National Technical University of Athens (NTUA)
Διερεύνηση της τεχνικογεωλογικής συμπεριφοράς των γεωλογικών σχηματισμών με τη χρήση πληροφοριακών συστημάτων
Abstract
dc:descriptionThe purpose of this thesis was to develop an integrated management framework of raw data that was designed to investigate the engineering geological behavior of the geological formations. Specifically, the investigation was to assess the landslide susceptibility, implementation of which was conducted in an area of Xanthi Prefecture, Greece. The development of landslide susceptibility zones was achieved through the application of spatial predictive models that made use of techniques and methods from the scientific branch of Machine Learning and Data Mining. For the purposes of the methodological approach presented in the thesis, eight (8) spatial variables were selected, which correspond to the primary geomorphological and geotechnical data. In particular, the created Layers of Information were related to: the altitude, the morphological gradient, the direction of morphological gradient, the engineering geological units, the distance from the geological boundaries, the distance from the tectonic features, the distance from the hydrographic network and the distance from the road network. The formation of the classes for each Layer of Information where achieved by combining the outcomes of the performed statistical analysis and the knowledge and experience of landslide experts. The indexing of past, archival landslide event material, the examination of the aerial photographs but mostly the field observation helped to create a relevant data set of 163 landslide events. The data set was "enriched" with data that correspond to areas that showed scant chance of failure, characterized as stable slopes. The estimation of the stable slopes was based on the metric distance Mahalanobis that was calculated between each grid cell of the study area and the "characteristic cell failure". The calculated distance was used by a classifier that created two (2) data sets with two (2) possible states that could characterize the grid cell, stable and an unstable. These two (2) data sets were combined to a single set that was used as the basic training data set for each of the prediction models developed in this thesis. Regarding the predictive models, four (4) predictions models were formed. Specifically: A prediction model, Similarity Index Model, which incorporates a modified machine learning algorithm, k-NN, nearest neighbor. A prediction model, Decision Tree Model, which incorporates a modified Decision Tree algorithm, with certainty factors for each produced rule, estimated by the statistical method, Certainty Factor. A prediction model based on the principles of Artificial Neural Networks, Artificial Neural Network Model, that incorporates a back propagation algorithm, and also a modified algorithm with the help of which, it was possible to estimate the optimal architecture of the Neural Network. A prediction model based on Bayesian Theory, Naive Bayes Model, which calculates the probability of failure in every grid cell of the research area. The results of the predictive models were compared with the results obtained by applying four (4) conventional predictive models that follow the qualitative and semi-quantitative techniques and qualitative assessment of landslide susceptibility (Analytical Hierarchy Process, Rock Engineering System, Statistical Index, Certainty Factor). The predictive models Artificial Neural Network Model and Naive Bayes Model were those that had the best performance, while the four (4) models that were formed following the principles of Machine Learning had better overall performance than those that followed the conventional forecasting models. In conclusion, the results of this study showed that the integration of Machine Learning techniques and methods in GIS, allowed the creation of reliable spatial prediction models in order to produce landslide susceptibility maps that provide basic knowledge in a preliminary study for the design on existing or future development.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2012
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Τσαγγαράτος, Παρασκευάς
- Tsangaratos, Paraskevas
Subjects
dc:subject × 22- Τεχνικογεωλογική διερεύνηση γεωλογικών σχηματισμών
- Κατολισθητική επιδεκτικότητα
- Μηχανική μάθηση
- Εξόρυξη δεδομένων
- Γεωγραφικά Συστήματα Πληροφοριών (ΓΣΠ)
- Τεχνητά νευρωνικά δίκτυα
- Δένδρα Απόφασης
- Αφελής ταξινομητής του Bayes
- Αλγόριθμος εγγύτερου γείτονα
- Data mining
- Engineering geological research
- Landslide susceptibility
- Machine learning
- Geographical Information Systems (GIS)
- Artificial neural networks
- Decision trees
- Naive Bayes classifier
- K-nearest neighbor algorithm
- Επιστήμες Μηχανικού και Τεχνολογία
- Περιβαλλοντική Μηχανική
- Engineering and Technology
- Environmental Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/27515