The analysis and prediction of slope stability are very important because slope failure can lead to large disasters. This paper focused on a performance comparison of four supervised learning methods for slope stability prediction. Based on characteristics of slope instability and analysis of data availability, six typical slope parameters – the unit weight, cohesion, internal friction angle, slope inclination, slope height and pore water ratio – were chosen to establish the evaluation index system. The gravitational search algorithm (GSA), random forest (RF), support vector machine (SVM) and naive Bayesian (Bayes) were proposed to establish classifiers. A data set from more than 10 domestic and abroad slope projects was established to train and test the four classifiers, and then key parameters of the four models were optimized by using the method of 10-fold cross validation. The prediction performances of the four supervised learning methods were compared and analyzed. The results of accuracy, Kappa and receiver operating characteristic (ROC) curves reveal that both GSA and RF models can achieve satisfactory results, and the GSA model can obtain the best results when compared with the other three learning methods. Finally, 7 models with varying indicators are investigated to obtain the parameter sensitivity based on RF and GSA models.

In view of the effect of water on the physical and mechanical parameters of natural gypsum rock, in this study, gypsum rock in the goaf of the gypsum mine in the Jiangsu Province was selected as the research object, and gypsum rock samples were prepared with different immersion times. In addition, uniaxially tests were performed separately on the gypsum rock samples. Compression and scanning electron microscopy experiments were carried out to analyze the effects of immersion time on the uniaxial compressive strength, elastic modulus, Poisson’s ratio, and microstructure of gypsum. The results show that the uniaxial compressive strength and elastic modulus of gypsum are inversely proportional to the water content. However, the Poisson’s ratio is direly proportional to the water content, and the failure mode is destroyed by the brittle fracture of a single crack and is transformed into the shear ductile failure of the Y-shaped crack. Microscopically, with increasing immersion time, the bonds in the crystal of the microporous cracks and microcrack tips are weakened by hydrolysis, and the macroscopic structure is complicated by the internal structure of gypsum, and the end of the crack is expanded by the compressive action of water. Based on the damage mechanics, the evolution equation of gypsum soaking softening damage based on time factor was derived. The relationship between the brittleness coefficient and softening damage variable is revealed, providing a theoretical basis for the determination of the softening degree of gypsum in the goaf.

Rock burst is a common disaster in deep underground rock mass engineering excavation. In this paper, a cloud model is applied to classify and assess rock bursts. Some main factors that influence rock bursts include the uniaxial compressive strength (σc), the tensile strength (σt), the tangential stress (sθ), the rock brittleness coefficient (σc/σt), the stress coefficient (sθ/σc) and the elastic energy index (Wet), which are chosen to establish the evaluation index system. The weights of these indicators are obtained by the rough set method based on 246 sets of domestic and foreign rock burst samples. The 246 samples are classified by normalizing the data and establishing an RS-cloud model. The 10-fold cross validation (10-fold CV) was used to obtain higher generalization ability of models. The classification results of the RS-cloud model are compared with those of the Bayes, KNN and RF methods. The results show that the RS-cloud model exhibits higher values of accuracy, Kappa and three within-class classification metrics (recall, precision, and the F-measure) than the Bayes, KNN and RF methods. Hence, the RS-cloud model, which is characterized by high discriminatory ability and simplicity, is a reasonable and appropriate approach to rock burst classification and prediction. Finally, the sensitivity of six indexes was investigated to take scientific and reasonable measures to prevent or reduce the occurrence of rock bursts.

With the development of urban underground space in China, foundation pits have followed a trend toward larger area and greater depth. To ensure the safe excavation of foundation pits, it is necessary to monitor their deformation and the internal force of supporting structures and then use assessments of pit and building stability to guide construction. This paper introduces the design and development of a safety monitoring and warning system for deep foundation pits and adjacent buildings. The monitoring information management system uses Visual Studio 2013 as its development platform and adopts the SQL and C# programming languages to realize its powerful data management and visualization functions. By combined consideration of the monitoring information and the layout of the measuring points, monitoring data on the deformation of the deep foundation pit, ground subsidence and deformation of adjacent buildings are analyzed comprehensively. The results show that: first, the horizontal deformation of the side wall of the rock foundation pit achieves its peak value when 50% of the excavation has been completed and tends to change steadily after the completion of anchor construction. Second, the ground surface around the Fengjing area exhibits upward vertical deformation closest to the side wall of the foundation pit, and then downward deformation at greater distances from it. The ground surface relatively far from the side wall exhibits settlement. Third, the internal force of a plate-ribbed anchor retaining wall in the Fengjing area becomes greater at greater depths. Finally, the excavation of a rock foundation pit has relatively little influence on the inclination of adjacent buildings.