Performance assessment of packed bed systems for humidity control in greenhouse applications: An experimental-based AI modeling approach

Optimal humidity control is essential for enhancing crop yields and ensuring favorable growth conditions in greenhouse agriculture. Packed bed systems have emerged as effective tools for regulating humidity levels, yet accurately assessing their performance remains unexplored, especially for temperate oceanic climates. This paper presents a packed-bed system with water as the working fluid to increase the humidity level during the winter for greenhouse cultivation. Accordingly, an experimental setup is developed, and a detailed parametric study is conducted. Further, an artificial intelligence (AI)-based modeling approach is developed for evaluating the performance of packed bed systems for greenhouse applications under varying environmental conditions with various inlet air flowrates, such as 176 m³/hr, 286 m³/hr, 383 m³/hr, and 428 m³/hr. Operating with water at an average temperature of 15.7 °C and a flow rate of 12.8 kg/min, the system achieves a significant 50% increase in humidity ratio, transitioning from an inlet humidity ratio of 6 g/kg_da to an outlet ratio of 9 g/kg_da, indicating its efficiency in elevating air humidity levels. The multi-layer perceptron neural network, trained with 112 non-repeated datasets and employing a 2-10-10-1 topology, demonstrates high accuracy and resilience in estimating Δω_a values for the packed bed system, with predictions closely aligning with experimental data and exhibiting a maximum discrepancy within ± 2.5%. This research contributes to the advancement of precision agriculture practices by providing a comprehensive framework for assessing and improving humidity control in greenhouse environments.

Introduction

Traditionally, greenhouse operators have relied on various methods to control humidity, including ventilation, misting systems, and evaporative cooling. However, these approaches often lack precision and efficiency, leading to suboptimal growing conditions and resource wastage [1-4]. In recent years, packed bed systems have emerged as promising solutions for humidity management in greenhouse applications. These systems employ porous/packing materials, that allow the working fluid to interact effectively and stabilize humidity levels within the greenhouse environment.

Despite their potential benefits, the performance assessment of packed bed systems in greenhouse applications remains a complex and challenging task [5-10]. The dynamic interplay of environmental factors, such as temperature fluctuations, airflow patterns, and plant transpiration rates, necessitates comprehensive methodologies for evaluating system performance and optimizing operation [11-12]. In this context, during the winter, we need to enhance the humidity level to maintain the required humidity level for optimum growth of the plants in the greenhouse.

Further, AI-based modeling leverages machine learning algorithms to simulate and predict the behavior of complex systems under varying conditions [13-16]. By training models on experimental data and simulation results, AI-based approaches enable accurate predictions of system performance and facilitate the identification of optimal operating strategies. In the context of packed bed systems for humidity control in greenhouses, AI-based modeling holds significant promise for enhancing the system’s applicability, minimizing resource consumption, and maximizing crop yields.

The objectives of this current research work are to study the packed bed system with water as a working fluid to increase the humidity level for greenhouse cultivation and to develop and validate an AI-based modeling approach for assessing the performance of packed bed systems in greenhouse applications. In this regard, an experimental setup is designed and fabricated to provide insights into the performance of packed bed systems for a temperate oceanic climate. The specific objectives of this research are:

• To design and characterize a packed bed system for humidity management in a greenhouse environment.
• To study the packed bed system with different working conditions
• To develop and validate AI-based models for predicting the performance of the packed bed system under various operating conditions.

By addressing these objectives, this research will contribute to the advancement of sustainable and intelligent humidity management solutions for greenhouse applications, with potential benefits for crop production and environmental sustainability.

Methodology

Experimental setup and procedures

An experimental setup is designed and implemented to obtain real-world greenhouse conditions specifically for the UK environmental condition. Figure 1 details the various components and their interconnections, which work together to increase the humidity level of the greenhouse. The experimental procedures involve the installation of packing material with variations in temperature, humidity, and airflow. Ambient air is drawn into the system through the air inlets. The air then passes through the packed bed, where the packing materials are installed. Here, the air interacts with the water, which increases the moisture level of the air. The relatively humid air is then directed to the air outlet for introduction into the greenhouse environment. The water, after releasing water vapor into the air, goes to the tank and is again pumped to the packed bed. Data collection instruments, such as sensors and data loggers, are utilized to monitor environmental parameters and system performance.

