Matthias Sommer

An Organic Computing Approach to Resilient Traffic Management

Growing cities and the increasing number of vehicles per inhabitant lead to a higher volume of traffic in urban road networks. As space is limited and the extension of existing road infrastructure is expensive, the construction of new roads is not always an option. Therefore, it is necessary to optimise the urban road network to reduce the negative effects of traffic, for example, pollution emission and fuel consumption. Urban road networks are characterised by their large number of signalised intersections. Until now, the optimisation of these signalisations is mostly done manually through traffic engineers. As urban traffic demands tend to change constantly, it is almost impossible to foresee all runtime situations at design time. Hence, an approach is needed that is able to react adaptively at runtime to optimise signalisations of intersections according to the monitored situation. The resilient traffic management system offers a decentralised approach with communicating intersections, which are able to adapt their signalisation dynamically at runtime and establish progressive signal systems (PSS) to optimise traffic flows and the number of stops per vehicle.

Learning a Dynamic Re-combination Strategy of Forecast Techniques at Runtime

Traffic experts try to optimise the signalisation of traffic light controllers during design-time based on historic traffic flow data. Traffic exhibits dynamic behaviour. Due to changing traffic demands, new and flexible traffic management systems are needed that optimise themselves during runtime. Organic Traffic Control is such a decentralised, self-organising system that adapts the green times of traffic lights to the current traffic conditions. Forecasts of future traffic conditions may result in a faster adaptation, higher robustness and flexibility. The combination of several forecasting techniques leads to fewer forecast errors. This paper presents three novel combination strategies from the machine learning domain using an Artificial Neural Network, Historic Load Curves and an Extended Classifier System.

Predictive Load Balancing in Cloud Computing Environments Based on Ensemble Forecasting

Cloud Computing allows the on-demand usage of computing resources in a pay-as-you-use manner. One major problem for Cloud providers is the trade-off between the huge amount of energy consumption resulting from the non optimal utilisation of their servers, and meeting the service level agreements. Virtualisation allows for a better utilisation of existing servers while maintaining the required quality of service, increasing the return on investment. Consequently, dynamic algorithms are needed that determine an optimal plan for the live migration and allocation of Virtual Machines (VM) during run time. The contribution of this paper is as follows: First, we present our forecast module for time series to future utilisation of VMs. Second, we demonstrate how forecasts of CPU utilisation can be used beneficially in Cloud Computing environments. We propose a novel proactive VM migration policy utilising forecasts(PRUF) in Cloud data centres using an predictive overload detection. It uses short-term VM utilisation forecasts based on an ensemble forecasting approach to estimate which host will be overloaded when the next migration process is triggered. A study in the cloud computing simulation toolkit Cloud Sim shows that our predictive approach reduces the number of service level agreement violations and the performance degradation due to VM migrations compared to VM migration algorithms implemented in Cloud Sim.

Ensemble Time Series Forecasting with XCSF

Time series forecasting constitutes an important aspect of any technical system, since the underlying data generating processes vary over time. In order to take system designers out of the loop, efforts for designing self-adaptive, learning systems have extensively been made. By means of forecasting the succeeding system state, the system is enabled to anticipate how to reconfigure itself for satisfying the upcoming conditions. Ensembleforecasting is a specific means of combining the forecasts of multiple independent forecast methods. In this work, we draw a picture of how the eXtended Classifier System for Function approximation (XCSF) can be utilised as a novel self-adaptive weighting approach for ensemble forecasting of univariate time series. We elaborate on the fundamental ideas and evaluate our proposed technique on the basis of several time series.

Learning Classifier Systems for Road Traffic Congestion Detection.

The increase in mobility leads to a higher number of kilometres driven per vehicle and more delay due to congestion which poses a recent and future problem. Congestion generates growing environmental pollution and more car accidents. We apply machine learning concepts to the task of congestion detection in road traffic. We focus on the extended classifier system XCSR, an evolutionary rule-based on-line learning classifier system. Experiments with real-world detector data demonstrate high accuracy of XCSR for congestion detection on interstates.

Adapting Signal Timings to Automated Incident Alarms within a Self-organised Traffic Control System.

Intersection management, routing, and congestion avoidance are key factors for improved mobility and better road network utilisation. Organic Traffic Control (OTC) is a self-organising traffic management system for urban road networks. Its main features are the self-adaptive traffic-responsive signalisation of intersections, the coordination of traffic light controllers, and dynamic route guidance of traffic streams. This paper aims at presenting how the automatic and fully distributed incident detection within OTC works and how OTC makes use of these incident alarms for the automated adaptation of signalisation.

Local ensemble weighting in the context of time series forecasting using XCSF

Time series forecasting constitutes an important aspect of any kind of technical system, since the underlying stochastic processes vary over time. Extensive efforts for designing self-adaptive learning systems have been made, to take system designers out of the loop. One goal of such systems is to transfer design-time decisions, e.g. parametrisation, to the run-time. By means of forecasting the succeeding system state, the system itself is enabled to anticipate, how to reconfigure to handle upcoming conditions. Ensemble forecasting is a specific means of combining and weighting the forecasts of multiple independent forecast methods. This concept has proven successful in various domains today. In this work, we present our self-adaptive forecast module for ensemble forecasting of univariate time series and draw a picture of how the eXtended Classifier System for Function approximation (XCSF) can be utilised as a novel weighting approach in this context. We elaborate on the fundamental ideas and evaluate our proposed technique on the basis of several time series with different characteristics.

Forecast-augmented Route Guidance in Urban Traffic Networks based on Infrastructure Observations

Increasing mobility and raising traffic demands lead to serious congestion problems. Intelligent traffic management systems try to alleviate this problem with optimised signalisation of traffic lights and dynamic route guidance (DRG). One solution for the former aspect is Organic Traffic Control (OTC), offering a self-organised, decentralised traffic control system. Based on OTC, this paper presents two proactive routing protocols, resembling techniques known from the Internet domain, applied to the traffic routing problem: Distance Vector Routing and Link State Routing. These protocols were adapted to utilise forecasts of traffic flows to offer anticipatory and time-dependant DRG for road users. The efficiency of these protocols is demonstrated with simulations of two Manhattan-type road networks under disturbed and undisturbed conditions. The results indicate their benefit in terms of lower travel times and emissions, even under low compliance rates.