Carbon Monoxide Prediction Using Artificial Neural Network And Imperialist Competitive Algorithm

Carbon monoxide (CO) is one of the main air pollutants produced by incomplete combustion process particularly in the urban areas and exposing to the CO polluted environments will definitely affect human health. Therefore, providing a comprehensive computer modeling based on the current and previous related information for further study, analyses and decision making is of paramount importance. There are number of approaches in air pollution modeling and prediction such as traditional statistical methods and more recent ones based on the artificial intelligence solutions. Successes of artificial neural networks (ANN) and evolutionary algorithms, like genetic algorithm, in simulating the nonlinear dynamic of environmental phenomena, have introduced them as powerful alternatives amongst researchers. In this paper, the CO concentration prediction problem is solved by a combination of ANN and a socio-political heuristic search method named imperialistic competitive algorithm (ICA). ICA is a global heuristic search method that uses imperialism and imperialistic competition process as a source of inspiration. The multi layer perceptron (MLP) network topology with Levenberg-Marquardt training technique has been applied in the prediction. In order to enhance the efficiency of the corresponding network, ICA has been hybridized in MLP network (ICA-NN) to optimize the network weights. This research utilized a set of that comprise toxic air samples gathered in monitoring network sites in Dallas-Fort Worth during 2001 to 2006. The performance of the proposed approach is measured by the mean squared error, for both training and testing phases. Comparison between the functionality of hybrid ICA-NN and the mentioned MLP network provides the fact that the ICA-NN is superior in terms of reliable performance with acceptable accuracy for CO pollutant prediction.