 36.73% by finding the appropriate inlet-temperature and water volumetric flow rate. In optimizing three variables simultaneously, it was possible to increase a random experimental test of up to 67.12%. Finally, the new ANNim method proposed may increase the thermal efficiency of a PTC in real-time because of the coupling of metaheuristic algorithms that allow obtaining optimal variables in the shortest possible time. Therefore, it can be a promising and widely used method for optimizing and controlling thermal processes.