Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.729200
Title: Integration of principal component analysis, fuzzy C-means and artificial neural networks for localised environmental modelling of tropical climate
Author: Mohd-Safar, Noor Zuraidin
ISNI:       0000 0004 6499 4892
Awarding Body: University of Portsmouth
Current Institution: University of Portsmouth
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Meteorological processes are highly non-linear and complicated to predict at high spatial resolutions. Weather forecasting provides critical information about future weather that is important for flooding disaster prediction system and disaster management. This information is also important to businesses, industry, agricultural sector, government and local authorities for a wide range of reasons. Processes leading to rainfall are non-linear with the relationships between meteorological parameters are dynamic and disproportionate. The uncertainty of future occurrence and rain intensity can have a negative impact on many sectors which depend on the weather condition. Therefore, having an accurate rainfall prediction is important in human decisions. Innovative computer technologies such as soft computing can be used to improve the accuracy of rainfall prediction. Soft computing approaches, such as neural network and fuzzy soft clustering are computational intelligent systems that are capable of integrating humanlike knowledge within a specific domain, adapt themselves and learn in changing environments. This study evaluates the performance of a rainfall forecasting model. The data pre-processing method of Principal Component Analysis (PCA) is combined with an Artificial Neural Network (ANN) and Fuzzy C-Means (FCM) clustering algorithm and used to forecast short-term localized rainfall in tropical climate. State forecast (raining or not raining) and value forecast (rain intensity) are tested using a number of trained networks. Different types of ANN structures were trained with a combination of multilayer perceptron with a back propagation network. Levenberg-Marquardt, Bayesian Regularization and a Scaled Conjugate Gradient training algorithm are used in the network training. Each neuron uses linear, logistic sigmoid and hyperbolic tangent sigmoid as a transfer function. Preliminary analysis of input parameter data pre-processing and FCM clustering were used to prepare input data for the ANN forecast model. Meteorological data such as atmospheric pressure, temperature, dew point, humidity and wind speedhave been used as input parameters. The magnitude of errors and correlation coefficient were used to evaluate the performance of trained neural networks. The predicted rainfall forecast for one to six hour ahead are compared and analysed. One hour ahead for state and value forecast yield more than 80% accuracy. The increasing hours of rain prediction will reduce the forecast accuracy because input-output mapping of the forecast model reached termination criterion early during validation test and no improvement of convergence in the consecutive number of epochs. Result shows that, the combination of PCA-FCM-ANN forecast model produces better accuracy compared to a basic ANN forecast model.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.729200  DOI: Not available
Share: