Predicting freshwater production and energy consumption in a seawater greenhouse based on ensemble frameworks using optimized multi-layer perceptron

Abstract

Water shortage in arid and semi-arid land is one of the most important challenges of decision-makers. The seawater greenhouse (SWG) is a useful solution for water supply in the agriculture sector. The optimal design of a SWG with lower consumption of energy and higher freshwater production is a real challenge for the decision-makers. This study used two ensemble models and multiple multi-layer perceptron (MLP) models based on non-climate data to predict freshwater production energy consumption in the SWG. The Copula Bayesian average model (CBMA) was used to develop the BMA model using different copula functions and distributions. In the first level, multiple MLP models using the dimension of SWG as inputs predicted freshwater and energy consumption in a SWG. In the next level, The CBMA and BMA were used to predict freshwater production and energy consumption. The uncertainty analysis of outputs, use of new models and non-climate data are the novelties of the current study. The results indicated that the CBMA decreased the mean absolute error (MAE) value of the BMA, MLP-SEOA, MLP-SCA, MLP-BA, MLP-PSO, and MLP models by 2.7%, 19%, 31%, 40%, 41%, and 42%, respectively for predicting freshwater production. The root mean square error (RMSE) of the CBMA was 40%, 49%, 56%, 57%, 62%, and 64% lower than those of the BMA, MLP-SEOA, MLP-SCA, MLP-BA, MLP-PSO, and MLP models, respectively for predicting energy consumption. The uncertainty analysis indicated that the CBMA and BMA provided the lowest uncertainty among other models. The current study results indicated that the use of ensemble models improved the accuracy of individual models for predicting energy consumption and freshwater production. The findings of the study indicated that the ensemble models using the dimension of SWGs as inputs successfully predicted energy consumption and freshwater production in a SWG.