MLP and MLR models for instantaneous thermal efficiency prediction of solar still under hyper-arid environment

Highlights

• Solar still was used to produce water.

• The instantaneous thermal efficiency (η_ith) of solar still was modeled.

• Multilayer perceptron (MLP) neural network and multiple linear regression (MLR) were used in the modeling process.

• The MLP model was better than MLR model.

• Using the MLP model provides η_ith values with high accuracy.

Abstract

The purpose of this study was to determine the viability of modeling the instantaneous thermal efficiency (η_ith) of a solar still, using weather and operational data with Multi-Layer Perceptron (MLP) neural network and multiple linear regressions (MLR). This study used weather and operational variables that were hypothesized to affect solar still performance. In the MLP model, nine variables were used as input parameters: Julian day, ambient temperature, relative humidity, wind speed, solar radiation, temperature of feed water, temperature of brine water, total dissolved solids of feed water, and total dissolved solids of brine water. The η_ith was the one node present in the output layer. The same parameters were used in the MLR model. Discussions of advantages and disadvantages are given from different points of view for both models. Performance evaluation criteria indicated that the MLP model was better than the MLR model. The average value of the coefficient of determination for the MLP model was higher by 11.23% than for the MLR model. The average value of the root mean square error for the MLP model (2.74%) was lower compared to the MLR model. The relative errors of predicted η_ith values for the MLP model were mostly in the vicinity of ±10%. Therefore, the MLP model is preferred as a highly precise model in predicting η_ith compared to the MLR model. It is expected that this study could be highly beneficial to those dealing with the design of solar desalination systems.

Introduction

Desalination is an appropriate means for providing pure water. Moreover, desalination is one of the ways to deal with the water crisis, because it is considered a stable source of water production (Karimi Estahbanati et al., 2015). Many sources of energy can be utilized for water desalination such as fossil fuel and electricity (Feilizadeh et al., 2010). The desalination process consumes a significant amount of energy (Dev et al., 2011), which is often provided by fossil fuels, and this causes many adverse effects on the environment owing to the emission of significant amounts of greenhouse gasses: around 1-kg CO₂-equivalents (CO₂-e) per kW-h (or 3.8 kg CO₂-e per) m³ of desalinated water (Ghaffour et al., 2014). However, renewable energy can be utilized for the desalination process to avoid these problems (Karimi Estahbanati et al., 2015, Rajaseenivasan and Murugavel, 2013). One of the available renewable energies for the desalination of low quality water is solar energy. A solar still is a simple way to desalinate water using solar energy and is one of the best solutions to confront the water crises in remote arid areas. With locally available materials, a solar still can be manufactured easily and the operation is simple, plus the maintenance is inexpensive and no skilled persons are required (Omara and Kabeel, 2014).

On the other hand, a solar still is not commonly used because of its low thermal efficiency and its low productivity (Kabeel et al., 2012, Buros, 2000, Tiwari, 1992), which depend on meteorological, design and operational parameters (Sivakumar and Ganapathy Sundaram, 2013). Accordingly, to optimize and enhance solar energy to be widely utilized, further modeling processes to forecast optimal performance and to identify vital parameters related to performance are required.

Solar stills should be designed and operated for optimum efficiency through modeling and prediction. Instantaneous thermal efficiency is one of the important parameters to be accurately determined and is therefore one of the most important parameters for judging the performance of solar stills since it details how effective the still is at absorbing solar energy and evaporating and collecting water. Classical modeling methods are too complex and the solutions require long calculations, and are occasionally completely unrealizable (Tripathy and Kumar, 2008, Khadir, 2005, West et al., 1997). In addition, the thermal performance analyses of solar systems are too complex, and analysis generally needs a large amount of computer power and requires a considerable amount of time for accurate forecasts. Moreover, the experiments and thermal analyses of solar thermal systems such as solar collectors and solar stills are complicated owing to numerous measurements and the heat transfer processes. On the other hand, it is very important for designers and engineers to be able to choose the best system speedily and accurately (Kalogirou, 2000, Kalogirou, 2001, Kalogirou, 2004). As an alternative to the classical methods of modeling, instantaneously thermal efficiency can be accurately modeled and predicted using artificial neural networks (ANNs). ANNs have been utilized in several engineering applications as they can model complex physical phenomena such as those in thermal engineering (Islamgolu, 2003). The ANN method, apart from decreasing the time required, is able to find solutions that make solar-energy applications more applicable and viable (Kaukal et al., 2011).

Several studies have demonstrated the application of ANNs in the modeling of many processes in solar thermal systems. For example, Lecoeuche and Lalot (2005) used an ANN to forecast the in-situ daily performance of solar air collectors. Farkas and Géczy-Víg (2003) developed ANN models for different types of solar thermal collectors to forecast their outlet temperature. An ANN was employed by Kalogirou (2006) for the calculation of the performance parameters of flat-plate solar collectors. The ANN method was also utilized to calculate the efficiency of flat plate solar thermal collectors by Sözen et al. (2008). Santos et al. (2012) used ANNs to determine the effectiveness of modeling solar still distillate production. Hamdan et al. (2013) developed ANN models to find the performance of a triple solar still operating under local atmospheric conditions in Jordan. Kalogirou et al. (2014) used ANNs for the performance forecasting of large solar systems. An ANN model was developed and used by Yaïci and Entchev (2014) to forecast the performance of a solar thermal energy system utilized for domestic hot water and space heating applications. Mashaly and Alazba (2015) conducted a comparative investigation of ANN learning algorithms for modeling solar still production. Moreover, the performance of an ANN and multiple linear regression (MLR) methods were evaluated comparatively for modeling processes in the field of solar energy in recent years. Citakoglu (2015) modeled monthly solar radiation values by ANN and MLR techniques. ANN and MLR were used for modeling the temperature of water in a solar cooker (Çakmak, 2014). Kassem et al. (2011) used MLR analysis and ANN for predicting drying efficiency during the solar drying process. The estimation capacities of MLR and ANN were investigated to estimate monthly-average daily solar radiation over Turkey (Şahin et al., 2013). Caner et al. (2011) used the ANN technique to estimate thermal performances of solar collectors. In order to analyze the ANN performance, they also developed a MLR model. The multi-Layer Perceptron network (MLP) is one of the ANN models that have attracted great efforts in solving a variety of problems (Piekniewski and Tybicki, 2004). In this study, the MLP neural network was used. The objectives are to develop an instantaneously thermal efficiency (η_ith) prediction model for a solar still using the MLP model, to compare the MLP model to the MLR model, and to assess the potential of MLP for predicting η_ith.