A Cement Free Calcium Soft-Sensing Method Based on Unsupervised and Supervised Learning

Abstract

The invention discloses a cement free calcium soft measurement method based on unsupervised and supervised learning. By analyzing the cement process, variables are selected as input variables of clinker fCaO soft measurement, and the time series of each variable is used as model input; Input variables, construct a prediction model combining unsupervised and supervised learning; remove the sparse self-encoded decoding layer, stack the encoding layer to form a deep network structure, use the determined initial parameters to initialize the deep network parameters, and use BP reverse error correction Algorithms for supervised learning; real-time prediction of cement clinker fCaO using a trained prediction model that combines unsupervised and supervised learning. The model of the present invention adopts a layer-by-layer greedy unsupervised learning method to extract high-level features of the data; combined with supervised reverse fine-tuning to further optimize parameters, the trained deep network is used to realize real-time prediction of clinker fCaO.

Description

technical field [0001] The invention relates to the field of industrial cement quality soft measurement monitoring, in particular to a cement free calcium soft measurement method based on unsupervised and supervised learning. Background technique [0002] The content of free calcium oxide (fCaO) in cement clinker is an important index to measure the quality of clinker in the production of new dry process cement. The clinker fCaO content not only affects the stability and clinker strength of cement, but also directly relates to the energy consumption of cement firing. At present, cement clinker fCaO is mainly measured by manual one-hour sampling and laboratory testing, but the off-line measurement results have obvious hysteresis during the cement firing process. optimization. The cement clinker firing process has the characteristics of large inertia, large time delay, multi-coupling, and few label samples, which makes it difficult to establish an accurate cement clinker fCa...

AI Technical Summary

Problems solved by technology

Considering that there are few labeled samples of clinker fCaO, the training of deep neural network with a small amount of sample data is prone to overfitting.

Images