Portland cement clinker strength prediction method based on DNN neural network

Abstract

The invention discloses a Portland cement clinker strength prediction method based on a DNN neural network. The method comprises the following steps: S1, acquiring factors influencing cement clinker strength; S2, neural network model selection: selecting a DNN neural network model; S3, carrying out data dimensionalization processing; and S4, constructing a neural network model which comprises an input layer, a group of hidden layers and an output layer, wherein the input layer is fully connected with the hidden layers, neurons of each layer are only fully connected with neurons of adjacent layers, no connection exists between the neurons in the same layer, no feedback connection exists between the neurons of each layer, the data subjected to dimensionalization processing serves as an input variable, and the intensity value serves as output. The Portland cement clinker strength is effectively predicted in advance, so that an effective support basis is provided for the subsequent improvement of the cement admixture doping amount, the production cost is saved, and the enterprise profit is improved.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a method for predicting the 28-day strength of Portland cement clinker based on a DNN neural network. Background technique [0002] The prediction of the 28-day strength of Portland cement clinker is a complex problem with multivariable, nonlinear, large time-delay and no fixed model. Various minerals in clinker undergo a series of complex chemical reactions in the process of cement hydration and hardening, so the influence of various minerals and their relative content in clinker on the strength of clinker is very complicated, and it is impossible to use a simple functional relationship To represent. The 28-day strength of the clinker in the enterprise needs to wait until the 28-day manual test to obtain it. The cycle is long and the lag time is long, so it is impossible to pre-adjust the amount of mixed materials. Therefore, it is necessary to find the main factors af...

AI Technical Summary

Problems solved by technology

[0003] Traditional regression prediction algorithms such as linear regression, least squares and other algorithms are not accurate enough to predict this complex scene, and it is easy to simplify the highly nonlinear relationship to a linear relationship, which affects the prediction accuracy. The DNN neural network is a neural network based on error backpropagation. Network algorithm is also one of the most widely used neural network algorithms

Images