Figure 1: Schematic of experimental setup.

The system is compact, with dimensions of 0.44 m x 0.17 m x 0.46 m, and utilizes a counter-flow pattern for efficient air and water interaction. The core of the system is the packed bed, which is constructed using PVC corrugated sheets with dimensions of 400 mm x 150 mm x 400 mm. The packed bed system’s specific surface area is 256 m2/m³ which facilitates effective interaction between working fluids. The system features two solution pumps, specifically Iwaki Magnet Pumps MX-70VM-13 with a capacity range of 90 – 100 l/min, which circulate the water between the packed bed and the water tank. Additionally, two blowers ensure a nominal air volume of 300 m³/h. The dehumidifier and regenerator are connected via 1-inch diameter pipes, and the outlet/inlet ducts have a diameter of 100 mm. The airflow rate is measured using a Testo 405I hot wire thermo-anemometer, which has an accuracy of ± 0.1 m/s and a range of 0 to 20 m/s. Water temperature is monitored with a K-type thermocouple, accurate to ± 0.5 °C, capable of measuring within a broad range of -200 to 1100 °C. Air humidity is assessed with Venatronics LLC THS14-A11-30-N, which offers an accuracy of ± 0.6 °C and ± 2.5% within a humidity range of 0 to 100% RH and an operating temperature range of -30 °C to 75 °C. The water flow rate is measured by an RS 257-133 radial flow turbine flow meter, which has a similar accuracy of ± 2% l/h and a flow range of 1.5 l/min to 30 l/min.

Methodology of intelligent models

The intelligent modeling approach involves machine learning algorithms to predict the behavior of packed bed systems. Training datasets are generated using experimental data collected from the greenhouse setups, including input parameters such as humidity, and airflow rates, as well as corresponding output parameters representing system performance (Δω_a). The artificial neural network is considered in the present study. An Artificial Neural Network (ANN) is a common method of artificial intelligence that is constructed in a manner that is analogous to the information-processing system of the human brain. They serve as computational tools for modeling complex non-linear relationships between independent and dependent variables, with extensive applications in various energy systems as documented in existing literature [17-20]. Among the various types of ANNs, the Multi-Layer Perceptron (MLP) is particularly notable for regression tasks. MLPs are structured with an input layer, one or more hidden layers, and an output layer, each comprising numerous units known as neurons. This architecture enables MLP networks to perform deep learning, allowing them to predict both simple and complex functional behaviors accurately. An MLP network is characterized by having at least three layers: the input layer, which receives the initial data; the hidden layers, which process the data through transfer functions to identify patterns; and the output layer, which delivers the final prediction. The connections between neurons across these layers are established through weights, which adjust over time to strengthen or weaken the influence of each input. The number of neurons in the input and output layers is determined by the number of input and output variables, respectively. In the hidden layers, neurons work to pattern the behavior of the input data, ultimately influencing the predictions made by the output layer neurons [21].

Eq. (1) outlines the process of adjusting the weight value for the connection between the p^th neuron in layer M and the q^th neuron in layer (M+1).

$\Delta w_{pq}=\beta {\delta }_q{o}_p\text{ (1)}$

Here, β signifies the learning rate, δ_q denotes the residual error between the q^th neuron in layer (M+1) and the p^th neuron in layer M, as detailed in Eqs. (2-3), and o_p represents the output value of index p in the M^th layer.

