【正文】 which may be roughly estimated by analysis but can only be determined accurately by testing runs. As for the control of trace pollutant emissions (in terms of ppm or ppd NOx, SO2, etc.) of a boiler, all of such uncertain factors as fuel positions, plexity of flow and temperature fields in the burning chamber, and the plicated mechanism of chemical reactions make model prediction highly unreliable. The inability of theoretical analysis makes empirical approaches necessary to explore the optimum operation conditions of a boiler. Nardin et al. [2] used fractional factorial design procedure and identified the important factors of NOx reduction in a fluidized bed bustor (FBC) with primary measures and selective noncatalyticreduction. Henryton et al. [3] performed an optimizingcontrol of NOx and SO2 emissions in the FBC process based on static models which were regressed from experimental data. George [4] suggested a boiler test system in order to monitor NOx emissions through optimizing various operation conditions in a prehensive and systematical manner, and claimed that achieving NOx emission reduction targets does not necessarily mean equipment retrofits or poor boiler performance. The capability of artificial neural networks (ANNS) as a universal modeling tool has been widely recognized in the last 20 years, and Cyberpunk [5] showed that ANNS could approximate any arbitrary nonlinear functions. ANNS offer an alternative approach to model process behavior, as they do not require a prior i knowledge of the process phenomena. They learn by extracting impede patterns from data that describe the relationship between the inputs and the outputs in any given process phenomenon. When appropriate inputs are applied to an ANN, the ANN acquires ‘knowledge’ from the environment in a process known as ‘learning’. As a result, the ANN assimilates information that can be recalled later. ANNS are capable of handling plex and nonlinear problems, processing information rapidly and reducing the engineering effort required in model development. ANNS have been successfully applied to a variety of problems such as process fault diagnosis, system identification, pattern recognition, process modeling and control, and statistical time series modeling. Karyological [6] gave a review on the application of ANNS in energy systems.Reif man et al. [7] developed an intelligent emissions controller for fuel gas reburning in coalfired power plant,and in their study, a feedforward neural network (FFN) was used to model the static nonlinear relationships between the distribution of injected natural gas into the upper region of the furnace of a coalfired boiler and the correspondingoxides of nitrogen emissions exiting the furnace. ZHOU et al. [8] used an ANN model and genetic algorithms to optimize low NOx pulverized coal bustion. Other instances of ANNS applied in solving bustion problems include: modeling the temporal evolution of a reduced bustionsystem [9], predicting coal ash fusion temperature [10], predicting coal/char bustion rate [11], conducting model predictive control in thermal power plant [12], etc. It is obvious that optimization of plex systems such as the bustion process of a boiler is a trialanderror process. In such an iterative process, experimental test is performed, and the test data are analyzed, and further test is suggested based on the analysis, and such iterations continue until satisfactory performance is achieved. In this study, a novel optimization procedure is proposed by extending the experimental design method developed by the authors [13] to constrained cases, and is used to search for the best operation conditions of a simulated coalfired bustion process. The proposed optimization procedure uses ANNs to model the relationship of the performance index with various operating variables, searches the built response surface under constraints to produce candidate points of next batch of test, determines the test points of next batch through information analysis. This procedure works iteratively and optimum conditions are expected after several batches of test. The major advantages of the proposed procedure are its abilities to deal with multivariable s, to precisely determine the number and location of future test experiments, to consider the noncatalytical constraints, and to locate multiple optima. The objective of this study is to demonstrate the effectiveness of the proposed optimization procedure in searching for the optimum operation conditions of a boiler. In the following context, a simplified model for the bustion process of a coalfired boiler is built, and theoptimization procedure is introduced and is applied to optimize the operation conditions of the simulated bustion process. It should be noted that the intent of this paper is to describe the proposed optimization procedure and to show its power in improving the overall performance of a boiler’s bustion process to achieve as high thermal efficiency and as low pollutant emissions as possible and not to provide an indepth analysis of a boiler’s bustion behavior. It should be noted that the above model for bustion is roughly estimated in the following aspects. (a) Plugflow is assumed through the bustion tunnel. (b) Homogeneous reaction is used in the bustion process and the contribution of unburnt char in the ash to NOx and COformation is not included. (c) Temperature profile used in this study is just an estimate because it is correlated to plex heat transfer, flow field and other factors. (d) Thermal efficiency defined in Eq. (4) is inplete, since heat loss caused by ash and water discharge, by inpletely burnt coal particles, and by surface convection with andradiation to the atmosphere, etc. is neglected. (e) The included chemical species and reactions are approximate to the true mechanism of chemical change in the bustion process. (f) Compositions of coal and air are simplified withou