A new scheme to predict chaotic time series of air pollutant concentrations using artificial neural network and nearest neighbor searching

Abstract

Air pollution time series is often characterized as chaotic in nature. The prediction using conventional statistical techniques and neural network with backpropagation algorithm, which is most widely applied, do not give reliable prediction results. The new algorithm is therefore proposed to predict the chaotic time series based on the artificial neural network technique. The training to the network is similar to the conventional technique of learning, whereas for testing and prediction purpose, the algorithm searches for the nearest neighbor of the presented test and/or prediction patterns in the training set. The case study of well-known chaotic time series, namely, Lorenz map and real data of ozone concentration are provided to examine the performance of proposed scheme. The most widely used backpropagation algorithm is also used for comparison purpose. The out performance of proposed scheme over backpropagation algorithm is observed. The proposed scheme provides the network, the ability to capture the underlying dynamics of the chaotic time series, as the input patterns are presented one by one to the network.