${\delta }_q=o_q\left(d_q-o_q\right)\left(1-o_q\right)$ ; d= target value of output layer for that neuron. (2)

${\delta }_q=T_q\left(1-T_q\right){\displaystyle \sum _k{\delta }_k{w}_{kq}}$ ; T= target value of hidden layer for that neuron. (3)

Eq. (2) serves to compute the residual error of the output layer’s neuron, while Eq. (3) is dedicated to the hidden layer’s neuron, which significantly relies on the alteration in the weight of the k^th neuron in the (M+2) layer adjacent to layer (M+1). The adjustment of β holds paramount importance within the network dynamics: small β values entail a lower convergence rate, whereas excessively large β values tend to induce oscillations. As suggested in the literature [17], the β value can be stabilized by introducing a momentum factor into the weight equation (Eq. 1). Consequently, Eq. (1) transforms into Eq. (4).

$\Delta w_{qp}(n+1)=\beta {\delta }_q{o}_p+\xi \Delta w_{qp}(n);n=iterationnumber$ (4)

Here, ξ represents a constant factor (positive) that governs the impact of previous variations in weight values on the current direction of movement within weight space. With ‘n’ inputs considered in the network, the net input function ‘A’ can be constructed by multiplying the input values by their corresponding weights. Subsequently, the output “Y” is derived by applying a transfer function to the resulting net function, as described in Eq (5).

$Y=F(A)=F\left[{\displaystyle {\sum }_{i=1}^n{X}_i{w}_i+b}\right]\text{ (5)}$

In the aforementioned context, X, b, and w respectively denote the input, bias, and weight. This procedure is elucidated through a schematic representation, as depicted in Figure 2 [permission obtained from Elsevier (License Number 5787270096839) and modified]. Additionally, the output of the perceptron model can be directly influenced by the choice of transfer (activation) function, which may exhibit either linear or non-linear characteristics. MATLAB offers three fundamental activation functions: purelin, logsig, and tansig, each tailored to different requirements. These functions are visually represented in Figure 3, accompanied by their respective function equations.

Figure 2: Schematic representation of an ANN illustrating overall model architecture and information flow.

Figure 3: Comparative analysis of activation functions in ANN (a: Purelin, b: Logsig, c: Tansig).

Performance parameters

Performance parameters are identified to assess the effectiveness of the packed bed systems in controlling humidity levels in the greenhouse. The key parameter is the humidity change during the operation. This parameter is quantified based on experimental measurements and simulation results obtained from the AI-based models. Humidity change (Δω_a) represents the amount of water vapor added downstream of the air and that can be expressed as Eq. (6).

$\Delta {\omega }_a=\left({\omega }_{a,o}-{\omega }_{a,i}\right)\text{ (6)}$

In this context, ω_{a, i} represents the specific humidity of the incoming air, while ω_a,o denotes the specific humidity of the outgoing air.

Results & discussion

Parametric study

Figure 4 is obtained from an experiment or a real-time monitoring system that is tracking the performance of a packed bed system. The constant air flow rate suggests that the system is designed to handle a specific volume of air per hour. Figure 4 expressed a comprehensive parametric analysis to evaluate system performance under various operating conditions. This would involve assessing the average water temperature, as well as the water flow rate. The water with a mean temperature of 15.7 °C, circulates through the system at a flow rate of 12.8 kg/min is considered in the present study. Figure 4 contains four sub-figures that illustrate the relationship between air flow rate and humidity ratio throughout different times of the day. It is beneficial to analyze the performance of a packed bed system designed to increase the humidity level of air using water as the working fluid. Figure 4a corresponds to an airflow rate of 176 m³/hr, we observe that the inlet humidity ratio remains relatively constant over the period shown, while the outlet humidity ratio exhibits more variability. This suggests that the packed bed system is effectively adding moisture to the air, as the outlet humidity is consistently higher than the inlet. Figure 4b, at an airflow rate of 286 m³/hr, shows a similar pattern, with the outlet humidity ratio again being higher than the inlet, indicating successful humidification. Figure 4c, with an airflow rate of 383 m³/hr, and Figure 4d, at 428 m³/hr, both continue to display the trend of a higher outlet humidity ratio compared to the inlet. More specifically, for instance, when the inlet humidity ratio is 6 g/kg_da and the outlet is 9 g/kg_da, this would represent a 50% increase in humidity ratio due to the system’s operation. In summary, Figure 4 demonstrates the effectiveness of the packed bed system in humidifying air, with the outlet humidity ratio consistently higher than the inlet across various air flow rates.

