Application of BP neural network in predicting the cement materials performance

Deng Xue-jie, Kang Tao and Wan

Abstract

The 4-10-5 prediction network model was established based on the improved BP neural network, considering that the performance of cement materialsare impactedby many factors and the multivariate cross of those factors are difficult and strenuous to study in laboratory. Furthermore, the network model was trained and tested with the data collected from the laboratory test results. The results showed thatthe prediction precision of the well trained network model is high and reliable, whose fitting correlation coefficient and average prediction error are 0.969 and 6.72% respectively. Finally, the well trained network model was applied to predict the optimal material ratio, which can meet the requirement. Specifically, the optimal material ratio of fly ash, quicklime, cement, and gangue in cement materials are 35%, 10%, 2%, and 53% respectively, and the concentration of slurry is 77% by the above proportion.The cemented backfilling materials under this ratio has good effect in field application which prove the reliable of this prediction model.