No abstract

No abstract

This paper presents a novel algorithm for training radial basis function (RBF) networks, in order to produce models with increased accuracy and parsimony. The proposed methodology is based on a nonsymmetric variant of the fuzzy means (FM) algorithm, which has the ability to determine the number and locations of the hidden-node RBF centers, whereas the synaptic weights are calculated using linear regression. Taking advantage of the short computational times required by the FM algorithm, we wrap a particle swarm optimization (PSO) based engine around it, designed to optimize the fuzzy partition. The result is an integrated framework for fully determining all the parameters of an RBF network. The proposed approach is evaluated through its application on 12 real-world and synthetic benchmark datasets and is also compared with other neural network training techniques. The results show that the RBF network models produced by the PSO-based nonsymmetric FM algorithm outperform the models produced by the other techniques, exhibiting higher prediction accuracies in shorter computational times, accompanied by simpler network structures.

This paper presents expressions correlating the exhaust emissions from a single-cylinder diesel engine with some of the most important properties of the fuels used, using a neural network approach. The exhaust emissions measured were carbon monoxide, hydrocarbons, nitrogen oxides, and particulate matter. The experiments were performed using a matrix of 59 fuels. The cetane number of the fuels covered the range 42-58, the density varied between 0.840 and 0.860 g/mL, and the sulfur content from 0.05 to 0.20 wt %. The predictions were based on specific points of the distillation curve, the cetane number, density, and kinematic viscosity of the fuels. In the case of particulate matter emissions, sulfur content was also employed. The predictions obtained were very good for all the emissions considered. The aromatic content was not used as a predictor variable, because it was found to have a strong inter-correlation with the cetane number, density, and two specific points of the distillation curve, the 50% and the 90% recovery point.

We show how microscopic modelling techniques such as Cellular Automata linked with detailed geographical information systems (GIS) and meteorological data can be used to efficiently predict the evolution of fire fronts on mountainous and heterogeneous wild forest landscapes. In particular, we present a lattice-based dynamic model that includes various factors, ranging from landscape and earth statistics, attributes of vegetation and wind field data to the humidity of the fuel and the spotting transfer mechanism. We also attempt to model specific fire suppression tactics based on air tanker attacks utilising technical specifications as well as operational capabilities of the aircrafts. We use the detailed model to approximate the dynamics of a large-scale fire that broke out in a region on the west flank of the Greek National Park of Parnitha Mountain in June of 2007. The comparison between the simulation and the actual results showed that the proposed model predicts the fire-spread characteristics in an adequate manner. Finally, we discuss how such a detailed model can be exploited in order to design and develop, in a systematic way, fire risk management policies.

No abstract

The popular radial basis function (RBF) neural network architecture and a new fast and efficient method for training such a network are used to model nonlinear dynamical multi-input multioutput (MIMO) discrete-time systems. The proposed training methodology is based on a fuzzy partition of the input space and combines self-organized and supervised learning. The algorithm is illustrated through the development of neural network models using simulated and experimental data. Results show that the methodology is much faster and produces more accurate models compared to the standard techniques used to train RBF networks. Another important advantage is that, for a given fuzzy partition of the input space, the proposed method is able to determine the proper network structure, without using a trial and error procedure.

No abstract