Prediction of Compressive Strength of Solidified Dredged Kaolin Based on a BP Neural Network

The kaolin obtained by dredging exhibits high water content and compressibility， making it difficult to be used directly as a filling material. To address this problem， calcareous sand， cement， and lime were used to stabilize the kaolin. Based on the results of 83 groups of unconfined compressive strength tests， a backpropagation （BP） neural network model was developed to predict the 28-day unconfined compressive strength of the solidified kaolin， and the optimal mix proportion was selected according to the predicted results. The BP neural network model was also compared with a traditional linear regression model. The test results showed that the combined use of the three materials significantly improved the compressive strength of the kaolin. Scanning electron microscope （SEM） observations revealed that an appropriate amount of lime （≤2%） promoted the densification of soil particles， calcareous sand filled the pores and precipitated calcium carbonate， and the hydration products generated by cement formed a dense honeycomb or network structure， collectively contributing to the improvement of the soil's mechanical properties. In terms of model prediction， the BP neural network model yielded a root mean square error of 142.515 and achieved a coefficient of determination of 0.985， whereas the traditional linear regression model showed a root mean square error of 154.203 and a coefficient of determination of 0.979. This indicates that the BP neural network model provides higher prediction accuracy.