Computational intelligence and transient thermal analysis methods for exploratory analysis of source air quality

Thumbnail Image
Sun, Gang
Major Professor
Steven J. Hoff
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Organizational Unit
Agricultural and Biosystems Engineering

Since 1905, the Department of Agricultural Engineering, now the Department of Agricultural and Biosystems Engineering (ABE), has been a leader in providing engineering solutions to agricultural problems in the United States and the world. The department’s original mission was to mechanize agriculture. That mission has evolved to encompass a global view of the entire food production system–the wise management of natural resources in the production, processing, storage, handling, and use of food fiber and other biological products.

In 1905 Agricultural Engineering was recognized as a subdivision of the Department of Agronomy, and in 1907 it was recognized as a unique department. It was renamed the Department of Agricultural and Biosystems Engineering in 1990. The department merged with the Department of Industrial Education and Technology in 2004.

Dates of Existence

Historical Names

  • Department of Agricultural Engineering (1907–1990)

Related Units

Journal Issue
Is Version Of


Over the last decade, with the trend toward larger more intensive animal feeding operations (AFOs) in the United States, ammonia (NH3), hydrogen sulfide (H2S), carbon dioxide (CO2) and particulate matter (PM10) generated and emitted from livestock production facilities have become a growing environmental concern for animal producers and nearby residents. Poor air quality inside the buildings can affect the health and productivity of farm workers and animals; while emissions of gas and dust beyond AFOs can influence the wellness of the neighboring residences, thus increasing the number of disputes and lawsuits against livestock operations.

To assess health and ecological environmental impacts caused by livestock pollutants, there exists a rich body of previous work to conduct numerous air pollutants experiments for different livestock facilities (Aarnink et al., 1995; Groot Koerkamp et al., 1998; Zhu et al., 1999; Ni et al., 2002; Gay et al., 2003; Jacobson et al., 2005; Guo et al., 2006; Hoff et al., 2006; Sun et al., 2008a, 2010). However, direct and long-term measurements of gas and PM10 concentrations and emissions (GPCER) at all animal operations are not practical since every gas source is different and animal and weather conditions change constantly. In the absence of effective and efficient means to directly measure GPCER from each livestock production facility, development of source GPCER mathematical prediction models might be a good alternative to provide reasonably accurate estimates. Additionally, due to the absence of a nationwide monitoring network in the United States, state and federal regulatory agencies have identified a need for air quality predictive (AQP) models to quantify long-term air emission inventories of livestock production facilities. State planners, environmental scientists, and livestock producers also need AQP models to determine science-based setback distances between animal feeding operations and neighboring residences, as well as to evaluate relevant emission abatement strategies, e.g., AQP models can be used by helping state planners and environment scientists to site new operations and to help livestock producers to understand the factors influencing air quality and odor and gas transmission so that they might make wise decisions regarding the selection and implementation of air quality mitigation techniques. In brief, air quality models could make an impact by helping to make government and livestock producers to be more profitable, sustainable and economically viable while protecting the environment and quality of life of all citizens. Up to now, three modeling approaches have been proposed for predicting source air quality: the emission factors method, the multiple regression analysis method, and the process-based modeling method.

Emission factors, expressed by the amount of each substance emitted per animal unit, are multiplied by the number of animal units to get average air emissions from animal operations. Arogo et al. (2003) attempted but could not assign empirical ammonia emission factors to estimate the average ammonia emission rates from various barns because of the many variables affecting air emissions. The under- or overestimated predictive results showed that using emission factors for all animals in all regions was not appropriate without direct and long-term measurements from a substantial number of representative animal feeding operations.

The regression analysis method uses standard least-squares multivariate regression equations to predict GPCER. The purpose of multiple regression analysis is to establish a quantitative relationship between various predictor variables (e.g., weather and animal conditions, production systems, etc.) and air emissions. This relationship is used to understand which predictors have the greatest effect and to forecast future values of the equation response when only the predictors and the direction of their effects are known. Sun (2006) developed statistical multiple-linear regression models to predict diurnal and seasonal odor and gas concentrations and emissions from confined swine grower-finisher rooms. However, the main weakness of this method is that the complex and sometimes nonlinear relationships of multiple variables can make statistical models complicated and awkward (Comrie, 1997). Moreover, these models are highly site-specific, making it difficult to apply to the data from other experiments. The only way to establish a robust set of equations is to sample hundreds of animal feeding operations under different meteorological conditions in the U.S. The lack of sufficient data is the main cause of the uncertainty of the statistical regression models.

