Radial basic function networks for atmospheric Cycloid combustion process modelling

Combustion of coal to generate heat for the generation of steam that drives a turbine to produce electricity is widely used. Considerable advances have been and continue to be made to improve the performance of the combustion process itself, either by staging the process or by better control. These have resulted in improved fuelefficiency and a reduction in emissions exiting the boiler. During the past years, the emission taxation procedures have made the minimization of harmful flue-gas emissions of power plants a profitable task. In addition to the developments in the plant construction and flue gas cleaners, the control of the process operating conditions is an important and cost-effective way to affect these emissions. In order to be able to optimise the plant operation, models for the variables in the overall cost function are required. Plants vary in design, and representative models based on first principles are difficult to build and maintain. When trying to model the non-linear, static/or dynamic behaviour of a given physical system, several different approaches can be chosen, physical system modelling (analytical modelling) is based on the first principles governing the behaviour of the system under investigation and requires the knowledge of parameters of the constitutive relations and phenomenological laws involved. Since quite often these system parameters are not known, the application of this approach is limited. Nonlinear system modelling based on measured inputand output data, i.e. non-linear system parameter estimation is a well-established discipline. Here, the system structure has to be assumed in advance, and the parameters are estimated using the least squares algorithm or other appropriate methods. Since the parameter estimation is usually performed in discrete time, the thus obtained parameter has no physical meaning. Artificial neural networks are widely accepted as a technology offering an alternative way to tackle complex and ill-defined problems. They can learn from examples, are fault tolerant in the sense that they are able to handle noisy and incomplete data, are able to deal with nonlinear problems and, once trained, can perform prediction and generalization at high speed. They have been used in diverse applications in control, medicine, power systems manufacturing and optimisations. They are particularly useful in system modelling, such as in implementing complex mappings and system identification. A neural network is a computing technique that simulates, at a primary level, the behaviour of the human brain for the treatment of information. A neural network consists of two parts: • Neurons, individual information processing centres. • Weights or synaptic connections between neurons. The most important features of neural networks, which are also their biggest advantage over traditional techniques, are as follows: • Adaptive learning: the ability to learn by example and adapt to new conditions using a learning algorithm which allows the maximum possible generality and precision, aims which are usually opposed in modelling processes. • Real time operation.