Search

Using WGAN and deep learning methods, a multiclass network intrusion detection model is proposed. The model uses the WGAN network to generate fake samples of rare attacks to achieve effective expansion of the original dataset and evaluates the samples through a two-classification method to ensure the effectiveness of the generated data. Through the CNN-LSTM network, the dimensionality reduction data is multiclassified and predicted. The network structure and parameters are effectively designed and trained to realize the identification and classification of network attacks. Experiments have proved that the model has improved the accuracy and recall index of network attack detection and classification compared with traditional methods. The proposed data generation method can improve the overall detection effect of the predictive model on rare attack types, and improve the accuracy rate and reduce errors reports.

High utility sequential patterns (HUSPs) are common patterns that can be discovered from the data collected in many domains (e.g. retail, bioinformatics, mobile commerce). To extract these patterns, high utility sequential pattern mining (HUSPM) has been proposed in recent decade. Although the HUSPM algorithms provide us a special perspective to analyze the knowledge behind the collected data, it also arises the risk of the privacy leakage and underlying security issues. This leads to the emergence of high utility sequential pattern hiding (HUSPH) whose purpose is to hide all HUSPs in the sequence database under a specified threshold. Around this topic, many algorithms were proposed. However, the existing algorithms are very time-consuming, which makes them unable to process the real massive data quickly. In this paper, we propose an efficient algorithm named FH-HUSP (fast algorithm for hiding high utility sequential patterns) for HUSPH. Substantial experimental results show that the proposed algorithm can hide all high utility sequential patterns quickly under the specific minimum utility with relatively small modifications.

With geometric fractal methods, the aftershock spatial-temporal fractal dimension of Wenchuan earthquake was calculated, and the variation rule of seismic b-value were compared with the results, which proved that the aftershock spatial-temporal fractal dimensions were shown with the overall decreasing trend in Wenchuan earthquake, which was the same to the variation rule of seismic b-value; strong aftershocks mainly occurred in the descending stage and the low valley area of the spatial-temporal fractal and b-value curve. This variation of spatial-temporal fractal was considered to represent the clustering variation of seismic spatial-temporal, reflecting the process of destruction and energy release with the tectonic stress action of earth crust rock mass.

As to substantial problems of seismic landslide disasters induced by Wenchuan earthquake, and based on the formation mechanism and movement features of Wenchuan seismic landslide, the seismic landslide is divided into two types of vibration-sliding and collapse-sliding. Of which, the vibration-sliding is also classified into the overall-sliding and fracture-sliding; the collapse-sliding is classified into the fall-sliding, dump-down-sliding and fall-ejection-sliding. Cases of various types of seismic landslide are cited in this thesis, where the formation mechanism and the developmental features are analyzed and summarized. It has certain significance for improving the understanding of seismic landslide and its induced geo-environmental hazards, and provides a scientific basis for the prevention of seismic landslide disasters.