The process-based models (also called mechanical models) determine the movement of elements (e.g., nitrogen, carbon, and sulfur) into, through, and out of the livestock production system, investigate the underlying chemical and physical phenomenon, and identify the effects of changing one or more variables of the system. In many cases, this modeling method uses mass balance equations to describe the mechanisms of gaseous emissions and estimate their characteristic and amount at each transformation stage. Recently, Zhang et al. (2005) established a comprehensive and predictive ammonia emission model to estimate ammonia emission rates from animal feeding operations using a process-based modeling approach. The main processes treated in the model included nitrogen excretion from the animals, animal housing, manure storage, and land application of manure. The results showed that the sensitivity analysis of various variables (e.g., manure production system, animal housing designs, and environmental conditions) needs to be quantified and that additional model validation is needed to improve model predictive accuracy. Other researchers also studied the process of mass (ammonia) transport and developed mechanical models for swine feeding operations (Aarnink and Elzing, 1998; Ni et al., 2000; Kai et al., 2006). Although there has been considerable value in the development and application of mechanistic modeling of ammonia volatilization from the main individual sources, some circumstances of gaseous emissions are not well understood and several parameters are difficult to determine experimentally. For example, adsorption, absorption, and desorption of ammonia from various materials in animal barns might be another emission source, but this mechanism is not easily acquired. Moreover, the gas release process is very complex due to abundant nonlinear relationships between gaseous emissions and the many variables that cause gas production. Therefore, a major effort would be required in future process-based model studies.

Due to the absence of adequate information available about the process of gas pollutant production, a black-box modeling approach using computational intelligence technology would be a powerful and promising tool for air quality prediction. Wikipedia (2010) defines computational intelligence (CI) as "CI is an offshoot of artificial intelligence. As an alternative to classical artificial intelligent, it rather relies on heuristic algorithms such as fuzzy systems, neural networks and evolutionary computation. Computational intelligence combines elements of learning, adaption, evolution, and fuzzy logic to create programs that are, in some sense, intelligent. Artificial neural network (ANN) is a branch of CI that is closely related to machine learning." It is noted that black-box models using CI technology do not need detailed prior knowledge of the structure and different interactions that exist between important variables. Meanwhile, their learning abilities make the models adaptive to system changes. In recent years, there has been an increasing amount of applications of ANN models in the field of atmospheric pollution forecasting (Hooyberghs et al., 2005; Grivas et al., 2006; Sousa et al., 2007). The results show that ANN black-box models are able to learn nonlinear relationships with limited knowledge about the process structure, and the neural networks generally present better results than traditional statistical methods. Sun et al. (2008b) developed backpropagation and generalized regression neural network models to predict diurnal and seasonal gas and PM10 concentrations and emissions from swine deep-pit finishing buildings. It was found that the obtained forecasting results of the neural network models were in good agreement with actual field measurements, with coefficient of determination values between 81.2% and 99.5% and very low values of systemic performance indices. The promising results from this work indicated that artificial neural network technologies were capable of accurately modeling source air quality within and emissions from these livestock production facilities.

Although AQP models can be used as a useful tool to forecast air quality over a time period that are beyond an actual monitoring period, the main input variables for the model must be known which require field measurements. These variables include indoor environment (indoor, inlet and exhaust temperatures and relative humidity), outdoor climate conditions (outdoor temperature, relative humidity, wind speed, wind direction, solar energy and barometric pressure), pig size and density (animal units), building ventilation rate, animal activity, overall management practices, and properties of the stored manure, to name a few. Sun et al. (2008c) performed a multivariate statistical analysis and identified four significant contributors to the AQP models: outdoor temperature, animal units, total building ventilation rate, and indoor temperature. The purpose of introducing fewer uncorrelated variables to the models is to reduce model structure complexity, eliminate model over-fitting problems, and minimize field monitoring costs without sacrificing model predictive accuracy. Conducting long-term field measurements of the identified four variables using current engineering approaches are still time consuming and expensive. Therefore, making use of simulation programs is a good alternative to obtain the required significant input variables for AQP models.

