The forecast of power consumption and freshwater generation in a solar-assisted seawater greenhouse system using a multi-layer perceptron neural network

This research is a simulation of the solar seawater greenhouse framework by using a hybrid model of multilayer perceptron neural network and different optimization algorithms. Four key factors, roof transparency, the front evaporator height, greenhouse length, and width, are considered decision-making variables to investigate power consumption and water production. In this regard, six metaheuristic approaches are employed to hyper-tune the learning process of the neural network. The accuracy of the proposed models is estimated via statistical in-dicators. The obtained results show that biogeography-based optimization and genetic algorithm had the min-imum RMSE value for both power consumption and water generation. Based on these criteria, the ACO method has the worst rank. It should be mentioned that some data are training set and some others are employed for the testing procedure. According to the results, less testing error is related to BBO and GA methods. Water con-sumption assessment represents that the least error is related to ACO strategy. It is found that by change of width and transparency, more water production is related to ES and GA methods with almost 120 m3/day and 105 m3/ day, respectively. Moreover, ES, ACO, and GA have less power consumption. Finally, it is found that when transparency was 0.4, the water production value could be 115 m3/day while for 0.6 of transparency this amount was 95.

Automated summary

This research simulates the solar seawater greenhouse framework using a hybrid model of multilayer perceptron neural network and different optimization algorithms. Key factors such as roof transparency, front evaporator height, greenhouse length, and width are considered to investigate power consumption and water production. Six metaheuristic approaches are employed to fine-tune the learning process, and the results show that biogeography-based optimization and genetic algorithm perform the best. The ACO method ranks the worst. Different methods also show variations in water consumption and power consumption.