Figure 4a-d: Effect of air flow rate on outlet humidity ratios.

Figure 5 represents quantitative data (Δω_a) collected over a specific time interval. The y-axis represents the change in a parameter denoted as Δω_a (g/kg_da), and the x-axis represents the duration in minutes. Figure 5 includes four different lines, each representing a different flow rate condition, as indicated by the legend: 176 m³/hr (purple line), 286 m³/hr (blue line), 383 m³/hr (red line), and 428 m³/hr (black line). The general trend of Figure 5 indicates that as the duration increases, the parameter Δω_a (g/kg_da) tends to decrease, especially at lower flow rates (176 m³/hr and 286 m³/hr). The higher flow rates (383 m³/hr and 428 m³/hr) show more stability in the parameter value over time.

Figure 5: Effect of air flow rate on humidity change, inlet and outlet humidity ratio basis.

AI- modeling and its results

The network inputs utilize two key variables, namely ṁa and ωa. To establish a model that is both accurate and resilient, Δω_a is precisely estimated through a multi-layer perceptron neural network. The structure of the suggested ANN model is depicted in Figure 6. This study employed 112 non-repeated datasets (comprising 56 input and 56 target data points) to map the input-output correlation. A feed-forward neural network, enhanced with a backpropagation learning algorithm, was employed for model training. From the total datasets, 70% were randomly selected for training purposes, while the remaining 30% were allocated for testing and validation. Various activation functions were utilized to ensure the ANN algorithm achieves commendable predictive accuracy. Details on the optimal hyper-parameters for the ANN model can be found in Appendix A.

Figure 6: Structure of proposed ANN model, featuring two hidden layers, each containing ten neurons (2-10-10-1 configuration).

The TrainLM training function was identified as the most appropriate for this present research study. The architecture of the neural networks utilized a tangent sigmoid activation function for both the output and hidden layers. To minimize the risk of overfitting during the training phase, a portion of the data samples, amounting to 15%, was allocated for validation purposes, while another 15% was set aside for testing the efficacy of the trained networks. The initial trials, which employed networks with a single hidden layer, did not yield satisfactory results due to a pronounced discrepancy between the experimental data and the predictions, attributed to the complexity of the problem at hand and regardless of other parameter settings. This observation led to the exploration of networks with multiple hidden layers for further testing and evaluation, aiming to enhance the accuracy of the predictions. Adjustments were made to the number of neurons within these layers, ranging from 8 to 12. The findings revealed that the optimal network configurations for predicting Δω_a were those with two hidden layers, specifically the 2-10-10-1 topology, which comprises two input neurons in the first layer, ten hidden neurons in each of the two hidden layers, and one output neuron, as depicted in Figure 6.

Figure 7 illustrates the predicted Δω_a values as determined by the ANN model, alongside a comparison with the corresponding experimental data. The packed bed system’s output parameters and predictive patterns generated by the AI model closely agreed on the experimental outcomes across all tested conditions. The greatest observed discrepancy between the ANN predictions and the experimental Δω_a values was within a margin of ± 2.5%.

Figure 7: Comparison between experimental and AI models outcomes for Δω_a.

Conclusion

The study investigates a packed-bed system employing water as a working fluid to boost winter humidity levels in greenhouse cultivation. Through an experimental setup and extensive parametric analysis, the system’s performance is evaluated. Key findings reveal the system’s ability to maintain consistent airflow rates (e.g., 176 m³/hr to 428 m³/hr), leading to higher outlet humidity ratios than inlet, indicating effective humidification. Operating with water at 15.7 °C and a flow rate of 12.8 kg/min, the system achieves a noteworthy 50% increase in humidity ratio. A multi-layer perceptron neural network, configured with a 2-10-10-1 topology, TrainLM training function, and tangent sigmoid activation functions, accurately predicts Δω_a values. This ANN model closely aligns with experimental data, demonstrating a maximum discrepancy of ± 2.5%, highlighting its precision in forecasting system performance.

