Prediction Model Of Cement Strength Based On Non-linear Artificial Network

With the increasing complexity of industrial processes, the model of traditional analysis is not far from the demands. A lot of multi-variable nonlinear redundancy data is boring the traditional analysis, but these high-dimensional data contains a lot of useful information .Therefore, when we analyses the high-dimensional data ,the raw data must be handled with reduction processing .The traditional methods of data processing had no effective means to extract useful information, resulting a tremendous waste of resources.28-day compressive strength of concrete value is a measure of the quality of cement. It is also an important indicator of a major label to determine the basis of cement, it plays an important role in the actual production process, which is a typical non-linear multi-variable system .It is very difficult to be used widely in practice with the traditional methods, because its accuracy is lower .In recent years, the neural network intelligence algorithms are applied to predict it and achieved the good results. Currently there are two kinds of neural network models widely to predict its strength One kind of is using the neural network directly, and the other is using statistical combined with neural network. The neural network based on statistical model has two shortages. First, when the sample correlation between the raw data is not too big, the principal component analysis can not effectively compress the data dimension Second the main component analysis method is based on the method of second-order statistical features, which can only extract the second order relationship between the random variables, while the higher order random variable relationship between the variable is not extracted by using of the principal component analysis .Some information of the raw data will be lost and not be used efficiently and it can influence its perdition accuracy.In this paper, we use the covariance function nonlinear principal component to deal with the raw data. Using the extracted components to be the new input variable of the neural network, and establish a model of 28-day compressive cement strength. This method is based on higher-order statistical relationship between the each variable, which can not only provide second-order relationship between the variables, but also extract hidden in the high-level relations between variables. In other words it can extract linear and nonlinear characteristics of the variable.