Basically, there are three steady-state models used to calculate indoor climate of livestock buildings which include those based on heat, moisture or carbon dioxide balances (Albright 1990). Pedersen et al. (1998) compared these three balance methods for estimating the ventilation rate in insulated animal buildings. They reported that the three methods could give good prediction results on a 24-hr basis when the differences between inside and outside temperature, absolute humidity and CO2 concentrations were greater than 2 , water per kg dry air and 200 ppm, respectively for the buildings tested in Northern Europe. A simple steady-state balance model (Schauberger et al., 1999) was developed for the sensible and latent heat fluxes and CO2 mass flows resulting in the prediction of inside temperature and ventilation rate of mechanically ventilated livestock buildings. The obtained variables were further applied for diurnal and annual odor emission estimates. Due to the lack of field measurements, the accuracy of the predicted parameters could not be determined. Morsing et al. (2003) released a computer program entitled StaldVentTM to help design and evaluate heating and ventilation systems in animal houses. They primarily used a steady-state energy balance method to predict the required ventilation rate and heat capacity, room temperature, CO2 concentration, and expected energy consumption throughout the year.

On the other hand, indoor climate can be predicted by studying thermal transients in buildings. Nannei and Schenone (1999) developed a simplified numerical model for building thermal transient simulation. The model can be applied to compute the room air temperature and the temperature of the inner surface of the walls. The good numerical results compared with the experimental data indicated that this model was useful for the study of unsteady thermal performance. Mendes et al. (2001) presented a dynamic multimodal capacitive nonlinear model to analyze transient indoor air temperature using Matlab/SimulinkTM (Matlab 5.0, 1999). This thermal model was improved by introducing internal gains and the inter-surface long-wave radiation. The predicted results were not experimentally validated however. Morini and Piva (2007) investigated the dynamic thermal behavior of residential heating and cooling systems with control systems during a sinusoidal variation of the outside temperature. The core of their program employed mechanical and thermal energy conservation equations implemented in the SimulinkTM environment. It was found that their transient model outperformed the standard steady-state approach.


The over-arching goal of this study is to predict indoor climate and long-term air quality (NH3, H2S and CO2 concentrations and emissions) for swine deep-pit finishing buildings using a transient building thermal analysis and air quality predictive (BTA-AQP) model and a typical meteorological year/specific weather year data base.

The specific objectives of this research were to:

1. Develop an artificial neural network based air quality predictive (AQP) model to forecast source air pollutants from swine deep-pit finishing buildings as affected by time of day, season, ventilation rate, animal growth cycles, in-house manure storage levels, and weather conditions.

2. Build a lumped capacitance model (BTA model) to predict the transient behavior of indoor environment (ventilation rate and indoor air temperature) according to the thermo-physical properties of a typical swine building, set-point temperature scheme, fan staging scheme, transient outside temperature, and the heat fluxes from pigs and supplemental heaters.

3. Evaluate the complete BTA-AQP model to estimate source air quality for a specific year and predict long-term air quality.

4. Apply the proposed BTA-AQP models to different husbandry management practices and geographical area scenarios in order to assess the potential simulated impacts of these scenarios on long-term air quality


This dissertation is organized in paper format and comprises five papers, corresponding to the four research objectives. The first paper entitled "Forecasting daily source air quality using multivariate statistical analysis and radial basis function networks" has been published in the Journal of the Air and Waste Management Association 58(12):1571-1578. The second paper entitled "Development and comparison of backpropagation and generalized regression neural network models to predict diurnal and seasonal gas and PM10 concentrations and emissions from swine buildings" has been published in the Transactions of the ASABE 51(2): 685-694. The third paper entitled "Prediction of indoor climate and long-term air quality using the BTA-AQP model: Part I. BTA model development and evaluation" and the fourth paper entitled "Prediction of indoor climate and long-term air quality using the BTA-AQP model: Part II. Overall model evaluation and application" have been published in the Transactions of the ASABE 53 (3): 863-881. The fifth manuscript entitled "Simulated impacts of different husbandry management practices and geographical area on long-term air quality" will be submitted to the Transactions of the ASABE. The five papers are followed by an overall summary of the major conclusions of this research and recommendations for future research. Three appendixes, which present sensible heat production procedures, APECAB (Aerial Pollutant Emissions from Confined Animal Buildings) daily data, and TMY3 (Typical Meteorological Year) weather data, follow the overall summary chapter. The acknowledgements are included at the end of this dissertation.

Fri Jan 01 00:00:00 UTC 2010