Partial Least-squares Regression Theory And Its Application In Dirt Prediction

To seek for the uniform standards to judge characteristics of coal boiler slagging, this thesis chooses the six parameters including ash softening temperature, alkaline acid ratio, silicon ratio, silicon aluminum ratio, average temperature of non-dimensional hearth furnace and non-dimensional cut circle diameter as inputs. Boiler actual slagging degree is chosen as output, and establishes nonlinear iterative partial least squares regression (PLS) digital pattern recognition model discriminating the coal boiler slagging tendency. The degree of hearth slagging can be divided into three kinds, namely mild, moderate, and severe. These three kinds can be written as 1st class (note for 1), notes 2nd class (notes for 2) and the 3rd class (notes for 3), respectively; and the predictive value which is less than 1.5 is divided into the first category which, is equal to or greater than 1.5 less than 2.5 and is divided into the second category, which is equal to or greater than 2.5. It is gathered for the third kinds. Results show that the accuracy of forecasting and the accuracy of the back sentence to all is 100%.In order to accurately predict the fouling of heat exchange equipment, the time and 10 main water quality parameters have effect on the scaling of Songhuajiang river are iron, chloride ion, bacteria, pH value, dissolved oxygen, turbidity, conductivity, bacteria, alkalinity, hardness, etc. They were used as input variables. This thesis chooses 20 groups of data at different times, with the former 15 groups as a model of training and the latter 5 groups as test sample, and establishes a prediction model of plate heat exchanger of fouling resistances. This model has achieved good prediction results. Then the paper establishes anti-removed part prediction model of water quality parameters, and analyses the influence of water quality parameters dirt according to its prediction results. Adopting ten components of coal ash as input variables, the model for ash fusion which was trained by 60 samples was set up by use of partial least-square regression method. Then the PLS regression model is used to forecast ash deformation temperature (DT) of the coal with orderly decreased input variables. The predicted results with different numbers of ash component as the input variables were compared. Results show that the predicting results using ten components of coal ash as input variables are most accurate.