Modelling of Energy Consumption in Wheat Production Using Neural Networks “Case Study in Canterbury Province, New Zealand“

An artificial neural network (ANN) approach was used to model the energy consumption of wheat production. This study was conducted over 35,300 hectares of irrigated and dry land wheat fields in Canterbury in the 2007-2008 harvest year. In this study several direct and indirect factors have been used to create an artificial neural networks model to predict energy use in wheat production. The final model can predict energy consumption by using farm condition (size of wheat area and number paddocks), farmers’ social properties (education), and energy inputs (N and P use, fungicide consumption, seed consumption, and irrigation frequency), it can also predict energy use in Canterbury wheat farms with error margin of ±7% (± 1600 MJ/ha). Keywords—Artificial Neural Network, Canterbury, Energy consumption, Modelling, New Zealand, Wheat I.INTRODUCTION EW ZEALAND is one of the countries with the highest energy input per unit (in agriculture) in the world. Furthermore, in terms of shipping, the influence of increasing fuel costs in the world is greater on New Zealand farming than in other countries. This study was conducted over 35,300 hectares of irrigated and dry land wheat fields in Canterbury, New Zealand, in the 2007-2008 harvest year, which reported 87% of the wheat area and 66% of arable area harvested in New Zealand. Using energy in developed and developing countries have created several environmental, commercial, technical, and even social sciences, which need to study. Analysing numerous amount of different sorts of information is necessary to reduce the energy consumption and its environmental impacts. Some of the energy sources in agriculture sector are classified in other sectors. For example, fuel consumption in farm operations may be classified in the transport sector or indirect energy sources (fertilizers, seeds, and agrichemicals) may be estimated in the industrial sector. Consequently, official national statistics usually do not show accurate energy use in agriculture and they pay very little attention to the energy consumption of agriculture sector [1]. Energy modelling is an interesting subject among engineers and scientists who are concerned about energy production and Majeed Safa, Department of Environmental Management, PO Box 84, Lincoln University 7647 Christchurch, New Zealand Phone: 64 3 325 3838, Fax: 64 3 325 3845, Email: [email protected] Sandhya Samarasinghe, Associate Professor, Centre for Advanced Computational Solutions (C-fACS) and Department of Environmental Management P O Box 84, Lincoln University 7647 Christchurch, New Zealand Phone: 64 3 325 3838, Fax64 3 325 3845 Email: [email protected] consumption and environmental impacts [2, 3]. In energy area, a wide range of models have been used, from geological models in research on natural resources, to modelling future energy demand [3]. In the past, regression analysis was the common modelling technique on energy studies. However, recently, neural networks (NN) have been increasingly used in energy studies [4]. Due to ability of neural networks to model complex nonlinear systems in a flexible and adaptive manner, NN have been used more and more in the recent years [5]. Several studies have used NN for classification, prediction, and problem solving in energy field. NNs have been applied by researchers in a wide range of application areas, such as mathematics, engineering, medicine, economics, environment, and agriculture [4]. Numerous number of researchers have applied neural networks in the modelling of various scenarios to solve different problems, in which no explicit formulations were available [6]. The main advantage of neural networks is that they are able to use prior information (historical underlying process data). In most studies, a feed-forward Multi-Layered Perception (MLP) paradigm trained by back propagation (BP) is used. Due to its documented ability to model any function, MLP trained with BP is selected to develop apparatus, processes, and product prediction models [5, 7, 8]. In the last twenty years, use of neural networks in energy studies has increased and a wide range of studies using neural networks (NNs) in energy systems has been done [9]. Javeed Nizami and Ahmed G. Al-Garni [10] applied seven years of data to develop a two layered artificial neural network forecasting model to relate the electric energy consumption in the Saudi Arabia to the weather data, global radiation, and population. A neural network was developed by Mohandes et al. [11] to predict the wind speed prediction. Kalogirou and Bojic [12] developed and applied a multilayer back propagation learning algorithm to predict energy consumption of a passive solar building. Kalogirou [13] has reviewed various applications of NNs in energy studies. Fang [6] developed a neural networks model to estimate energy requirements for the reduction of cultivated wheat area. Aydinalp [14] used a simple neural network based energy consumption model for the Canadian residential sector. An artificial neural network model to predict the regional peak load of electricity in Taiwan has been used by Hsu [15]. The benefits of using NN models are the simplicity of application and robustness in results. The application of ANNs has developed into a powerful tool that can approximate any nonlinear input-output mapping faction to any degree of accuracy in an iterative manner. ANNs have many attractive properties for the modelling of complex production systems, universal function approximation capability, resistance to noisy or missing data, accommodation of multiple nonlinear variables with unknown interactions, and good generalization ability [16]. In the processing of inputs by network, each neuron processes the weighted inputs through a transfer function to produce its output. The function may be a linear or a nonlinear function. There are several transfer function, such as Logistic, Hyperbolic-tangent, Gussian, and Sin. The output depends on the particular function. This output is then sent to the neuron in the next layer through weighted corrections M. Safa, S. Samarasinghe Modelling of Energy Consumption in Wheat Production Using Neural Networks “Case Study in Canterbury Province, New Zealand” N World Academy of Science, Engineering and Technology International Journal of Biological, Biomolecular, Agricultural, Food and Biotechnological Engineering Vol:4, No:12, 2010 915 International Scholarly and Scientific Research & Innovation 4(12) 2010 scholar.waset.org/1999.1/15114 In te rn at io na l S ci en ce I nd ex , A gr ic ul tu ra l a nd B io sy st em s E ng in ee ri ng V ol :4 , N o: 12 , 2 01 0 w as et .o rg /P ub lic at io n/ 15 11 4 and these neurons complete their outputs by processing the sum of weighted inputs through their transfer functions. When this layer is the output layer neuron output is the predicted output. Several methods of error estimation have been proposed. The Mean square error (MSE) is the most commonly used error indicator of the prediction over all the training patterns. MSE is a very useful to compare different models; it shows the networks ability to predict the correct output. The MSE can be written as: