Pork farm odour modelling using multiple-component multiple-factor analysis and neural networks

• Authors: Kevin R. Janes;Simon X. Yang;Roger R. Hacker
• Venue: Applied Soft Computing
• Year: 2005

The number of non-farming rural residents surrounding pork farm operations has increased greatly in many pork-producing nations. These non-farming rural residents consider pork farm odour emissions to be a serious health threat. This has led to significant lobbying to limit the expansion and development of pork operations, and consequently reduced economic growth in the pork industry. This has led pork producers to search for effective methods to mitigate the odourous emissions, and researchers to seek methods to adequately model the odour. Extensive research has been performed on the use of single-component analysis to model the odour. Several researchers have used multiple-component analysis to extend the effectiveness of previous models. Since odour generation factors, like temperature, also contribute to the odour, the next logical approach to modelling pork farm odour is multiple-component multiple-factor analysis. It is proposed that the multiple-component neural network model be extended to make use of multiple-component multiple-factor analysis. First, a neural network model and a linear multiple regression model are developed and compared using multiple-component analysis to demonstrate the better modelling technique for pork farm odour. The neural network model of the pork farm odour yielded more accurate and precise odour intensity predictions than the linear multiple regression models, indicating that neural networks are the better modelling technique for this application. Subsequently, a multiple-component multiple-factor neural network model was developed and compared with the multiple-component neural network. The multiple-component multiple-factor neural network model generated performance gains, indicating that this approach is relevant to modelling pork farm odour. It is hypothesized that the extension of the multiple-component multiple-factor analysis to include additional significant odour components and odour generation factors in the neural network model will further improve model performance.