Study On Acid Gaseous Emission And Its Artificial Neural Networks Predication In An MSW-Fired Fluidized Bed

Incineration technology was one of the main municipal solid waste (MSW) disposal measures introduced in the world, massive acid gaseous will pollute the environment seriously in the MSW-fired fluidized bed, its prediction and control was of great significance for decision-making. However, MSW has the characteristics of different density, multi-component, multi-particle size, multi-pollution source, high-moisture, multi-ignition point and multi-heat value. Gaseous pollutant emission is a multi-variable and non-couple complicated system, and it will sustain for a long time, so it was very difficult for us to construct a definite mathematic model. Artificial neural networks have good capture capability for non-linearity regularity, and it is very suitable for predicting gaseous pollutant emission.This dissertation was composed of some main parts as follow:This dissertation summarized international application and development status of MSW incineration technology and represented model, algorithm and latest progress report for artificial neural networks systematically.Formation mechanism of single and admixture component waste for acid gaseous emission characteristics varied with bed temperature and excess air, etc. were studied in the 150mm fluidized bed. This paper studied utilization efficiency of calcium-based desulfurizer and NOX mechanism influence for different calcium-based desulfurizer variety, size and Ca/S ratio.This paper introduced the process of constructing predication models for acid gaseous pollutants, including mainly the selection of model structure, algorithm, node excitation function, number of neural networks, learning precision, node number of hidden layers, characteristics function, initial values of weights and biases and learning rate.This paper studied input variables and output variables by canonical correlation analysis of emission influencing factors, and checked the results by test of significance. This paper studied the predication models by principal component analysis and simplified the neural networks by calculating variance percentage of input nodes. This paper studied the predication models by confidence interval analysis and checked degree of reliability of predicated results. This paper also compared the difference of multi-variant linear regression.Bad generality ability, cause and improvement means of local minimum for predication models has been studied. The scale of input patterns, organization principle and means, the selection, pretreatment and its reasonable standard of the training set and the test set.Based on a lot of patterns whose range was wide, the models were checked repeatedly, and estimated performance of main influencing factors for cid gaseous emission predication models of the MSW-fired BFB was analyzed carefully.Based on advanced in research and mechanism of PCDD/Fs and heavy metal, the difference of used experimental bench and actual incinerator, this paper analyzed forecasting feasibility for PCDD/Fs and heavy metal by adopting artificial neural networks technique.