Appendix A: Details of the ANN model

For the Artificial Neural Network (ANN) in this study, the Levenberg-Marquardt algorithm, also known as TrainLM, was selected due to its superior convergence properties, particularly in regression tasks. Before the commencement of training, bias and weight values are assigned randomly [20-24]. These values are then iteratively updated by the TrainLM function, which employs the gradient descent method to optimize the network. The training of the multilayer perceptron (MLP) model adheres to specific stopping criteria, namely a minimum gradient of 10-7 and a maximum of 10,000 epochs [25-30]. The ideal hyper-parameters for the ANN model are identified in Table A1. The tangent sigmoid function was the chosen transfer function for the output layer. Analysis of Table A1 indicates that the configuration utilizing a hyperbolic tangent sigmoid transfer function for both hidden layers, with each layer containing ten neurons, results in the lowest mean squared error (MSE) when compared to other tested algorithms.

1. Vadiee A, Martin V. Energy management in horticultural applications through the closed greenhouse concept, state of the art. Renew Sustain Energy Rev. 2012;16:5087–100. https://doi.org/10.1016/j.rser.2012.04.022.
2. Singh RD, Tiwari GN. Energy conservation in the greenhouse system: A steady state analysis. Energy. 2010; 35:2367–73. https://doi.org/10.1016/j.energy.2010.02.003.
3. Pasqualin P, Lefers R, Mahmoud S, Davies PA. Comparative review of membrane-based desalination technologies for energy-efficient regeneration in liquid desiccant air conditioning of greenhouses. Renew Sustain Energy Rev. 2022;154. https://doi.org/10.1016/j.rser.2021.111815.
4. Muniandy JM, Yusop Z, Askari M. Evaluation of reference evapotranspiration models and determination of crop coefficient for Momordica charantia and Capsicum annuum. Agric Water Manag. 2016; 169:77–89. https://doi.org/10.1016/j.agwat.2016.02.019.
5. Öztürk HH, Başçetinçelik A. Energy and exergy efficiency of a packed-bed heat storage unit for greenhouse heating. Biosystems engineering. 2003;86(2):231-245.
6. Jain D. Modeling the performance of greenhouse with packed bed thermal storage on crop drying application. J Food Eng. 2005; 71(2):170-178.
7. Bouadila S. Assessment of the greenhouse climate with a new packed-bed solar air heater at night, in Tunisia. Renew Sustain Energy Rev. 2014; 35:31-41.
8. Kumari N, Tiwari GN, Sodha MS. Thermal modelling for greenhouse heating by using packed bed. Int J Agric Res. 2010; 5(9):788-798.
9. Gautam A, Saini RP. A review on technical, applications and economic aspect of packed bed solar thermal energy storage system. J Energy Storage. 2020; 27:101046.
10. Nahhas T, Py X, Olives R. Life cycle assessment of air-rock packed bed storage system and its comparison with other available storage technologies for concentrating solar power plants. Waste Biomass Valorization. 2020; 11(5):2357-2365.
11. Ghoulem M. Greenhouse design and cooling technologies for sustainable food cultivation in hot climates: Review of current practice and future status. Biosyst Eng. 2019; 183:121-150.
12. Soussi M. Comprehensive review on climate control and cooling systems in greenhouses under hot and arid conditions. Agronomy. 2022;12(3):626.
13. Cigizoglu HK. Estimation and forecasting of daily suspended sediment data by multi-layer perceptrons. Adv Water Resour. 2004; 27:185-195. https://doi.org/10.1016/j.advwatres.2003.10.003.
14. Eberhart RC. Neural Network PC Tools a Practical Guide. Academic Press, Elsevier; 2014.
