|
Incineration technology is commonly used in the international waste disposal technology, waste incineration process acid gas pollution problem has aroused extensive concern. Municipal waste has a different proportion of multi-component, multi-particle scale, multi-sources of pollution, high moisture, ignition point, multi-calorific value characteristics, pollutant emissions fluidized bed incineration is a multi-variable uncoupled complex system, and reaction time long, large inertia, it is difficult to put forward a deterministic mathematical model to describe and control this process, artificial neural network technology with a strong ability to capture nonlinear variation, very suitable for refuse fluidized bed incineration pollutant emissions forecast . This study include the following aspects: the application and development of domestic and foreign garbage fluidized bed incineration technology, systems elaborate artificial neural network model and algorithm, the latest research progress at home and abroad artificial neural network technology; φ150mm bubbling stream fluidized bed incineration of a typical single-component and mixed component waste acid gas emissions characteristic test, the study bed temperature, combustion atmosphere acidic gases and specific mechanism. In φ150mm bubbling fluidized bed incineration test of typical components of garbage, the study of the efficiency in the use of calcium-based sorbent, and calcium-based sorbent species, particle size and the Ca / S ratio of NO transition rate of calcium-based sorbent NO increase in the conversion rate mechanism; establish the the acidic gases BP neural network prediction model, model structures, algorithms, node activation function, network layers, learning precision, hidden layer nodes, the objective function error, initial weights value threshold, the selection and setting of the learning rate; through the typical analysis and forecasting model to study the relationship between the input variables and pollutants emissions output variables parameters and analysis of the results of the test of significance; principal component analysis and forecasting model come to the variance contribution of model input node, and in order to simplify the model structure; draw the model to predict the results of analysis and forecasting model by the confidence interval confidence level; analysis and comparison of the predicted effects of the linear regression and BP neural network model; analysis discusses the network generalization ability is weak, local minima causes and improvement measures, the size of the network model input sample set, organizational principles and methods, training sample set and the testing sample set selection, pre-processing and select reasonable criteria; using repeated testing of a wide range of representative samples of the training model and detailed examination of the model prediction performance of fluidized bed incineration of garbage a major impact parameter of the acid gas emissions; mechanism, the paper test through the analysis of dioxins and heavy metals Research the actual garbage incinerator differences, to explore the feasibility of using neural networks to predict dioxin and heavy metals.
|