Outlier Detection and Forecasting in the context of the Floodcontrol Project

From Master Projects
Jump to: navigation, search


has title::Outlier Detection and Forecasting in the context of the Floodcontrol Project
status: ongoing
Master: project within::Technical Artificial Intelligence
Student name: student name::Gabriel Mititelu
Dates
Start start date:=2011/02/14
End end date:=2011/08/15
Supervision
Supervisor: Zoltán Szlávik
Second reader: has second reader::Wojtek Kowalczyk
Company: has company::IBM Netherlands
Poster: has poster::Media:Media:Posternaam.pdf

Signature supervisor



..................................

Abstract

Current flood protection is primarily concerned with strong dikes. But the greatest gain lies in making the total system smarter: the dike, the decision-maker, and their environment. FloodControl 2015 integrates these three aspects in advanced forecasting- and decision-supporting systems. IBM, TNO and Deltares are among the initiators and participants of the project, together with ITC, Arcadis, Fugro HKV and Royal Haskoning.

The visual inspection of dikes, embankments, other sea defense structures and dunes is a crucial aspect in maintaining the structural integrity of these objects. Flood control starts with information. Sensor networks in dikes provide real-time information about behavior and the potential failure site. Satellites permanently monitor the flooding in large areas. Wide variety parameters are measured by sensors in and on levees. These include water tensions, temperatures, tilt, strain, self potential, acceleration, heat and hydraulic heads.

The project involves searching and detecting outliers (specific events that differ from normal behavior) in the levee data, and then based on the characteristics of the outliers building a forecasting model that will detect such events.

Validation of outliers can be done by making a selection of them, giving them to experts who would say if indeed something strange was identified or not.

The data involved will come from real dikes across The Netherlands. The research involves also analyzing the sensor behavior and how they are correlated.

The research will be carried within IBM in the context of the FloodControl project (http://www.floodcontrol2015.com).


1st KIM Abstract

Fighting water is one of the challenges of the current century. More than half the world’s population lives along coastlines, lakes and rivers. These deltas are densely populated and are becoming economically more valuable. The potential damage from flooding is therefore enormous. And the probability of it happening is continuing to increase. Ground is sinking due to water extraction and settlement. At the same time, water levels are rising. Climate change means an additional rise in sea levels and more extreme weather conditions. Flooding will occur more frequently and also be more extreme. Deltas are vulnerable, and this vulnerability is increasing.

With this threat on site, protection from water and levees especially will have a more important role. The visual inspection of dikes, embankments, other sea defense structures and dunes is a crucial aspect in maintaining the structural integrity of these objects. Flood control starts with information. Sensor networks in dikes provide real-time information about behavior and the potential failure site. Satellites permanently monitor the flooding in large areas. Wide variety parameters are measured by sensors in and on levees. These include water tensions, temperatures, tilt, strain, self-potential, acceleration, heat and hydraulic heads.

FloodControl 2015 integrates these three aspects in advanced forecasting- and decision-supporting systems. IBM, TNO and Deltares are among the initiators and participants of the project, together with ITC, Arcadis, Fugro HKV and Royal Haskoning. In the first KIM presentation the context of the problem will be provided along with real life examples. Also an overview of the experiments and tests done is provided.

Having the data from the experiments and also from real levees data at disposal we plan to answer to the following questions.

- What qualifies as an abnormal behavior? What are the sensor data values that characterize a macro-stability problem on a levee?

- What are the correlations between sensor reading?

- What is the best model to characterize a levee?

- Can we predict future behavior of dikes based on the real-time sensor data? Can we forecast the behavior of dikes under certain condition? (Step to be done depending on the answers to the first question)


The output of the project will be an outlier detection model and depending on that, a forecasting model.