15. Ghobadian B, Rahimi H, Nikbakht AM, Najafi G, Yusaf TF. Diesel engine performance and exhaust emission analysis using waste cooking biodiesel fuel with an artificial neural network. Renew Energy. 2009; 34:976-982. https://doi.org/10.1016/j.renene.2008.08.008.
16. Caner M, Gedik E, Keçebaş A. Investigation on thermal performance calculation of two type solar air collectors using artificial neural network. Expert Syst Appl. 2011; 38:1668-1674. https://doi.org/10.1016/j.eswa.2010.07.090.
17. Güler NF, Übeyli ED, Güler I. Recurrent neural networks employing Lyapunov exponents for EEG signals classification. Expert Syst Appl. 2005; 29:506-514. https://doi.org/10.1016/j.eswa.2005.04.011.
18. Sheela KG, Deepa SN. Review on methods to fix number of hidden neurons in neural networks. J Sci Ind Res. 2014; 73:686-688. https://doi.org/10.1155/2013/425740.
19. Yassin MA, Alazba AA, Mattar MA. Artificial neural networks versus gene expression programming for estimating reference evapotranspiration in arid climate. Agric Water Manag. 2016; 163:110-124. https://doi.org/10.1016/j.agwat.2015.09.009.
20. Ranka S, Mohan CK, Mehrotra K, Menon A. Characterization of a Class of Sigmoid Functions with Applications to Neural Networks. Neural Netw. 1996 Jul;9(5):819-835. doi: 10.1016/0893-6080(95)00107-7. PMID: 12662565.
21. Suranjan S, Reddy SVK, Kumar S. Experimental Investigation and Neural network-based parametric prediction in a multistage reciprocating humidifier. Appl Energy. 2021; 293:116958. https://doi.org/10.1016/j.apenergy.2021.116958.
22. Bhowmik M, Muthukumar P, Anandalakshmi R. Experimental-based multilayer perceptron approach for prediction of evacuated solar collector performance in humid subtropical regions. Renew Energy. 2019; 143:1566-1580. https://doi.org/10.1016/j.renene.2019.05.093.
23. Negnevitsky M. Artificial Intelligence: a Guide to Intelligent Systems. Germany, Addison-Wesley Pearson Education. 2005.
24. Mohandes MA, Rehman S, Halawani TO. A neural networks approach for wind speed prediction. Renew Energy. 1998;145-147. https://doi.org/10.1016/S0960-1481(98)00001-9.
25. Zandie M. Multi-input multi-output machine learning predictive model for engine performance and stability, emissions, combustion and ignition characteristics of diesel-biodiesel-gasoline blends. Energy. 2023; 262:125425. https://doi.org/10.1016/j.energy.2022.125425
26. Rahmawati A, et al. Analysis of a Backpropagation Neural Network Training Algorithm for EEG-Based Drowsiness Identification. In: 2023 International Conference on Radar, Antenna, Microwave, Electronics, and Telecommunications (ICRAMET). IEEE. 2023.
27. Ibrahim S, Abdul Wahab N. Improved Artificial Neural Network Training Based on Response Surface Methodology for Membrane Flux Prediction. Membranes (Basel). 2022 Jul 23;12(8):726. doi: 10.3390/membranes12080726. PMID: 35893444; PMCID: PMC9394359.
28. Anta DK, et al. Identification of Nitrogen Content of Vernonia amygdalina Leave Based on Artificial Neural Network Modeling. Proceedings of the 7th International Conference on Food, Agriculture, and Natural Resources (IC-FANRES 2022). Springer Nature. 2023.
29. Bhatt AN, Shrivastava N. Application of Artificial Neural Network for Internal Combustion Engines: A State of the Art Review. Arch Comput Methods Eng. 2022;29(2):897-919. doi: 10.1007/s11831-021-09596-5. Epub 2021 May 3. PMID: 33967576; PMCID: PMC8090920.
30. Stanley UC, et al. Prediction of the Phytochemical Properties of Luffa Cylindrica Seed Oil Using Artificial Neural Network. Path of Science. 2023; 9(1):1001-1010.